The benchmarks page tries to answer that. It takes the full coverage universe of quarter-level blended dilution data and groups it by company scale, IPO maturity, and revenue growth. The result is a set of reference curves that give you an empirical baseline for what dilution looks like across different company profiles.

I’ll walk through the three main charts using the rideshare and delivery cohort (Uber, Lyft, DoorDash, and Instacart) as a worked example on each one. These four companies share a sector but sit at very different points on the maturity spectrum, which makes them a useful illustration of how the benchmarks work.

Subscribe now

Dilution By Market Cap

The first chart groups dilution by market cap at the time of each quarterly observation. Smaller companies dilute more. A lot more. Companies under $5B are running blended dilution in the range of 3.5–5%. By the $25–50B range, that’s dropped to roughly 2.5%. At $100B+ it flattens out around 1–1.5%, and the trillion-dollar cohort sits near 0.8%.

Overlaying the rideshare cohort tells the story clearly. Lyft, sitting in the $5–10B market cap range, is running dilution at 4.1%, right at the top of the confidence band for its size. Instacart, in the $10–25B range, is at 3.8%, also near the upper edge. Both are diluting at a rate that is high even by the standards of companies their size.

DoorDash, at $50–100B, is running at 2.1%, which is slightly above the benchmark line but still within the confidence band. And Uber, now in the $100–200B range, is at 1.3%, right in line with what you’d expect for a company of that scale.

The shape of this curve matters. The steepest drop happens between <$5B and $25B. That’s the zone where companies are scaling past the initial post-IPO grant overhang, buyback programs are starting to kick in, and comp committees are getting more serious about managing dilution. Once a company crosses into mega-cap territory, dilution converges toward a narrow band. The market basically forces convergence.

Dilution By Age Since IPO

This chart groups dilution by how many years a company has been public. The pattern here is different from market cap, and more interesting in some ways.

Dilution increases slightly in the first year post-IPO, peaking around 3.5% at roughly the one-year mark as the initial wave of IPO-related grants vest. After that peak it enters a long, gradual decline. This decline is a lot shallower than I would’ve expected but this is likely reinforced by poor ZIRP behavior as companies could behave almost like children for longer.

The rideshare overlay makes the divergence in this cohort impossible to miss. Instacart, about 2 years post-IPO, is running above the benchmark line but at least partially explainable by its stage. DoorDash, roughly 5 years out, is at 2.1%, which is slightly below the benchmark for that age. Uber, at about 6 years post-IPO, is at 1.3% meaningfully below the line, which at that age sits closer to 2.5%. Uber has matured its dilution profile faster than the typical public company trajectory would predict.

And then there’s Lyft. Nearly 7 years post-IPO and running dilution at 4.1%, well above the benchmark line and outside the upper confidence band. For a company that has been public that long, this level of dilution is an outlier. The benchmark for a 7-year public company is somewhere around 2.3%. Lyft is nearly double that. This is a company that has failed to bring dilution down on the timeline you’d expect.

The confidence band is widest around years 8–11, which suggests that’s the period where company outcomes diverge the most. Some companies are actively managing dilution down by that point. Others are still granting at elevated rates, often because they haven’t achieved the stock price appreciation that would let them reduce grant volumes without cutting comp.

Subscribe now

Dilution By Revenue Growth

This is the chart that surprises people the most, and it’s the one I think is most underappreciated.

Dilution increases with revenue growth. Companies with declining or low-single-digit revenue growth cluster around 0.5–1% dilution. By 20–25% revenue growth, dilution is around 2%. And the fastest growers, 40%+, are running at 2.5% or higher. This makes sense as faster-growing companies are hiring aggressively, competing for expensive engineering talent, and using equity as a primary compensation tool.

But the revenue growth dimension adds something the other charts don’t: it lets you evaluate whether the growth justifies the dilution. This is where the rideshare cohort gets really interesting.

DoorDash is growing revenue around 25–30% and diluting at roughly 2.1%. That’s right on the benchmark line. Shareholders are getting rapid growth in exchange for the dilution. The trade is roughly fair by historical standards.

Uber, growing at roughly 15–20%, is diluting at 1.3%, below the benchmark line for that growth rate. This is a company whose dilution is lower than what its growth rate would typically justify. That’s a sign of SBC discipline.

Now compare those to Lyft and Instacart. Both are growing revenue in the 5–15% range, but diluting at 4%+. The benchmark for that level of growth is somewhere around 1–1.5%. They’re diluting at 3–4x the rate you’d expect for their growth profile. This is the chart that makes the sharpest case: you’re getting modest growth and elevated dilution, which is the worst combination for shareholders.

How To Use This

The benchmarks page on the site lets you overlay individual company positions across all three charts, filterable by sector and tag. The rideshare cohort is just one example, you can do the same exercise with software, big tech, or any sector in the coverage universe.

The framework is straightforward. Take any company and plot it against the relevant curve for its market cap, IPO age, and growth rate. If a company is above the line on all three dimensions, that’s a red flag. If it’s below on all three, that’s a company with stronger SBC discipline.

The full benchmarks page is live at dilution.platformaeronaut.com/benchmarks.

Performance & Valuation Snapshot

Note: Email renders these as images, click through for interactive filters or view on Platform Aeronaut.

What I Read This Week

• Tripadvisor showed an end-to-end AI planning prototype at Nvidia GTC: Tripadvisor partnered with Nvidia, Nebius, and Nexla to demo a more personalized AI planning flow built around travel data, video, and data integration

• The next travel SEO fight is AI visibility: PhocusWire reports brands are shifting from classic SEO toward GEO and AEO as AI answers become a discovery layer; a companion piece says nearly half of marketers report lower search traffic because of AI-generated answers.

• Grab made a $600 million move beyond Southeast Asia: Grab agreed to buy Foodpanda Taiwan, calling it its first expansion outside the region and part of a broader strategy built around AI, newer services, and selective overseas deals.

• Europe’s first commercial robotaxi push is forming in Zagreb: Uber, Pony.ai, and Verne said testing is already underway in Croatia; Uber will distribute rides, Verne will run the fleet, and Pony.ai says it expects more than 3,000 robotaxis across 20+ cities this year after posting its first quarterly profit.

• AI has crossed into the mainstream for U.S. travelers: Phocuswright says 56% of active U.S. travelers used AI for planning, booking, or in-destination help in the past 12 months, up from 43% in late 2025 and 33% in early 2025.

• Wing is bringing drone delivery to the Bay Area: Alphabet’s Wing plans Bay Area home delivery in coming months as it scales small-item and meal delivery; Reuters says the company has completed more than 750,000 deliveries so far.

• Zoox is widening its U.S. footprint: Amazon’s robotaxi unit is expanding in San Francisco and Las Vegas and will start public-road testing in Austin and Miami before broader rollouts later this year.

Transcript Highlights

Adobe (ADBE) Q1 2026 Earnings Call

• Generative credit consumption grew over 45% quarter-over-quarter, with Firefly usage skewing to higher-value modalities such as video (over 8x year-over-year growth) and audio (2x year-over-year growth).

• “In Q1, we delivered breakthrough innovations enabling GenStudio-created assets to flow directly into activation workflows across the Adobe stack and a broad ecosystem of advertising platforms, including Amazon Ads, Google, LinkedIn, and Meta.”

• Credits act like tokens across creative apps and that more users are generating higher-resolution and higher-modality content as AI becomes embedded in workflows. Cited that broader adoption, Firefly as a destination, and purchases of credit packs by creative professionals are driving the consumption increase.

HelloFresh (HLFFF) Q4 2025 Earnings Call

• Tenured meal kit customers (over 50 lifetime orders) now represent 51% of meal kit orders, up from 13% in 2022 and up from 42% at end-2024 in Fabien’s comments, indicating higher revenue quality.

• Marketing spend declined by over €200M in 2025, with U.S. meal kit marketing spend down more than 40% in USD as the company prioritized marketing ROI over volume.

• Stock-based compensation expenses were reduced by over 30% year-on-year as part of equity compensation restructuring.

Frontier (ULCC) 2026 JP Morgan Industrials Conference

• The company has four stated focus areas: rightsizing the fleet, reducing unit costs, building loyalty/repeat customers, and improving customer service metrics such as on-time performance and completion factor.

• Bookings have remained resilient in the immediate period after the Iran-related oil-price spike, with the company describing no meaningful short-term booking disruption.

• "You will have the front two rows with first-class seats by the end of this year."

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

Platform Aeronaut Dilution

Today I’m launching Platform Aeronaut Dilution, a live, queryable platform for public-company SBC and dilution analytics. It covers 80+ companies across Big Tech, Internet, Software, and Non-Tech, updated quarterly as earnings roll in. Think of it as the dilution analysis I’ve been publishing, except interactive, permanent, and filterable.

Below I’ll walk through what the site does, why the methodology is built the way it is, and a few of the current data points that I find most interesting.

Subscribe now

Why Dilution Deserves Its Own Platform

The one thing I keep coming back to is that dilution is the most persistently underweighted factor in how investors evaluate technology companies. It’s not that people don’t know about SBC (it comes up in every earnings call and every analyst model). The issue is that it gets waved away. “Non-cash.” “It’s already in the diluted share count.” “Everybody does it.”

None of those are actually good answers. When a company runs at 4% dilution per year, the market cap of that business needs to compound 22% over 5 years just to keep the stock price flat. That math is real regardless of how the expense is classified on the income statement.

What I kept finding as I dug deeper into the data was that it’s genuinely hard to do this analysis across companies at scale. Every management team has their preferred framing. Some use forward dilution, some trailing, some SBC-as-a-percent-of-revenue. The numbers are scattered across 10-Ks, proxy statements, and earnings supplements. There’s no single source of truth.

The platform is my attempt to build that source of truth: a consistent, cross-company lens on dilution that doesn’t follow each management team’s preferred presentation.

What’s On the Site

The homepage gives you a quick snapshot of the current state of the market: a run-rate dilution aggregate across covered public companies, SBC as a percent of FCF, and the leaderboard movers for the current quarter. The dilution decelerators and accelerators widget is one of my favorite features: it’s the fastest way to see who’s improving and who’s getting worse quickly.

From the current data: YELP is the top dilution decelerator meaning their forward dilution today is better than their blended dilution implying future deceleration to their dilution rate. On the dilution accelerators side, Wayfair increased 2.2 points and continues to be one of the more egregious diluters in the coverage universe. Lyft, Snap, and Atlassian are also on the wrong end of the movers list. None of those should surprise anyone who’s read the prior posts.

Dilution Leaderboard

The Leaderboard page is where the full ranking lives. Right now the top of the tech leaderboard looks like this:

Netflix and Nvidia are tied at the top with 0.3% run-rate dilution. BKNG comes in at #3 with 0.5% (consistent with how I’ve always characterized them as the gold standard for SBC discipline in the internet cohort.) Apple, Shopify, and Spotify round out the top six. The leaderboard is filterable by cohort (Mega Cap, Large Cap, Mid Cap, Small Cap), sector, and quarter going back to 2018.

That last piece is important, the historical archive lets you go back and see how a company’s rank has changed over time. Palantir, for example, dropped 21 spots from last quarter to this one. Coinbase fell 16. Robinhood, on the other hand, has climbed to 9th (a remarkable improvement from the 10%+ dilution death spiral they were running in 2022).

Subscribe now

Individual Company Pages

The Companies page lets you drill into any individual ticker. Each company page includes the core dilution metrics across the most recent quarters, cohort and sector comparisons, and the underlying methodology inputs. The goal is to make it auditable: you can see not just what the number is, but roughly how it was derived.

Custom Dilution Charts

The Charts page has the ability for users to customize a dilution chart and compare companies any way that they wish. If you’re logged in you can save, edit, and share publicly if you choose. I’ll continue to refine this and add additional chart types and analysis in the future.

Dilution by Sector

Finally, the Sectors page breaks the universe into Big Tech (10 companies), Internet (25 companies), Software (31 companies), and Non-Tech (14 companies). The non-tech cohort matters because it’s the benchmark. Best-in-class non-tech dilution runs at 10-30 basis points per year. at 0.1%. at 0.1%. That’s the bar that tech companies should be aspiring to as they mature.

Subscribe now

One Interesting Call-Out on Tesla

One thing worth calling out that’s impacting some of the sector and cohort level metrics is Tesla’s huge pay package to Elon. It’s single handedly boosted the mega cap and big tech metrics up based on it pushing dilution to almost 8%

A Note on Methodology

I’ve gotten pushback over the years on methodology (sometimes from management teams, sometimes from readers). The most common objection is about which share count to use: basic shares, diluted shares, or fully diluted. It’s a legitimate question and I’ve addressed it in prior posts. The honest answer is that getting to true FDSO for 100+ companies in a time-efficient way isn’t really possible, so there are tradeoffs.

The platform’s methodology uses a blended approach that averages forward and trailing dilution, with SBC as a % of FCF as a secondary metric. The forward dilution methodology (net RSU and option grants divided by shares outstanding) is what most management teams reference when they talk about “run-rate dilution.” It’s forward-looking and captures decisions being made today about grants, cancellations, and RIFs. The trailing methodology is what actually happened over the last twelve months. Averaging the two cuts through both the noise of timing and the optimism sometimes baked into forward framing.

The cohort and sector relativization is equally important. Comparing Confluent to Apple on raw dilution doesn’t tell you much because they’re at different stages of maturity, different capital intensity profiles, different employee cost structures. The leaderboard ranks companies against their peer set so the context is relevant.

I’ve published a full methodology page that covers the core metrics, the ranking approach, and the important limitations, including the fact that this is a directional analytical tool, not investment advice.

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

The more important question now is: who is best positioned to turn AV supply into a scaled rideshare network? That is a different question entirely. And when you frame the market that way, the setup today looks something like this:

• Waymo is the clear leader in actual commercial robotaxi deployment.

• Tesla is still the most hyped player, but operationally and regulatorily still has a lot to prove.

• Lyft has made sure it is not completely out of the game, but it still looks stuck in no man’s land.

• Uber meanwhile is increasingly putting together the broadest AV aggregation layer in the market, and by the end of 2027 I think there is a very real chance that is what matters most.

Subscribe now

To level set: this is no longer just about the best AV stack

Waymo is the only player that has really crossed from demo to scaled commercial reality.

The company said in February it was already doing more than 400,000 paid rides per week across its operating markets, and it raised another $1.6 billion while laying groundwork for expansion into more cities. Its new Arizona manufacturing facility with Magna is designed to produce “tens of thousands” of autonomous vehicles per year at full capacity, with new Zeekr RT and a multi-year Hyundai partnership Ioniq 5 vehicles joining the Jaguar I-PACE platform.

That is the most real robotaxi business in the U.S. by a mile. But the leap from “best AV operator today” to “winner of AV rideshare economics” is not automatic.

Because scaled rideshare is not just a software problem. It is a supply problem, a dispatch problem, a maintenance problem, a financing problem, and maybe most importantly a utilization problem. That is where Uber’s setup starts to look much more interesting than the market gives it credit for.

Uber is not trying to win autonomy. It is trying to win the network.

Uber’s strategy now looks pretty clear: let others build the autonomous brain, while Uber becomes the default marketplace, demand layer, and utilization optimizer. That may end up being the smarter economic position.

Nuro / Lucid: hard unit supply

This is Uber’s clearest and most tangible AV supply commitment: 20,000 or more Lucid Gravity vehicles equipped with the Nuro Driver, launching first in a major U.S. city in 2026 and ultimately intended for dozens of markets globally. The vehicles will be owned and operated by Uber or third-party fleet partners and offered exclusively through Uber. That makes this one of the biggest hard-number AV supply announcements in the market, and the cleanest proof that Uber is moving beyond pilots and toward real fleet formation.

Wayve / Nissan: city breadth, not fleet size

The Wayve relationship matters, but for a different reason than Nuro / Lucid. What Uber, Wayve, and Nissan announced is a late-2026 Tokyo pilot, framed as part of a planned global rollout across more than 10 cities, including London. That gives Uber geographic breadth, international optionality, and another autonomy stack integrated into its marketplace. But unlike Nuro / Lucid, there is still no hard vehicle count attached. So this is best framed as network breadth and expansion potential rather than concrete disclosed supply.

Nvidia: ecosystem architecture and very large long-term ambition

The NVIDIA partnership is broader and more strategic than the others. NVIDIA said Uber will begin scaling its global autonomous fleet starting in 2027, targeting 100,000 vehicles over time, supported by a common L4-ready architecture and a growing ecosystem that already includes names like Nuro, Wayve, May Mobility, and Avride. This is less a near-term fleet launch than a statement that Uber wants to become the operating layer for a very large multi-partner AV ecosystem. It is long-dated, but it reinforces the same core point: Uber is building a portfolio, not betting on a single stack or OEM.

Volkswagen / MOIA: another large U.S. fleet path

The Volkswagen / MOIA deal is easy to overlook, but it adds another meaningful U.S. lane. Uber and Volkswagen said they plan to deploy a fleet of thousands of autonomous ID. Buzz vehicles across multiple U.S. markets over the next decade, starting in Los Angeles, with testing beginning in late 2025 and commercial launch expected in 2026. This is not as near-term concrete as Nuro / Lucid, but it is another proof point that major OEM-backed AV supply is choosing to plug into Uber rather than build a standalone demand network from scratch.

Zoox: premium U.S. partner, but no disclosed unit count

The Zoox partnership adds another important U.S. AV partner to Uber’s roster. Zoox robotaxis will launch on Uber in Las Vegas this summer and in Los Angeles by mid-2027, while Zoox continues to run its own app in parallel. Strategically, the most interesting part is that this is the first time Zoox has partnered with a third-party platform. So again, this is not about hard unit count. It is about Uber increasingly becoming the distribution layer even for AV developers that once looked likely to stay fully closed and vertically integrated.

Waymo: real U.S. deployment already happening through Uber

Waymo is different from the others because this is not just future intent. Uber and Waymo expanded their partnership to Austin and Atlanta, with Waymo vehicles available only through the Uber app in those markets. The original announcement framed the relationship around a dedicated fleet expected to grow to hundreds of vehicles over time. So this is less about a giant headline number and more about proof that Uber is already integrating live commercial AV supply into its network in major U.S. cities.

Avride: live proof of smaller-scale rollout

Avride is important mainly as evidence of operating cadence. Uber and Avride launched robotaxi rides in Dallas in December 2025, initially across a 9-square-mile service area. This is not a major disclosed fleet-scale announcement, but it does show Uber turning partnerships into real launches rather than just stacking press releases.

May Mobility: meaningful U.S. pipeline below the top tier

May Mobility is not as high-profile as Waymo, Zoox, or Nuro, but the economics matter. Uber and May said May aims to deploy thousands of AVs on the Uber platform over the next few years, starting in Arlington by the end of 2025. That makes it another substantive U.S. supply relationship, especially because it broadens Uber’s partner bench beyond the handful of headline names investors focus on most.

Subscribe now

A lot of commentary around AV tends to sloppily bundle “partnerships” together as if they are equal. They are not, some partnerships are real supply, some are geographic options, but Uber increasingly has both.

Why Uber may end up with better AV economics than the AV developers themselves

This is the key point investors should focus on. The biggest challenge in rideshare has always been demand variability. Morning commute is different from midday. Friday night is different from Tuesday afternoon. Airports behave differently than suburbs. Weather, events, and local density all matter. I’ve extensively written about fleet sizing and demand variability in the past:

A pure first-party robotaxi operator has a real utilization problem. Build too much capacity and your assets sit idle off-peak. Build too little and you fail riders at peak. Uber’s hybrid model is structurally better suited to that world.

AVs can be concentrated where autonomy works best and where utilization is highest. Human drivers can still absorb peak overflow, edge cases, bad weather, difficult geographies, or cities where AV supply is not there yet. Uber does not need AVs to replace the whole network. It just needs AVs to improve the mix of the network.

And if Uber can route scarce AV supply into the densest, cleanest, highest-frequency demand pools, it can likely drive materially better utilization than a standalone AV operator trying to build demand market-by-market from scratch. That is what makes Uber so dangerous here.

Waymo is still the current leader, but it may not be the biggest winner economically

To be clear, Waymo deserves the premium perception it has today.

It has the product.
It has the data.
It has the rider base.
It has the regulatory progress.
It has the most scaled operations.

But as the market broadens, Waymo also looks increasingly like a company testing multiple go-to-market paths at once. It has first-party operations in some markets. It has partnered with Uber in Austin and Atlanta. It has partnered with Lyft in Nashville for a 2026 launch. It is also expanding its manufacturing footprint and vehicle platform mix.

That tells me something important: even Waymo appears to recognize that AV scaling is not just about the autonomy stack. It is about matching that stack with the right operational and distribution layer.

In other words, the likely end state may not be one AV company vertically owning everything. It may be an ecosystem. And if that is true, Uber is very well positioned.

Tesla is still more narrative than scaled reality

Tesla remains the company with the most upside in the bull case and the least patience left in the bear case. Yes, Tesla has now started robotaxi rides in Austin without safety monitors in the car. That is real progress, but the broader picture still looks messy.

Reuters reported in late February that Tesla logged zero autonomous test miles in California in 2025, the sixth straight year with no such miles reported there, and that the company has not taken the regulatory steps necessary for a commercial robotaxi launch in California. By contrast, Waymo logged over 13 million autonomous test miles before securing its full commercial permissions in the state. That is not a trivial difference. That is the difference between a live commercial system and a story still trying to become one.

At the same time, Tesla’s core auto business is wobbling. Reuters reported this week that analysts increasingly expect Tesla to post a third straight year of declining deliveries in 2026, while capex ramps sharply and concerns about future cash burn rise.

So yes, Tesla still has the biggest theoretical prize if it can crack low-cost, vision-based autonomy at scale. But sitting here today, Tesla looks far less like a clean AV rideshare winner and far more like a company still trying to bridge the gap between product narrative and real commercial deployment.

That is what I mean by floundering. Not dead. Not irrelevant. Just still nowhere near as concrete as the stock and the discourse often imply.

Subscribe now

Lyft is in the most awkward strategic position

Lyft deserves some credit for at least making sure it has a seat at the table.

It announced partnerships with May Mobility and Mobileye in late 2024, launched its first May fleet in Atlanta in 2025, partnered with Waymo for Nashville in 2026, and even struck a Baidu partnership for Europe. But compared to Uber, Lyft still looks stuck between identities.

• It is not the AV technology leader like Waymo.

• It is not the moonshot vertical integration story like Tesla.

• And it does not have the same magnitude of disclosed AV pipeline as Uber.

The May Mobility rollout was important, but it did not come with the type of hard fleet scale disclosures that Uber’s Nuro/Lucid tie-up did. The Waymo Nashville deal is meaningful, but it is also incremental rather than market-defining. The Baidu deal is outside the U.S. and does little to change the domestic competitive picture.

So Lyft is not absent. But it is hard to look at the current setup and conclude that Lyft is driving the market structure. It looks much more like a company trying to stay attached to whichever AV suppliers will have it. That is a very different strategic posture than Uber’s.

By the end of 2027, the U.S. AV rideshare map may look completely different

If you zoom out, the likely shape of the market by the end of 2027 is becoming much easier to imagine.

• Waymo will probably still be the premium gold standard in actual AV operations.

• Tesla may still matter enormously, but it still has to prove it can move from small controlled pilots and repeated promises into scaled, repeatable, regulatorily durable commercial service.

• Lyft will likely still participate, but right now it looks more like a follower than a shaper.

• Uber is the one increasingly assembling the broadest demand and supply clearinghouse.

That matters because the eventual winner in AV rideshare may not be the company with the single best robotaxi. It may be the company that can best aggregate robotaxis, finance them, route them, fill them, and keep them busy at higher utilization than anyone else.

Today that company looks increasingly like Uber.

And if that is right, then by the end of 2027 the U.S. rideshare market may look much less like Uber vs Lyft and much more like Uber sitting at the center of a hybrid human-plus-AV network, Waymo supplying part of the premium autonomous layer, Tesla still trying to force its own vertically integrated path, and Lyft trying not to get marginalized.

That is a very different market than the one investors were underwriting even a year ago. And in that world, Uber may not need to build the best AV stack. It may just need to become the best place for every AV stack to work.

The obvious risk to this thesis is that if Tesla or Waymo drive vehicle costs low enough, Uber’s utilization advantage matters less than I think.

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

Actual vs. Incremental: The Chart Worth Staring At

Booking. sits at the top on both dimensions: 5.3% actual EBITDA margin and a 7.9% incremental margin. That incremental figure is the highest in the group. When a business is adding margin on new revenue at nearly 8 cents on the dollar, that’s a compounding story. The gap between where they are today and where they can go is a function of volume, not structural change (pending AI impact).

Uber presents a more interesting case. At 4.5% actual and 7.3% incremental, the gap is large (nearly 300 basis points) which implies meaningful operating leverage as the business scales. Uber has been on a deliberate path toward EBITDA discipline after years of growth-at-all-costs, and the incremental margin profile here suggests that path is working. Every dollar of incremental GBV is being converted at a much higher rate than the blended base. The delivery and freight businesses still compress the consolidated number, but mobility’s underlying economics are pulling the incremental rate up materially.

Airbnb is the puzzle. At 4.7% actual EBITDA margin, the absolute number isn’t embarrassing, but the 3.0% incremental margin is the second-lowest in the group. For a marketplace with virtually no physical supply, no employees operating the assets, and a brand that doesn’t require nearly the performance marketing spend of an OTA, a 3% incremental margin figure is hard to explain away. There’s a real question about whether the platform has hit an S-curve on core alternative accommodations and whether experiences and services can pick up the slack. Management flagged that 2025 was a heavier investment year and telegraphed margin expansion in 2026.

Subscribe now

The 2.9% club Expedia, Instacart, and Lyft share the same actual margin but diverge on incrementals. Expedia converts at 6.5%, Instacart at 5.4%, and Lyft at 6.1%. Those incremental figures are meaningfully above the base, which suggests all three still have operating leverage to harvest.

DoorDash has the lowest actual margin at 2.7%, with a 4.0% incremental. The company just announced it plans to ramp 2026 tech and autonomy spend. That investment cycle will likely compress near-term incrementals further. The thesis there is a multi-year one (international expansion, restaurant tech, autonomous delivery) and is real, but they require patience before the margin profile reflects it.

The broader takeaway: actual margins tell you where a business is today; incremental margins tell you the direction and velocity. A business with a low actual margin but high incremental (Uber, Expedia) is on a convergence path. A business with a high actual margin but low incremental (Airbnb) is worth watching for signs of structural compression. Booking’s combination of both is what makes it the benchmark.

New Tool: Query All Platform Aeronaut Transcripts via CustomGPT

I’ve built a CustomGPT on ChatGPT that lets subscribers search and query across the full archive of Platform Aeronaut transcript records. Rather than scrolling back through issues to find what a specific exec said about AV unit economics or margin guidance, you can now ask directly.

Try it here →

A few things to know:

• Authentication currently only works on desktop so if you’re on mobile, bookmark it and come back when you’re at a computer

• It’s early, and I want to improve it based on how people actually use it

• Feedback is genuinely useful if a query returns something off, or there’s a category of questions it handles particularly well, I want to hear it

The goal is to make the transcript archive as queryable as possible. There’s a lot of signal buried in what management teams say quarter over quarter, and pattern-matching across that manually is time-consuming. This should help.

Give it a spin and let me know what you think.

Performance & Valuation Snapshot

Note: Email renders these as images, click through for interactive filters or view on Platform Aeronaut.

What I Read This Week

• Congress probes AI pricing at major travel platforms: The House Oversight chair asked Uber, Lyft, Expedia, Booking.com, and Instacart to explain whether they use AI-driven “surveillance pricing” and to provide documents by March 19

• OTA stocks jump after report that OpenAI is backing away from direct checkout in ChatGPT: Expedia, Booking, and TripAdvisor rose after a Reuters-reported market reaction to The Information’s report that OpenAI is scaling back direct-booking ambitions inside ChatGPT

• GetYourGuide names a new CPO as the experiences market keeps scaling: GetYourGuide hired Rob Rekrutiak and said it surpassed €1 billion in revenue with 33 million experiences booked in 2025, reinforcing the size of the category.

• Kroger leans harder into partner-led last mile: Kroger’s new CEO said the retailer aims for e-commerce profitability next year by using stores as fulfillment hubs and leaning on Instacart, DoorDash, and Uber Eats for delivery

• Navan launches an AI assistant for unmanaged business travelers: Navan Edge targets travelers outside employer-managed programs with AI trip planning, itinerary management, and disruption support.

• Agentic AI in travel is running into a trust problem: A fresh industry roundtable featuring Booking.com, Google, Sabre, and Skyscanner framed the current bottleneck as consumer trust and operational readiness, not just model capability.

Transcript Highlights

Instacart (CART) Morgan Stanley TMT Conference

• Gen AI productivity gains that include a 40% higher output per engineer on average, top 10% engineers at 80% output improvement, and end-to-end project velocity up to 4x in some initiatives.

• Agentic interfaces and AI-driven discovery are expected to create incremental small-basket use cases while preserving large-basket weekly habits.

• Company prioritized four execution areas for 2026: marketplace growth, enterprise expansion, advertising scale, and AI/agent development.

Roblox (RBLX) Morgan Stanley TMT Conference

• Views AI as a tailwind and is investing aggressively in gaming-specific AI to accelerate creator productivity.

• Roblox is building interoperable tooling that allows creators to use both internal and third-party AI capabilities to produce content.

• Not dependent on hits, noting over 14 million experiences and that roughly two-thirds of engagement time in 2025 occurred outside the top 10 titles. The company reported that experiences outside the top 10 grew bookings ~55% year-on-year and pointed to strong Q4 bookings growth (63%) without a new viral hit.

Expedia (EXPE) Morgan Stanley TMT Conference

• Expedia reports roughly 2/3 of traffic as direct and is prioritizing diversification of acquisition channels including early AI agent partnerships.

• Views AI as an opportunity to diversify acquisition channels and improve product personalization, citing roughly 2/3 of traffic as direct and over 50% self-service. The stated approach combines experimentation with AI-enabled product improvements, self-service scale, and supplier onboarding enhancements

• Expedia reported a 10% increase in listed properties in Q4 and plans continued onboarding of chains, independents, and Vrbo inventory across 2026 and beyond.

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

• Ride-hailing platform buys parking marketplace

• Urban mobility meets urban parking

• Cross-sell airport rides with airport parking

All of this is true, but if you zoom out and view this through the lens of autonomous vehicles, fleet utilization, and unit economics, this is not simply a parking acquisition, it’s an infrastructure acquisition. More specifically it’s an AV parking and fleet management play.

Reframing Parking in an AV World

In a human-driver rideshare world parking is a friction. Drivers cruise, double-park, idle, and occassionally pay for short-term parking. It’s inefficient but it’s not Uber or the 1P provider’s problem since the driver bears the cost.

In an AV world, the fleet provider beats the cost whether it’s Uber, Tesla, or Waymo. The moment vehicles become autonomous and fleet owned (or fleet-managed), parking shifts form a nuisance to a core cost center.

If an AV sits idle 6 hours per day and urban structured parking averages $20–$40 per night, that’s $7,000–$14,000 annually per vehicle in parking alone. Across 20,000 vehicles, that’s a $140M–$280M annual cost center. Based on conversations I’ve had around trying to build out AV specific depots at scale, that can come down to $80-120 per vehicle per month, but that’s assuming opportunistic asset availability of 1,000+ vehicle parking lots.

Awhile back I put together a fleet optimizer system and one of the charts is vehicle status by hour:

But let’s try to shift our mind into 3D and think about how this demand is not only distributed across hours, but across geography. In the future when there are 14,000 AVs driving around Los Angeles where do they go during periods of latent demand? The obvious first answer is charging and cleaning depots. We’ll likely need 10-20 depots geographically distributed around a city with 500-1,500 parking spots each. But that’s naturally a bit less efficient as there’ll be significant deadhead driving to get to depots and there’s also the reality that there just isn’t enough large scale parking available in most urban environments.

How do you optimize depot location, sizing, and availability while working within physical constraints? I think SpotHero can be a big help with this.

Subscribe now

How SpotHero Can Help with AV Fleet Management

After a Waymo or Nuro AV drops off a rider and it’s 2am and demand is latent what does it do next? It can’t circle endlessly, it can’t double-park forever, and what if it’s clean and mostly charged already? Does it warrant driving to a depot? What if the depot is 25 miles away and Uber’s demand forecasting knows that there’ll be demand at 6am for rides in that location?

SpotHero can help provide ad-hoc on-demand parking for downtime for AVs to help optimize vehicle location and reduce costs. If Waymo’s are just randomly parking on the street during latent demand there’s the obvious risk to damage, but also think about the political or publicity issues. I lived in Somerville, MA outside of Boston for a decade and if you’re familiar at all with the area there’s a big reputation for people slashing tires or keying cars if you park in the “wrong spot” or you’re not “from the neighborhood”.

I think cities will continue to restrict curb space, regulate AV staging, impose idle restrictions, and generally crack down if we get anywhere near the scale forcasted for AV robotaxis. This makes parking a potentially valuable asset.

The optimal solution very well may be matching up periods of low demand for parking spot asset owners with AVs that need a place to sit and wait for additional demand. If you’re a business or real estate owner in a quasi-urban area and your operating hours are 9am to 5pm and you have 5-10 parking spots, those can be available through SpotHero to AV fleets just as latent/spare demand during periods of your choosing. It’s a win-win-win for the parking owner, the AV fleet, and SpotHero/Uber.

This doesn’t solve the charging and cleaning issue, that’ll need to be solved through actual depots with significant infrastructure, but SpotHero and Uber can help dial up utilization and fleet optimization through ad-hoc parking.

If Uber can integrate SpotHero inventory directly into fleet routing algorithms, parking stops become programmable. Vehicles can stage exactly where demand forecasting predicts a surge. Parking becomes part of dispatch logic.

Performance & Valuation Snapshot

Note: Email renders these as images, click through for interactive filters or view on Platform Aeronaut.

What I Read This Week

• Waymo expands robotaxi service to 10 U.S. cities: Waymo added Dallas, Houston, San Antonio, and Orlando, aiming to scale toward 1M+ weekly paid trips by end of 2026.

• DoorDash posts strong demand but warns of heavy 2026 spend: Q4 orders and revenue rose sharply; the company says unifying DoorDash/Wolt/Deliveroo onto one platform will be expensive and pressure near-term profitability.

• Booking CEO details the “connected trip” vision: Emphasis on seamless travel (automatic updates when disruptions hit) with AI as the glue across trip components.

• Uber to acquire parking-reservations app SpotHero: Uber adds in-app parking (commuters/events/airports) as part of its “super app” push; SpotHero operates in 400+ cities.

• Expedia’s annual filing calls out “agentic AI” as a competitive threat: Expedia explicitly flags AI-driven travel search/booking as a risk from faster-moving entrants.

• Winter disruptions renew focus on “agentic travel” workflows: Piece argues MCP-style coordination could help agents carry structured traveler constraints (identity/loyalty/payment) across systems.

Transcript Highlights

Airbnb (ABNB) - Q4 2025 Earnings Call

• A custom AI customer support agent is resolving roughly a third (about 30%) of support issues in North America and is reported to reduce resolution times.

• Brazil moved from a top-ten market to a top-five market and was the second largest contributor to first-time bookers in Q4 behind the U.S.

• AI-enabled product experiments (including AI search and sponsored listings tests) are underway, but no material sponsored advertising revenue is yet disclosed or banked into guidance.

Booking (BKNG) - Q4 2025 Earnings Call

• Generative AI integration into customer service contributed to roughly a 10% decline in customer service costs per booking

• Genius loyalty program Level 2 and 3 members represented over 30% of the active base at Booking.com and accounted for a high 50% share of room nights for the full year, up from the prior year.

• Social media investment grew 13% year over year in 2025 as the company leaned into new social channels with attractive ROIs

Instacart (CART) - Q4 2025 Earnings Call

• The 2025 new-customer cohort generated the largest GTV contribution since 2022 and cohorts are retaining at higher rates year over year.

• Restaurant integrations (including with Uber Eats and Grubhub embedding) materially contributed to mix and small-basket growth.

• Over the past year, average output per engineer is up nearly 40%.

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

I’ve taken that personal tool and productized it into a full fledged website: Platform Aeronaut Transcripts to help ease that scavenger hunt.

Subscribe now

What it is

Platform Aeronaut Transcripts is a workflow-first research site focused on public-company earnings calls that provides one place to monitor what’s new, read the call, skim a structured summary, grab the deck, and track what’s next.

At a high level, it’s one place to:

• Read and search transcripts in a clean format

• Use structured AI summaries as a fast first-pass

• Pull up the slide deck without hunting

• Monitor upcoming earnings dates

• Compare communication patterns across companies / execs / analysts

I’ve already built out 70+ companies, 1,400+ transcripts and 800+ summaries and growing.

Why I built it

Two reasons: one practical, one philosophical.

Practical: time and focus are the scarce resources.
The raw materials (transcripts, decks, dates) are abundant. The friction is in retrieval + formatting + cross-referencing. If you cover a basket of companies, the meta-work becomes its own job.

Philosophical: the edge is rarely “having the transcript.”
It’s being able to track narrative and intent over time and to separate what management said from what the market heard. That only happens when your workflow is consistent enough to notice the deltas.

How to use it (3 simple workflows)

Homepage → Company Page → Calendar

1) The “morning scan”

Open the homepage, and use it like a dashboard:

• What new transcripts dropped?

• What new AI summaries are available?

• What calls are coming up next?

• Any new decks worth pulling?

This is the “don’t make me think” entry point. The goal is to replace the mental overhead of “where should I look today?” with one page (for those who don’t have Bloomberg)

2) The “Company Page”

Every company page is designed to answer three questions immediately:

1. What just happened?

2. What’s next?

3. How has the stock moved on prior earnings?

From there, the company page becomes your hub: event history, links to the transcript, AI summary, and the deck (when available), plus a quick way to contextualize the market’s reaction.

Subscribe now

3) Personalization: Make it Yours (Watchlist + Alerts)

If you want the site to work “for you” instead of you checking it manually, log in and build a watchlist. Your watchlist powers the calendar filters and alerts do you don’t have to check manually.

From there you can enable alerts for new transcripts and new AI summaries and choose immediate, daily digest, or both.

The daily digest is designed to be a predictable “earnings inbox” that shows up on schedule and lets you plan your day.

The Fun Stuff

Statistics

Now that I have a giant database of transcripts, executives, Q&A, and other details I can compile statistics. There’s a lot of heavy lifting to do and I’ll continue to flesh this out, but for now I have tagged stats like what % of time is an executive’s Q&A response evasive vs a partial answer vs direct?

Login Using Your Substack Subscription

If you’re a Platform Aeronaut substack subscriber, you already have access. With all of the work put in I decided to gate the AI summaries and the presentations into behind a login.

At the pop-up just type in the same email you’ve subscribed to my substack with and you’ll get a magic link email to login, no additional effort required. If you’re not a subscriber simply subscribe and you’ll get automatic access. This is part growth hack and part a gut feeling of wanting some sort of reward for the effort put in. Call it a perk, if you support me you get the parts that take the most work.

User Requests

Coverage is never perfectly complete. So instead of emailing me “can you add X” or “can you summarize Y,” the site has visible request flows to request a new company coverage or request specific AI summaries when the transcript exists but the summary doesn’t yet. There’s a little bit of manual tagging required for new companies are first added, but I’m continuing to add names.

Roadmap

Next plans include something I’m really excited about: a transcript and summary MCP server that you can plug into ChatGPT or Claude. Rather than doing the heavy lifting myself and setting up a full AI chat interface into the website, I’m working on a server anybody can use that can do full expansive analysis across transcripts and time periods and companies while offloading the inference costs to your existing ChatGPT subscription.

There’s also some data quality issues around event dates for some tickers that I’m still working out, but if you notice anything obvious let me know.

Performance & Valuation Snapshot

Note: Email renders these as images, click through for interactive filters or view on Platform Aeronaut.

What I Read This Week

• Marriott says Google’s “AI Mode” will process hotel bookings: First major hotel chain to discuss building with Google’s agentic booking tools.

• DoorDashers paid to close Waymo robotaxi doors: New Atlanta pilot pays nearby couriers to shut doors so vehicles can depart.

• Tripadvisor flags AI Overviews drag: CEO says changing search (AI summaries) is pressuring traffic; options on the table.

• Waymo unveils 6th-gen robotaxi stack: Fewer cameras, upgraded lidar/radar; targets high-volume production and scaling.

• Uber Eats adds “Cart Assistant”: AI builds a grocery cart from a text or photo list; launches with major U.S. grocers.

• TikTok launches “Local Feeds” in the U.S.: Opt-in, GPS-based feed for nearby travel, dining, events and shopping content.

• Coveo launches a Hosted MCP Server: Enterprise-grade MCP hosting to wire LLMs into systems with governance.

• Chesky: Airbnb’s native AI is a moat: CEO argues rivals’ chatbots are copyable; building AI deep into product is harder to replicate.

Recent Posts

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

Bill Gurley has frequently asked the question of whether Waymo should build to peak demand or average? What makes economic sense? What is optimal from either an individual per ride basis or more importantly on a fleet-wide total profitability basis?

To answer that question I created a Fleet Optimizer within the website that is pre-populated with top 20 US cities/MSA data and computes based on population, density, trip distance and average speed among other variables. Below you can see the headline chart of optimal fleet size for cumulative top 20 MSAs in the United States that results in 319k vehicles:

In this model, “optimal” fleet size is defined as the point that maximuzes total fleet-level profit, not per-vehicle utilization and not per-ride margin.

High Level Conclusions:

• Optimal AV fleet sizing is not peak demand, but it’s also not average.

• For major MSAs, the economic optimum is 98-99% demand coverage

• Oversizing fleets destroys pricing power faster than it increases utilization

• National Top 20 MSAs imply ~319k vehicles, not millions

• AV unit economics only work with aggressive off-peak price differentiation

• The hybrid human+AV model matters most during ramp, not at maturity

Subscribe now

Fleet Size Optimization Drivers

In creating this I worried most about garbage in - garbage out in terms of model drivers and how to arrive at reasonable assumptions to create a demand curve. I also utilized research studies, NYC and Chicago taxi/rideshare data, and the ridesharing demand variability chart that Uber posted in their earnings presentation a couple quarters ago:

MSA Data

For each MSA I pulled down population, density, and other relevant statistical data and applied a regression against a variety of known data for Uber and Lyft rides today to spit out an average trip distance, driving speed, and deadhead % (time between trips). That provides an output for trip duration (23.5min for SF), trips per hour (2 for SF), and maximum trips per day (43.1).

Pricing and Demand Curve

What is most challenging is building up that actual demand curve and applying the right price elasticity. You can’t just apply one price elasticity to all of ridesharing as consumer demand and elasticity differs over the course of a day. A rider at peak time versus off-peak versus normal periods has vastly different elasticity.

The reality for human drivers is that they naturally self select and match supply to demand because when demand is low at 2am most drivers are also at sleep. With an AV fleet that no longer naturally matches up supply with demand you need to have greater price differential in off-peak hours vs peak than you would with a human driver.

After a ton of research and adjustments I settled on price elasticities that feel right and are research supported for ridesharing.

Additionally in the model I set up three default market phases that affect pricing, penetration, and addressable trips per capita (as well as some of the cost levers). I’ve defaulted it to “Mature” which implies an addressable TAM of 2.5 trips per day per person and AV penetration of 20%. We are nowhere near that today with most cities at 2-3% penetration and high density metros at 5-8%. But in a world where AV rides might cost $10-12 perhaps 20% is possible. If that feels too aggressive to you there are “Early” and “Hybrid” defaults or you can change assumptions yourself.

Pricing vs Utilization Tradeoff

The central tradeoff in AV fleet sizing is between utilization and pricing power and it is fundamentally non-linear. At small fleet sizes adding vehicles increases utilization and captures unmet demand, but eventually once the fleet approaches full demand coverage incremental vehicles no longer create meaningful new trips, they simply reallocate demand across more idle assets. At that point utilization falls faster than volume rises, as incremental trips are spread across an expanding base of idle vehicles rather than concentrated on high-value hours.

The optimal fleet size is one that sits where nearly all demand is served, but scarcity still exists often enough to preserve price discipline.

Unlike human drivers, AV fleets do not naturally exit the market when prices fall, so oversupply persists and price becomes the only clearing mechanism. This is also why cost optimization alone cannot fix an oversized fleet. Once pricing power is lost, no reasonable reduction in depreciation, charging, or maintenance can restore profitability.

Costs

For the cost structure I generally kept the Mature market phase default matched up exactly with the default for the main AV Rideshare Economics model. For earlier phases I’ve scaled costs assuming less cost efficiency.

One note is that I also made a few changes to the main AV Rideshare Economics model:

1. I added in Taxes, Fees, and Tolls as a COGS to the model defaulted to 6% based on feedback received. This was a missing portion of the model beforehand.

2. Lifetime miles default I changed to 300k based on additional feedback and research I received.

3. Fleet Management cost defaults went up particularly around cleaning ($20/day) and parking/real estate ($100/mo) based on research and feedback.

Optimal AV Fleet Size for San Francisco Bay Area

So with the explanation of methodology and modeling out of the way what does it tell us about the optimal fleet size for autonomous vehicle ridesharing in San Francisco?

Based on optimizing total profitability, at maturity the optimal fleet size for AV ridesharing in the San Francisco Bay Area MSA is around 13,600 vehicles. Beyond that you start to get diminishing returns from underutilized vehicles and you also get hit with too many vehicles = overcapacity = pricing pressure. Over the course of a week though how does the demand vs capacity profile look like?

Subscribe now

SF Bay Area Robotaxi Demand vs Capacity Weekly Profile

To answer Bill Gurley’s question of how to size a fleet, the answer is just shy of peak. If you optimize around having 1% of demand unmet and you can load charging, cleaning, maintenance, and downtime into off-peak periods then you are best served sizing your fleet to capture the vast majority of demand that exists.

To more accurately model this dynamic I had the model include active vs downtime vs idle vehicles by hour of day and allocate downtime to off-peak periods so that vehicles are available for as much peak demand as possible.

What about Pricing?

In this mature market phase scenario where does pricing sit? With an average price around $12 and trough around $6 you fall pretty squarely in the sweet spot for a fleet that Waymo or Tesla might run. Ideally you want your pricing below the above yellow line where human rideshare drivers breakeven. Below $16-18 per trip, a human driver will stop driving as it no longer makes financial sense.

It’s important to remember these are averages not absolutes and don’t encompass hyper-peak day activity or rain or snow or natural disasters that cause price spikes.

Who Wins: 1P vs 3P?

One conclusion is that fleet sizing logic favors operators who control pricing. Marketplaces that can dynamically blend human + AV supply are advantaged early. Fully autonomous 1P networks dominate only after scale and density.

Waymo’s recent fundraising rumors gives them the cash to fully pay for capex required to build out a US-wide network, but will they actually be able to replicate the scale and density required to be successful? We’re a long way from a mature market with >20% penetration of the ridesharing TAM and until we reach that point with at-scale costs and trips that cost 50% less than today, there’s a big disadvantage for a 1P only provider.

Will Autonomous Ridesharing Replace Personal Vehicles?

No, not in the short term. If you play with the pricing and cost levers of the model it’s almost impossible to see how Waymo or Tesla could reasonably charge $3-5 for an average trip to be equivalent to personal vehicle costs per trip. Even with optimistic assumptions around depreciation, maintenance, insurance, taxes/fees/tolls, and every other cost line item, you can maybe get cost per trip down to $6. The $3 robotaxi narrative is a fantasy under almost any realistic scenario.

The caveat I’ll make here is that there’s a scenario where a savvy operator could monetize their audience in other ways. If you price a ride at breakeven and let’s say in 5 years a ride structurally costs $5, how else could you make money? You could extensively advertise, do paid loyalty programs, or turn Waymo or Tesla into the Spirit Airlines of mobility with low base fares and upcharges for bags, priority, or amenities.

Fleet Optimizer Website: https://www.avrideshareeconomics.com/fleet-optimizer

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

The meta-point across all seven questions is simple: AI doesn’t need to “kill” a business model to permanently impair it. If you lose control of discovery, you lose pricing power, and in travel pricing power is the whole game.

Subscribe now

Does AI mean the OTAs are done for? Are the glory days over?

No, OTAs are not “done.” But the glory days are absolutely over. OTAs were built for a world where discovery lived in a browser, routed through search, and ended in a click. That structure created a compounding advantage: more demand → more supply → better conversion → more marketing efficiency → more demand. The problem is we’re now exiting the era where “ten blue links” is the primary UX for discovery. As discovery moves inside AI interfaces, OTAs get pushed down the funnel and down the value stack, and that’s where take rates compress.

I used the Yellow Pages analogy on stage for a reason. Yellow Pages didn’t vanish overnight; it threw off cash for years. But the market correctly understood that once distribution control was broken, the multiple was terminally impaired. That’s the OTA setup: still profitable, still relevant, but increasingly abstracted behind whoever owns the interface where intent is created and routed.

Will Generative AI become the primary front door for travel booking?

Yes, and it’s basically the continuation of Question 1. The key point isn’t “AI is cool.” The key point is that AI collapses the funnel. Discovery, planning, comparison, and eventually booking can happen in one conversational workflow, instead of the current tab-sprawl of: search → metasearch → OTA → supplier → back to OTA → price paranoia → “did I just click an ad by mistake?”

On stage I framed this as a consolidation of eyeballs problem. Over time, super-cycles reduce the number of “front doors.” There were ~170,000 travel agents in the U.S. pre-internet. That collapsed. Search consolidated to Google. OTAs consolidated to a handful of scaled winners. In AI, it’s increasingly a two-front-door world: ChatGPT/OpenAI and Google/Gemini. If the router of intent sits upstream at that scale, booking becomes a routing decision, not a browsing decision.

Is “zero-click” travel marketing an existential threat to the industry?

To the industry broadly? No. Travel demand doesn’t evaporate. But to the click-dependent distribution model that sits underneath OTAs and metasearch economics? Yes, it’s existential. OTAs were built as marketing machines. Search sends clicks. Clicks convert. Take rates sit on top of that flow. If the user gets an answer inside an AI interface and never needs to visit an OTA page, the substrate changes.

The uncomfortable part: OTAs aren’t structurally designed for an “answer world.” Their muscle memory is paid acquisition + conversion optimization on their own properties. If AI interfaces compress clicks dramatically (and early data suggests they do), you don’t just lose volume, you lose leverage. And when leverage shifts to the interface owner, margin structure becomes negotiable. Zero-click doesn’t mean zero value. It means value moves up the stack to whoever owns the interface.

Are we overestimating how fast consumers will let AI book travel for them?

No, I think we’re underestimating it. The trust hurdle is real, but trust isn’t linear. It’s lumpy. It feels impossible right up until it suddenly feels normal. People don’t wake up and delegate a $10,000 family vacation on day one. Delegation starts small: a repeat hotel, a simple domestic flight, a rebooking during a disruption. Once it works twice, the behavior shift is fast.

I used a Waymo analogy on stage because it captures the adoption psychology. The first time you get in, you’re tense. Five minutes later you’re staring at your phone like it’s an Uber. If a robotaxi can earn trust when the consequence of failure is catastrophic, travel booking can absolutely earn trust when the consequence is… annoyance and a customer service escalation. Humans are terrible at intuitively modeling exponential adoption curves and we’re about to relearn that lesson again.

Will AI travel slop trigger a “Fyre Fest of 2026”?

No, not one big, theatrical implosion that discredits AI travel. The risk is more subtle and arguably more important: homogenization. The two lines that matter are AI capability and AI personalization. If capability grows faster than personalization, the agent recommends the same ten hotels, the same three neighborhoods, the same “top experiences” to everyone. That crushes the long tail and accelerates overtourism. At scale, that becomes systemic.

I’m actually optimistic here because personalization is the natural counter-force. An agent in your pocket can learn your constraints, preferences, and tolerance for trade-offs far better than an OTA ever could. The future failure mode isn’t “AI lies and ruins travel.” It’s “AI makes travel boring.” The solution isn’t rejecting AI it’s making sure personalization outpaces raw capability.

Will independent hotels finally be able to go direct to consumer?

Yes, and if you’re an independent hotel group, now is the moment. OTAs have historically earned 15–25% commissions largely because they controlled discovery and standardized the marketplace. AI weakens that choke point. The biggest unlock isn’t “cheaper customer acquisition”, it’s richer offers. Hotels can expose structured, differentiated inventory and packages directly to agents: bundles, experiences, flexibility, loyalty perks, hyper-specific room attributes. OTAs flatten hotels into a template; AI can unflatten them.

The underappreciated catalyst is that AI collapses the hotel tech disadvantage. Hotel tech has been chronically underfunded and understaffed. But with modern AI coding tools, five developers can do the work of fifty. That changes what’s feasible for independents: better booking flows, better data exposure, better merchandising, better conversion, better servicing. Direct doesn’t mean abandoning OTAs, it means rebalancing power. A few points of mix shift is a step-function change in hotel economics.

Will vibe coding unleash a tsunami of startups that revolutionize travel?

I wish it would, but I don’t think the big wave is in consumer booking. AI makes building easier, but it also centralizes distribution harder. That’s the paradox: lower build cost doesn’t automatically create sustainable businesses when the “front door” is consolidating into a couple of superhuman agents. On stage I talked about my own experiment: I vibe-coded a lie-flat-seat “OTA” in ~1.5 days. It proves that building is becoming trivial, and it also proves how irrelevant building alone is.

The real opportunity isn’t “compete with Booking.com’s homepage.” The opportunity is selling into industry pain points: infrastructure, ops, supply tech, servicing, data normalization, and the messy physical reality layer where incumbents are slow and fragmentation is high. Ironically, vibe coding might be most disruptive inside the industry, not at the consumer front door. And yes, if a credible MVP takes 1–2 days, it does raise a very uncomfortable question about what 40,000 people at the major OTAs are doing when the consumer UX hasn’t meaningfully changed in a decade.

Bottom line

AI doesn’t “kill” OTAs. It re-prices them.

• Discovery is moving upstream into AI interfaces.

• OTAs get pushed down the funnel → take rates compress.

• “Zero-click” threatens the click-dependent arbitrage core.

• Consumer trust will flip faster than most expect.

• The biggest risk isn’t fraud, it’s homogenization.

• Independents finally have a real shot at direct mix shift.

• Vibe coding spawns a million toys…and a smaller number of real businesses (mostly B2B).

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email.

What I Read This Week

• Waymo opens Miami to the public: First riders are entering a 60-sq-mile driverless service zone, with airport service “soon.

• EQT leads $300M into AI-forward PMS Mews (valued $2.5B): Fresh growth capital earmarked for AI-driven operations and payments.

• Op-ed: MCP may add vendor-controlled complexity. Industry critique argues “layers of simplicity” could mean more cost/lock-in for hotels.

• Amazon shutters Go & Fresh stores, doubles down on delivery/Whole Foods. Physical footprint consolidates as grocery strategy pivots.

• Zipline raises $600M; expands to Houston & Phoenix. Big funding and new U.S. markets signal drone delivery scaling beyond pilots.

• NYC delivery worker rules take effect. Judge denies Uber/DoorDash bid to block tipping laws; DCWP confirms $21.44/hr minimum pay

• Uber backs Waabi’s robotaxi push. Up to $500M from Uber as Waabi pivots from trucks; plan calls for 25,000 AVs on Uber’s network.

Recent Posts

Transcript Highlights (Exec Signals)

1. About American: “We are not going to allow them to win a single gate at our expense in 2026.”

2. “The strategy we’ve had to build a revenue diverse, brand loyal airline...is remarkably resilient in tough times.”

3. “Economic gravity always wins in the end.”

1. “At the midpoint of our EPS and CapEx guidance, we would hit our 2027 goal to have total debt below $35 billion a year ahead of schedule in 2026,”

2. “It’s a combination of both macro and strategic efforts that we expect to drive improved profitability moving forward.”

3. “American invented airline loyalty, and the Advantage program continues to lead the industry.”

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

That gap is what motivated me to build AVRideshareEconomics. I wanted a way to move the conversation from narratives to numbers and force assumptions out into the open and let people stress-test them. If autonomous rideshare really is a multi-hundred-billion-dollar opportunity, the unit economics have to close. This model is an attempt to make that explicit.

Subscribe now

What the Website and Model Are

At its core, AV Rideshare Economics is an interactive, bottoms-up unit economics model for autonomous rideshare. It’s designed to let you toggle assumptions and immediately see how they flow through to per-trip economics, contribution margin, EBITDA, and operating profit. Rather than baking in a single “right” view of the world, the site is built around the idea that reasonable people will disagree on inputs and that the only productive way to debate those disagreements is to model them.

The framework supports both first-party (1P) and third-party (3P) operating structures, explicitly separating the economics of the autonomous vehicle provider, the marketplace/aggregator, and the fleet manager where relevant. The goal isn’t to predict winners, but to define the narrow corridors where each model actually works.

Who This Model Is For (and Isn’t)
This model is built for investors, operators, policymakers, and technologists who want to understand when autonomous rideshare works, not to argue that it must. It is not a valuation tool, a demand forecast, or a prediction of winners. It is a framework for pressure-testing assumptions.

Fleet and Utilization: The Real Unlock

If there’s one takeaway embedded throughout the model, it’s that utilization matters more than autonomy itself. Removing a human driver does not automatically make rides cheap. A robotaxi that drives a few hours a day is still an expensive piece of capital sitting idle.

The fleet section forces you to define:

• Vehicle purchase cost and useful life

• Annual miles per vehicle

• Average speed and miles per trip

• Daily downtime for charging, cleaning, and maintenance

From these inputs, utilization is derived as an output (not something you can hand-wave away). Push utilization higher and capital efficiency improves dramatically, but not without consequences. Higher utilization increases depreciation per day, compresses maintenance cycles, and raises operational complexity. The model makes those tradeoffs visible instead of implicit.

What Usually Breaks First
This is where much of the debate around companies like Waymo and Tesla ultimately converges: not just who has better autonomy, but who can sustainably operate vehicles closer to 24 hours a day without breaking the system. In most scenarios, the economics fail not because of autonomy quality, but because utilization, capital cost, or operational overhead fails to scale as assumed.

Subscribe now

Pricing & Marketplace Dynamics

Cost per Mile vs. Cost per Trip

One of the most common mistakes in autonomous rideshare discussions is mixing cost-per-mile logic with cost-per-trip pricing. Riders don’t buy miles; they buy trips. The model explicitly tracks both.

Cost per mile is driven by:

• Vehicle depreciation

• Energy

• Maintenance and cleaning

• Insurance and teleoperations

Cost per trip layers in:

• Average trip length

• Deadhead miles

• Marketplace fees and commissions

This distinction matters because short trips can look great on a per-mile basis while still being uneconomic on a per-trip basis once fixed costs are allocated. The model makes it very clear where that breakage happens.

3P Commission and Utilization Efficiency

The second pricing lens focuses on third-party marketplace dynamics. In a 3P model, the marketplace takes a commission, which at first glance looks dilutive. But higher utilization enabled by better demand aggregation can more than offset that take rate.

This is the crux of the argument from players like Uber: even if you give up 30–50% of gross bookings as a commission or revenue share to the 1P Fleet Operator, materially higher utilization and faster market expansion can improve absolute dollar economics. The model lets you explicitly test that claim by adjusting utilization uplifts, commission rates, and pricing power side by side.

Early data and disclosures suggest higher effective utilization in Waymo-on-Uber cities relative to Waymo’s 1P-only operations. The big question is how much higher utilization and better unit costs can an aggregator like Uber or Lyft drive than a 1P solution from Waymo or Tesla?

Costs and Cost Allocation

A major design choice in the model is explicit cost allocation. Rather than lumping everything into a single margin line, costs are broken out and allocated based on what actually drives them.

Key cost buckets include:

• Energy

• Maintenance and repairs

• Cleaning

• Insurance

• Teleoperations

• Support

• Platform and overhead costs

The allocation logic matters. Some costs scale with total system miles, others only with 1P-operated trips, and others are shared across the entire network. The model forces consistency so that per-trip and per-mile metrics reconcile cleanly to the annual P&L or per trip or per mile income statement. This avoids the common mistake of double-counting or under-allocating costs when mixing 1P and 3P operations.

Fleet Manager Economics

An area that’s often ignored in public discourse is the economics of the fleet manager. In many realistic scenarios, the autonomous technology provider will not directly manage charging, cleaning, parking, and maintenance. That role will sit with a fleet manager earning a return on capital and operations.

The model explicitly separates fleet manager revenue and costs, allowing you to see:

• Management fee applies to 1P or 3P operator for their services

• Cost sliders for those costs under the responsibility of the Fleet Manager

• Sensitivity if you click the on/off toggle to remove the fleet manager from the model.

This helps frame an important question: are fleet managers a low-margin utility, or a structurally attractive layer once scale is reached? The answer depends heavily on utilization stability and cost discipline, both of which are easy to assume and hard to achieve.

Vehicle Financing Economics

Finally, the model incorporates vehicle financing, which is often glossed over despite being critical. Autonomous fleets are capital-intensive, and who finances the vehicles (and at what cost) materially changes outcomes.

You can toggle:

• Loan-to-value assumptions

• Interest rates

• Amortization periods

• Who bears the financing burden (1P vs fleet manager vs shared)

Interest expense flows below EBITDA and Operating Profit by design, reinforcing the distinction between operating leverage and capital structure. This makes it clear when a model “works” operationally but fails once financing reality is layered in. I’m still thinking through how I want to integrate cash flows into this and the capital structure of each player but that’s a future add-on.

Crowdsourced Presets: Saving and Voting

One feature I’m particularly excited about is the ability to save and share presets. Any user can create a scenario (optimistic, pessimistic, or somewhere in between) and generate a link others can view.

More importantly, presets can be up-voted or down-voted. Over time, this creates a crowdsourced view of which assumptions the community finds most credible. Instead of arguing in threads, the debate happens through models. The highest-voted presets should converge toward something resembling collective wisdom, or at least clearly define where disagreement lives.

A Living Framework

This model definitely is not a finished product. As more real-world data emerges from Waymo, Tesla, Uber, and others, the default assumptions will be updated. Utilization curves, cost structures, and pricing power will all evolve, and the model will evolve with them.

My hope is that this becomes a shared sandbox for thinking rigorously about autonomous rideshare economics. If you think the model is wrong, save a preset and show where the economics break and share it with me. If autonomous rideshare is going to reshape transportation, it will do so because the numbers work. This tool is my attempt to keep that conversation grounded in reality.

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

What surprised me wasn’t that I could build a lie-flat-only OTA in a day, it’s that nothing structurally prevented it from existing before. The constraints weren’t product vision or capital, they were developer time, integration friction, and iteration speed. Those constraints are collapsing fast.

All of this took about one day’s work utilizing Claude Code, Github, Vercel, and the Amadeus self-service API for data. I have minimal coding experience beyond PHP back in the day and it spit out an impressively good copy of an OTA including some cool pages like Inspiration (pulling recently searched travel), AI Search (utilizing OpenAI API and facilitating search of the underlying data), and Lie-Flat Routes (pulling cached data from searches to demonstrate where and who flies lie-flat business class seats.

I will caveat that I haven’t hooked up the actual booking mechanism (either through Amadeus or sourcing a meta-esque solution through Kayak or direct with airlines) mostly because I don’t want to deal with the hurdle of contracts and legal setup since this is more playing around. Additionally, Amadeus is pretty ass-backwards from a technical perspective and the ability to move from limited test API data to production has been having technical issues for awhile and even the test API goes down sometimes so it’s a bit frustrating. Hence the actual flight results are limited right now, it’s probably 20% of actual flight availability and mostly close in with cached data but I’ll fix it as soon as Amadeus does.

Subscribe now

In terms of the learnings that are useful from an investor or industry observer perspective:

Data Integrity Varies Significantly

When I was looking at various APIs for flight specific data you get a huge range of data. Amadeus appears to have the most depth (they even have seat-tilt information by flight segment), while others like Skyscanner and Kayak are actually shockingly limited in the data they provide through API.

It’s no small wonder that Google doesn’t offer an external API for their travel data and closely protects it as I’m sure they’re doing a lot of data massaging. For the lie flat detector I was originally utilizing a call to a Seat Map API at Amadeus but that’s pretty rate limited and makes it difficult to limit when you get so many flight results back so I had to switch to a custom heuristic. I had to set a ton of rules utilizing data based on route, aircraft type, airline, segment duration, and more to properly try to predict lie flat seat availability.

Development Speed and Accessibility has 100x’ed

The fact that I was able to make this in a day and with a reasonable budget (maybe $50 including compute) means the ability for the average developer or consumer to develop a customized solution is more accessible than ever. You don’t like how Expedia displays or searches hotels or flights? Just make your own solution. That’s not going to be super impactful to Expedia’s business, but it’s interesting.

More important is what I believe to be the impact on the direct booking ecosystem. If the average hotel can now have a more robust home grown system, or it enables smaller companies like Mews or Directbooker to reach feature or speed parity with the OTAs on offerings, it could have an enormous impact on the speed that OTA disruption happens.

The Importance of Caching

I got a bit of an inside look at the importance of caching and what that means for result and price integrity. The information received per search is like a firehose, and the more specifics you want the bigger the data pull and it can quickly get overwhelming. Then you have to make cache trade off decisions of how frequently, how to real-time update during searches, and how much data to store. The good news is none of this is really terribly complicated and with AI can be optimized even in the average developer’s hands. I’d guess if I set a budget of $1k I could easily have a more real-time but optimized cache solution for tens of thousands of daily searches.

A single broad transatlantic search can return tens of thousands of fare/segment permutations once you explode it by cabin, aircraft, and connection logic. Without aggressive caching, you’re either bankrupt or blind.

Enabling AI is Easier Than Ever

The fact that I could exposure AI to my Amadeus cached results and enable searching and results in a decently enough user friendly UX means the agentic world is even closer than previously thought. It’s easy to envision a world this year that every hotel or airline has an AI agent that can communicate rapidly with caching and surface the most relevant data to other agents and consumers.

In that world, the battle isn’t OTA vs direct, it’s whose agent is trusted as the system of record. Discovery, pricing, loyalty, and servicing collapse into a single conversational interface. OTAs don’t disappear, but their value migrates from traffic arbitrage to infrastructure, payments, and exception handling.

Vibe Coding Changes the Cost Curve, Not Just the UX

It’s not about prettier interfaces or faster prototyping, it’s about eliminating the translation friction between intent and software. When a single person can go from idea to production-grade product in a day, the constraint shifts from engineering capacity to imagination. That dynamic disproportionately benefits narrow, high-intent use cases that were never worth building before, and it’s exactly where incumbents’ scale and distribution moats are weakest. I spun up a real time lie-flat map organized by hub and airline with one prompt, 5k tokens, and 2 minutes:

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email

What I Read This Week

• Google’s new UCP lets AI agents handle checkout: Google teases agentic checkout for partners (incl. Expedia/Marriott), pushing AI from “search” to “transaction.”

• KAYAK CEO tried to take company private: Booking’s metasearch unit explored a take-private; parent CEO Glenn Fogel vetoed.

• Robotaxi outlook: Waymo, Zoox, Tesla on Bloomberg: Leaders debate timelines, economics, and regulatory risk for autonomous ride-hailing

• Walmart × Wing expand drone delivery to 150 more stores: Drone drop-offs head to LA, Miami, Cincinnati, St. Louis; 270 locations targeted by 2027.

• Kroger goes nationwide on Uber apps: Nearly 2,700 Kroger-family stores now on Uber Eats/Uber/Postmates with promo incentives.

• Travel’s wallet-share is shrinking: New data: travel fell from 12% of consumer wallets (Jan ’24) to 8% (Dec ’25), raising stakes for conversion/upsell.

• Sabre invests in BizTrip AI: Corporate-travel-focused LLM startup gets a strategic stake from Sabre.

• UCP vs MCP: the race for an agentic commerce stack: PhocusWire breaks down how Google UCP and Anthropic’s MCP will coexist in travel workflows.

Recent Posts

Transcript Highlights (Exec Signals)

1. “We expect the transaction to be EPS accretive in the first full year post closing.”

2. “MSP will be a major strategic hub for the combined company.”

3. “Since the vast majority of our fleet is owned… the combined fleet will have significant embedded equity value.”

1. “The US economy remains on firm footing, and consumers continue to prioritize experiences with travel among the top spending categories,”

2. “We plan to grow capacity by 3% for the full year, with all new seat growth concentrated in premium cabins...”

3. “...this consistent and integrated strategy positions Delta with a sustained unit revenue premium of nearly 115% relative to the industry.”

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

And if we look at cumulative enterprise value it sits at $2.1T, up from $1.3T right before covid. Despite the volatility of covid, it ended up being great for leisure margins and growth. It shook the weak hands out, rationalized capacity, and allowed companies to push pricing. Regardless of this though, the multiples across the space do not appear terribly demanding. At a high level the sector looks fairly priced on headline multiples, but increasingly asymmetric beneath the surface.

EBITDA margins across the spectrum are now at a peak of 23% vs 20% long term average. Drilling into Free cash flow margins shows us sitting at 12% up from 8% long term average (pretty much where the market was pre-covid as well)

Subscribe now

I like to look at scatter plots of NTM revenue growth vs EV / EBITDA multiples to get a gut check of where various companies sit. As we sit today here’s the platform and leisure space scatter plot for names:

Investors tend to pay more for growth (although not always, quality of growth matters (take Shift4 for example). It’s a good way to mentally filter names and see how it shapes up vs peers then you can do your own qualitative overlay. For example MakeMyTrip trades at a premium more due to India than quality of the business. Hertz business is theoretically temporarily impaired. Trivago has existential threats to its existence despite solid top-line growth.

OTA & Hotels

One interesting call-out is within the hotel & OTA space. We’ve seen EBITDA margins (and FCF margins as well) just absolutely explode from historical levels. These businesses generated low 20% EBITDA margins pre-covid, and have just been on a historic run with NTM consensus EBITDA margins now sitting at 29%!

But despite this we’ve seen multiples generally at the same levels as they were in Jan 2020:

So you have great margins, multiples in-line with historical levels, slightly slower growth, and a potential risk (at least on the OTA side) from AI disruption. I’ve been adamant that margins at the OTA level will likely go down over time as AI adoption and capability accelerates.

Cyclical Risks: demand normalization, ad efficiency plateauing
Structural Risks: AI agents collapsing discovery + servicing layers

If the middle man layer gets collapsed (similarly to what happened to the GDS layer during the age of the internet), are we likely to continue to see the OTAs put up the margins that they do? Does more get re-allocated to the direct suppliers and property owners? As a side note I’m going to be doing a debate-style discussion at Skift Megatrends 2026 around travel and AI disruption.

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email.

What I Read This Week

• “Travel Slop” is here: Skift warns AI-generated itineraries and content are flooding travel, risking brand trust if quality controls lag.

• MCP shows up in 2026 tech roadmaps: Hospitality Investor flags MCP as the standard to connect hotel data to AI agents; execution details still fuzzy.

• Amazon’s Alexa+ to integrate Expedia: Travolution reports a 2026 integration that makes Alexa+ a bookable travel assistant via Expedia’s inventory.

• Uber, Lucid & Nuro unveil a robotaxi at CES: Production-intent AV built on Lucid Gravity with Nuro autonomy and Nvidia Thor; Bay Area testing underway.

• Waymo maps out 2026 city expansions: 9to5Google tallies upcoming markets including Dallas, Miami, Nashville (with Lyft), DC and more.

• New California law forces real refunds for missing items: AB 578 now requires food-delivery apps to refund undelivered items and disclose fees clearly.

• Action-taking AI needs guardrails, says WaPo op-ed: Great primer on risks as AI agents start booking travel and spending money on users’ behalf.

• Leaders’ 2026 predictions: Priceline, Cloudbeds and more share near-term AI, distribution and demand outlooks.

Recent Posts

Transcript Highlights (Exec Signals)

1. Lyft will invest $10–15 million to construct a purpose-built Waymo depot in Nashville: “We’ll invest about $10 million to $15 million in getting that done.”

2. “Riders who come to the platform through a partnership tend to take a higher mix of more profitable rides.”

3. “AVs are really opening a new lens in terms of bringing riders into the marketplace.”

1. “We saw an inflection point in the U.S. direct channel... a very, very healthy indicator for the future.”

2. “We’re growing faster in flight tickets than any airline... faster in rental car days than any rental car company... faster in Attractions..”

3. “Alternative Accommodations… will continue to outgrow our Traditional Accommodations.”

4. “We outgrew the U.S. in general from a room night perspective in the third quarter by quite a wide margin.”

5. “We are the largest technology travel platform outside of Mainland China by really a length.”

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

I’ve written in the past about how hotels can thrive in an agentic AI world as we move from SEO to MCP, as well as the challenges the OTAs will likely face but I wanted to spend some time doing an overview of what technological and product changes the CRS, PMS, and Booking Engine vendors themselves are doing to enable the hotels themselves to take control.

At a super high level there’s a few definitions I’d like to provide first:

Property Management System (PMS): the hotel’s system of record for on-property operations. Manages guests once they exist as a reservation and responsible for everything that happens between booking and checkout. In an AI-first world it enables reservation modifications, amenities, inventory checks, and actions.

Central Reservation System (CRS): the hotel’s commerce and distribution system that manages rates, availability, and booking rules. It’s the revenue operating system. The CRS creates the booking and pushes it to the PMS for fulfillment.

Booking Engine (BE): the hotel’s direct-to-consumer storefront and transaction UI.
It sits on top of the CRS and converts real-time rates & availability into guest-facing offers. The BE orchestrates pricing display, room selection, upsells, payment capture, and confirmation.

MCP (Model Context Protocol): is the emerging standard that enables AI agents to call tools, query APIs, and execute transactions. Effectively giving booking agents direct programmatic access into a hotel’s systems.”

CRS = Where AI Agents “shop, compare, select, and book”
PMS = Where AI Agents “service, modify, fulfill requests, and manage”
BE = Where AI Agents “display, package, upsell, and transact”
MCP = Where AI Agents “access, understand, and digest”

Subscribe now

With that said I’ve put together a hotel tech stack graphic for distribution with the general flow from guest to hotel so we can identify where the best positioned players are for an AI Agentic future.

When you zoom out at the full distribution flow, the misalignment becomes obvious: the top of the funnel is built for human browsing, not agentic transactions.

Who is Best Positioned to Enable MCP Style AI Bookings?

If you look at the above diagram the further down the tech stack you get the richer, more accurate, and most up to date data you get, and those are the players who are best positioned to enable MCP style AI Bookings direct.

The CRS / Booking Engine / PMS is where the hotel controls rate, availability, inventory, policies, room categories, content, LOS rules, restrictions, and real-time booking workflows. This is exactly the data an MCP needs to:

• Query availability

• Reason about room types

• Price compare across dates

• Evaluate cancellation policies

• Transact the booking

• See customized offers/packages

You don’t get the full fidelity from OTAs, TMCs, metasearch, or GDS. Top of the funnel intermediaries are structurally misaligned with MCP-native booking execution. They worked in a pre-AI age of the internet because of disaggregated supply and tech-challenged hoteliers who didn’t know how to best display their data.

OTAs don’t own the ARI, they use cached availability, weren’t built for reasoning-based agents, sit outside the canonical data path, and pricing logic is opaque.

#1 CRS (Central Reservation System)

This is the canonical source of truth for ARI (rates, availability, inventory). They already power Brand.com, mobile apps, GDS feeds, OTA extranets, wholesales, TMC links and call centers. This means one CRS integration gives an MCP agent the ability to book at any hotel distribution endpoint. Additionally the CRS already has the transactional logic to perform the actual reservation commit + confirmation and CRS vendors already expose APIs ready-made for MCP adapters.

An MCP powered by the CRS gives AI Agents full life-cycle visibility:
Availability → Booking → Modify → Cancel → Loyalty → Upsell

SynXis CRS: Hospitality Solutions (Sabre sold this to TPG at the worst time) has rolled out SynXis Concierge.AI and an AI “Booking Agent” embedded in the SynXis Booking Engine, a gen-AI chatbot that recommends rooms, answers questions in 50+ languages, and completes bookings on top of SynXis APIs.

Amadeus iHotelier / CRS: Amadeus positions iHotelier as a highly personalized, data-driven CRS + booking engine and has publicly said it wants to be the real-time pricing and content provider to AI platforms rather than compete with them, ie they’re building the pipes MCP agents would hit for live rates/availability.

Cendyn (Pegasus CRS + CRM): Cendyn combines Pegasus CRS with an AI-enhanced CRM/CDP that uses AI-powered insights for personalization and has explicitly said it’s investing in APIs and MCPs to feed “AI helpers,” making it the only one today name-checking MCP directly.

SHR Windsurfer: Markets an “AI-driven booking engine” and CRS with an AI assistant that auto-generates content/translations and optimizes UX, so they’re already using AI inside the CRS/IBE and would be relatively straightforward to front with an MCP agent.

Conclusion: CRS vendors are the new gatekeepers of AI travel because they own the canonical ARI dataset, the pricing logic, and the transactional rail, which are the three ingredients AI agents require to book autonomously.

CRS vendors have the right economic incentive: unlike OTAs, they’re paid by the hotel and are not built on arbitrage or consumer acquisition spend.

#2 BE (Direct Booking Engines)

Booking engines (TravelClick, Mirai, Avvio, Mews BE, Cloudbeds BE are the closest to the guest-facing UX but still depend on the CRS or PMS for availability, pricing, and rules. They are well positioned to expose AI-native storefront UIs but they’re not the source of ARI truth.

CRS = the source of truth
BE = the source of UX

In an AI-mediated flow, the guest never sees the booking engine, an MCP agent may call its endpoints directly. AI is compressing the gap between CRS and BE. Many booking engines may evolve into API layers rather than UI layers, with MCP servers orchestrating the commerce flow instead of a traditional web interface.

SynXis Booking Engine: SynXis BE now includes gen-AI Booking Agent / Concierge.AI, so the booking engine itself exposes a conversational UI that can handle search, recommendations, and checkout which is an agentic booking flow you could either embed or call from MCP.

Mirai Booking Engine: Mirai has launched Sarai, an AI agent that chats with guests, routes them into the Mirai booking engine with trackable links, and is experimenting with a 3D-native booking UI. So they’re actively treating BE as an AI-first storefront, not just a form.

Cloudbeds Booking Stack (BE + Platform): Cloudbeds’ unified platform includes a direct booking engine and is now wrapped with “Signals,” a causal AI / gen-AI layer plus AI messaging and voice concierge that can answer questions and complete bookings, making it a natural candidate for MCP-style agents to sit on top of.

Conclusion: BEs are where the MCP conversation experience can be surfaced, but the CRS is where it is best executed.

#3 PMS (Property Management System)

PMS is not the best place to run MCP-driven OTA bypass because it’s operationally focused and often syncs ARI from the CRS or RMS systems and doesn’t handle multi-channel distribution. PMS should receive the reservation but not handle the AI booking logic.

But while PMS is the wrong place to book, it’s the right place for an MCP server to look for servicing. Whether it’s modifications, extensions, mid-stay requests, loyalty benefits, room moves, upsells, or in-stay commerce, this can all be managed through the PMS and so it stands to see a bit of a renaissance as well. In an AI-first stay, most guest interactions (extensions, upgrades, chat-based requests) flow through the PMS, not the CRS.

Mews PMS: Mews is leaning hard into AI with “AI Smart Tips” and AI-powered guest insights (summaries of reviews, stay history, lifetime value) plus a broader “agentic AI for hotels” narrative, so their PMS is already exposing AI-enhanced servicing and is philosophically aligned with MCP-style agents.

Cloudbeds PMS: Cloudbeds bills itself as an “AI-powered hospitality platform” and uses its Signals model plus gen-AI for forecasting, revenue intelligence, and AI messaging / voice concierge that can take calls, answer questions, and handle bookings, so PMS + distribution + guest comms are all being instrumented for agentic workflows.

Oracle OPERA Cloud: OPERA Cloud has a built-in Digital Assistant chatbot for natural-language operations tasks and offers a robust Integration Platform (OHIP) that third-party AI tools like Myma.ai and Canary already use for automated guest messaging and personalization, making it more of an “API fabric” that MCP agents could plug into than a branded AI agent itself.

Conclusion: Hybrid/vertically integrated players like Mews and Cloudbeds could attempt this given more robust RMS and CRS offerings natively integrated, but traditional PMS vendors like Oracle are not structurally aligned with distribution innovation.

The big shift underway is that AI collapses the layers of hotel distribution

OTAs won the web era because guests needed aggregation, comparison, and UX.
But AI agents don’t browse, they transact. They go directly to the systems with the richest, most accurate, most up-to-date data. And that’s the CRS.

The vendors who sit closest to the canonical source of truth for rates and availability are the ones best positioned to power MCP-native bookings. Over the next five years, the influence of CRS vendors will grow dramatically, booking engines will evolve from interfaces to APIs, and PMS platforms will become the operational brain for AI-managed stays.

Implications for OTAs:
OTAs don’t disappear, but their role shifts from primary booking originators to marketing surfaces. Their margin structure becomes pressure-tested as hotels regain control of direct programmatic access.

If OTAs were the UI layer of the last generation, CRS will be the API layer of the next. Hotels that understand this shift early will own their distribution economics again, maybe for the first time in 20 years.

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

When non-tech firms are diluting at <0.5% levels and big tech is doing so at ~1% the incremental FCF growth over 5yrs to offset that additional dilution is <5% which is a pretty reasonable deal.

But look at Pinterest and Snap, with Snapchat diluting >6.5% the incremental required FCF growth just to offset that dilution is >40%. Do longer term investors think that the high dilution levels at that company are justified by expected FCF or EPS growth?

Subscribe now

After publishing my Q3 deep dive on dilution metrics for technology last week, I wanted to post a few more charts on specific companies that were interesting:

Airbnb: Worsening Dilution from New Grants

For a company that burst onto the public markets and post-covid hit long term margin targets well ahead of schedule, the dilution metrics at Airbnb have continued to worsen. I weigh trailing dilution lower than the other metrics as it’s fairly backward looking, but we’ve seen forward dilution continue to increase from <2% in 2024 to >2% so far in 2025. They’re undergoing a heavy investment cycle into experiences and services, but there’s definitely not a lot of discipline.

Pinterest: Recovered from 2022 but Worsening

Pinterest dilution in 2022 was particularly egregious among internet businesses but it’s been slowly creeping up again. YTD 2025 they’ve issued 26m RSUs/options versus 18m total for FY24 and FY23. And that’s with a stock that’s been dead money since 2023.

Snap: Dilution Machine Masquerading as a Public Company

Long held as one of the most egregious diluters in the public markets, Snap continues to worsen approaching 2023 levels of dilution. They’ve issued 93m net options/RSUs YTD 2025 through Q3 versus 50m for all of 2024.

Infra Software: High Growth but High Dilution

Ignoring the 2023 chaos of Twilio slamming into a post-pandemic wall and then revamping the cost structure and dilution of the business, generally infrastructure software is diluting between 2-5%. Utilizing my first chart at the top you could justify Snowflake dilution given 27% top-line growth and similarly for Datadog, but Confluent is particularly bad for a mid-teens grower (on consensus), and Twilio remains ugly for a 10% grower.

All of this can be rolled up into a dilution vs revenue growth chart with some nice bubbles breaking companies down into potential dilution problems, dilution issues today, and what I’ll call the death zone of high dilution and low top-line growth. Companies in the death zone will find it hard to escape the weight of dilution around their neck and deliver shareholder value.

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email.

What I Read This Week

• Barry Diller on Google, AI & the future of travel search: The IAC/Expedia chair weighs in on AI-era distribution dynamics at Phocuswright.

• Sabre unveils Concierge IQ (gen-AI chat for airlines): A direct-channel assistant to plan, book, service and redeem loyalty in one conversation.

• Waymo approved to expand across Bay Area & SoCal: Big California footprint green-lit; CPUC approval still needed for paid rides in new zones.

• Tesla gets ride-hailing (TNC) permit in Arizona: A necessary step toward robotaxi ops; not an approval for driverless service.

• Kroger to shut 3 Ocado CFCs; pivots to Instacart/DoorDash/Uber: $2.6B impairment now; e-comm profit target +$400M by FY26;

• Peek raises $70M; buys ACME Ticketing & Connect&GO: Capital goes to AI and category consolidation in attractions/experiences infra.

• Why a Grab–GoTo merger may be too late: Strategic/market realities complicate long-rumored Southeast Asia superapp tie-up.

Recent Posts

Transcript Highlights (Exec Signals)

1. “More number of monthly active users in the first 9 months… twice the number compared to 12 months of last year.”

2. “When we generate a dollar of efficiency… the best use of that dollar is to put it back because that drives more profit dollars in the future.”

3. “We are replatforming our underlying tech infrastructure… the larger portion of our investment… The new platform natively comes with AI capabilities… speeding up development and personalization.”

1. “Travelers are increasingly willing to invest in high-quality travel experiences, highlighting strong consumption power.”

2. “Our international bookings grew by around 60% year over year... demonstrating robust growth despite macroeconomic uncertainties.”

3. “In the long run, we do not see any structural limitations to our profit margins.”

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

Looking at Big Tech dilution specifically, we have three main components driving how I calculate dilution:

Trailing Dilution: Shares issued for RSUs/Options divided by shares outstanding
This is essentially backwards looking for past grants, given stock comp tends to vest over 4 years it’s a moving average of past decisions.

Forward Dilution: Net grants of RSUs/Options (granted minus cancelled) divded by shares outstanding
This is decisions made today on grants and RIFs and cancellations that will impact future dilution. When management teams say things like “Our dilution run rate is <2.5%” it often refers to this.

SBC Dilution: Stock-based compensation expenses divided by market cap
My least favorite method because it utilizes valuation and SBC expense itself isn’t a perfect metric, but it’s a good gut check.

Looking at Big Tech specifically I want to call out a few things:

1. Trailing dilution is jumping because of Tesla’s CEO award for Elon. This isn’t the mega package, but rather 96m shares issued to Elon in the short term:

   

2. Forward dilution is jumping up because of Meta specifically. YTD 2025 through Q3 Meta has issued 49m RSUs net of cancellations versus 38m net issued for all of FY2024. AI engineers are expensive.

3. SBC dilution is coming down mostly because of stock market price expansion. If the market pulled back 10% this would shoot right back up because SBC dilution is a combination of trailing SBC expense and current market cap. It’s not my favorite metric but it’s decent as a subcomponent of my total dilution metric.

Subscribe now

My favorite chart to try to separate and see outliers and outperformers on dilution is this updated chart on tech company dilution versus market capitalization:

Naturally smaller companies in technology are seeing higher dilution. That can be for a few reasons:

• Post growth laggards like TRIP, BOX, BILL and GTM

• Hypergrowth at any cost without a focus on dilution like SNOW and MDB

• Generally worse business models than the Mag 7 with a higher employee count requirement like W, PINS or LYFT.

So the above chart and analysis is a more fair way to cohort companies based on dilution versus the market capitalization/size. Just note when you look at the mega caps like META how much more they’re diluting than NVDA, MSFT, and AAPL.

Cohort Leaders and Laggards

Here is a non-exhaustive list of some of the cohorts and individual companies charted out in some of the above graphs. I’ve highlighted in red those companies that are notably diluting higher than their peer set.

Tech Company Dilution Leaderboard

You can find the leaderboard from 1Q25 in a previous post. The most notable change at the top are that NFLX has taken over the top spot from BKNG.

Generally the way to read this is that names in the top couple groups are doing pretty good all else being equal. Those in the bottom half are doing pretty poorly. The numbers are where they rank in total and I’ve highlighted the top and bottom decile for each category.

Notable Y/Y Changes in Leaderboard

Since 3Q24 we’ve seen some drastic changes in the dilution leaderboard. HOOD has skyrocketed to 4th place from the middle of the pack while META has plummeted from middle of the pack to one of the worst across all of tech.

Tickers Mentioned:

The information presented in this newsletter is the opinion of the author and does not necessarily reflect the view of any other person or entity, including Altimeter Capital Management, LP ("Altimeter"). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for information purposes and should not be construed as an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

With Q3 earnings nearly wrapped for the major rideshare, delivery, and travel platforms, we’re starting to see a clearer picture of who’s compounding operational leverage and who is hitting a ceiling. Below is one of my favorite charts that captures the state of play: EBITDA margins (as a % of GBV) versus incremental margins across the Experience & Mobility ecosystem

Incremental margins show how much profit falls through from each new dollar of growth. Plotting that against absolute EBITDA margins gives you a quick read on who’s expanding efficiently versus who’s buying growth. Caveat that Instacart hasn’t reported yet so that’s still TTM through Q2 2025, but there’s a couple interesting takeaways:

Incremental margins for the majority of the space are super healthy

Booking continues to defy gravity with 5.3% TTM EBITDA margins on $185B of GBV, and still expanding. For a company that should be in harvest mode, it continues to compound like a growth stock.

Uber, Expedia, Instacart, Expedia, and DoorDash I would characterize as very solid. There’s a bit of reinvesting into the product at each, but incremental margins > actual margins = margin expansion which is what matters.

What is going on at Airbnb?

While they’re certainly re-investing into the product and trying to re-launch experiences and services and create a host marketplace, the incremental margins at Airbnb are horrendous and have consistently been getting worse:

And this is while the business is essentially growing in-line with peers at Expedia and Booking. It’s no wonder that the stock has been dead money with slower growth at industry average and no margin expansion:

Is the S-curve for alternative accomodations over? Are Booking and Expedia just getting better at competing for supply and customers? Is is a struggle of affordability and value for STRs versus traditional hotels? Is it lack of hyperfocus on the cost structure? It could be a bit of all of it but it’ll be interesting to see what/if Brian Chesky and the team can do over the next 12mo with experiences and services and accelerating the core alternative accomodations market.

On the Q3 2025 call, they did hint at margin expansion next year and 2025 being a heavier investment year vs 2026:

”And as we look to ‘26, we anticipate that, one, obviously, we’re scaling the revenue associated with those businesses. And while there is ongoing investment, we don’t have the same heaviness of the kind of first year launch. So you should anticipate that across experience services across hotels, across AI, we will be investing in those next year to drive growth.”

Looking at FCF Margins: Lyft Still Way Overearning

I like to look at EBITDA margins (to GBV) vs FCF margins to see if there’s any outliers, and Lyft remains the big one, generating way more FCF as a % of GBV than they are EBITDA. There’s just a significant amount of working capital benefit that Lyft is accuring that none of the other similar companies are seeing. I’d expect this to reverse and normalize over time but something to keep an eye on.

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email.

What I Read This Week

• Google’s AI Mode can now book tickets & appointments: Google expands “agentic” capabilities inside Search from info to transactions.

• Amazon sues Perplexity over agentic shopping: Case spotlights governance risks as AI agents automate e-commerce workflows (relevant to travel commerce too)

• Oversee & FCM add NDC “reshopping”: New corporate-travel capability auto-optimizes fares post-booking.

• Waymo to expand robotaxis to Las Vegas, San Diego & Detroit: Biggest footprint yet; new Zeekr-designed vehicles coming

• Tripadvisor to lay off ~20% amid TRIP–Viator merger: Boston Globe reports several hundred roles affected as brands combine.

• Uber Eats × Toast global partnership: POS-to-marketplace tie-up lets restaurants run Uber ads/promos from Toast.

• Kroger extends Instacart as primary delivery partner: Expanded AI-powered collaboration aims at faster fulfillment.

• DoorDash will ramp 2026 tech & autonomy spend: Wall Street balks at heavier investment in robots and unified platforms.

• How NYC became a nightmare for Airbnb: Enforcement of Local Law 18 slashes listings from ~60k (2018) to ~3k in 2025

From My Recent Work

Transcript Highlights (Exec Signals)

1. “Our philosophy has always been to reinvest back in the business”

2. “We have the largest audience with the most number of frequency.”

3. “Overall unit economics are improving across the business”

1. “The average cross-platform consumer is spending three times more than single-product users”

2. “Our partnership with Waymo continues to be excellent from an operational standpoint.”

3. “AV is not profitable today… but we expect that AV won’t be profitable for a few years.”

1. “Our win rates...are up year over year against all major competitors”

2. “It took us more than 10 years to reach our first billion in ARR, and just two years to double it.”

3. “Q3 in the summer was GPV per location exceeded expectations.”

From My Recent Work

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

Uber’s latest wave of AV partnerships continues the deluge of announcements and drive towards being the dominant autonomy platform. The company recently announced an exclusive AV partnership with Nuro and Lucid with ongoing Lucid Air pilots in premium ride-hail tiers. But the more strategic move is Uber’s deepening integration with Stellantis and Nvidia, both of whom it’s collaborating with to standardize the data and simulation layer for future robotaxi deployments. Uber’s AV team is already leveraging Nvidia Drive Sim and Omniverse Cloud XR to model traffic behavior, and Stellantis has begun feeding those simulation data sets into its own in-vehicle stack built on Nvidia Drive Thor. It’s a closed feedback loop: Uber provides the real-world telemetry and consumer demand, Nvidia provides the compute, and Stellantis builds the production hardware.

For Stellantis, the partnership is existential. The group plans to roll out Drive Thor-powered ADAS and Level 3+ autonomy across 14 brands starting 2026 (Jeep, Chrysler, Ram, and Maserati among them) turning its fleet into a software-defined network of potential Uber endpoints. Nvidia’s role here goes beyond chips; it’s becoming the operating substrate for automotive AI, similar to how AWS underpins cloud computing. With Uber as a downstream mobility partner, Nvidia gains access to petabytes of anonymized ride-hail data that can train perception models at scale, while Stellantis secures an immediate commercial channel for its autonomous-ready vehicles. It’s a three-way flywheel: data → inference → deployment → monetization.

At the same time, the U.S. testing footprint is widening. Waymo is rolling out in Denver, expanding its 1P operations from California into a high-altitude, four-season environment. Zoox and Waymo are both scaling in Seattle, a key testbed for lidar performance in rain and low-light conditions. The Waymo × Lyft partnership in Nashville aims for 2026, and Zoox has announced launches and testing in Austin, Miami, and DC as it transitions from validation to early commercial service. Taken together, these deployments underscore a broader shift: the autonomous race is no longer about who perfects the robotaxi first, but who builds the connective tissue between AI models, carmakers, and consumer demand. Uber’s alignment with Nvidia and Stellantis may quietly be the most durable architecture in that ecosystem.

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email.

What I Read This Week

• Stellantis teams with Nvidia, Uber & Foxconn on Level-4 robotaxis: Production targets 2028; Uber to seed initial fleets.

• Uber to take on Waymo in SF with Lucid/Nuro robotaxis: Plans point to late-2026 rollout and a broader AV scale-up.

• The Next Shift in AI Distribution: From MCP to A2A: Why it matters for hotels

• Grubhub pilots sidewalk robots with Avride in Jersey City: Urban navigation & reliability under test; broader city rollout eyed.

• Expedia Group: new study on content & AI: Video drives decisions; travelers open to AI but still want human input.

• Google hints at a deeper push into travel with agentic AI: Focus is on more “helpful” planning experiences that do more than search.

• Tesla expands Austin robotaxi area to ~245 sq mi: Fourth expansion since launch; still limited fleet and waitlist.

Poll of the Week

From My Recent Work

Transcript Highlights (Exec Signals)

1. “The reality situation is that bookings have grown faster than our ability to deploy the appropriate growth investments.”

2. “We estimate now that 3.2% of global gaming bookings or revenue is going through Roblox, and that’s up from 2.3% last year.”

3. “We believe there’s a wonderful opportunity to share and develop what we believe is going to be the industry standard in communication for social and mobile apps,”

1. “As we enter the fourth quarter, we continue to observe stable levels of global leisure travel demand.”

2. “What we are seeing in terms of traditional search that we still see volume growth.”

3. “Connected trip transactions...grew mid-20% year-over-year in the third quarter.”

4. “Genius members book more often, convert at higher rates, and cancel less.”

1. “We are not just a rental car company. We are a service company delivering a dependable product.”

2. “Our focus has to be if we want to offer a differentiated product, our stand is going to be on customer experience.”

3. “Demand is held up better than many expected, but it’s uneven across segments and geographies.”

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

Uber is quietly turning the Driver app into an on-ramp for AI work. Select U.S. drivers and couriers can now complete bite-sized “digital tasks” (voice clips, menu/photo uploads, document snippets, multilingual prompts, etc) to train models and earn between trips. It’s classic Uber and the push into monetizing the flywheel: find latent supply → instrument it with product → route demand through a marketplace. Instead of rides and deliveries, the “ride” is RL/labeling instead of a passenger. The pilot is live and positioned explicitly as new, flexible work layered on top of mobility and delivery, with Uber itself advertising data-labeling and testing services on the corporate site.

The scale math is where this gets interesting for Uber:

~33M trips per day * ~8.8M drivers and couriers = An ocean of micro-idle windows the company already optimizes with heatmaps and batched dispatch.

Even single-digit participation and minutes-per-day can translate into millions of labeled units at U.S.-grade quality, a supply pool that rivals specialist shops and Mechanical Turk while avoiding their cold-start and QA headaches. If ads became the high-margin “second engine” for delivery networks, RL/labeling can rhyme for mobility: a capital-light, counter-cyclical revenue line that monetizes the long tail of driver downtime without cannibalizing trips. Turning operational exhaust into P&L is the same discipline I’ve highlighted in marketplace S-curve stacking and incremental margin engines before.

Uber hasn’t published task rates or contribution targets, so think in illustrative ranges: if ~5% of active U.S. drivers opted in and averaged ~10 minutes/day at micro-task rates typical of on-shore annotation, you’re plausibly talking tens of millions of annualized, high-gross-margin dollars at pilot scale. This is small next to $12B quarterly revenue, but with attractive unit economics (no pick-up deadhead, native fraud controls, built-in identity) and strategic spillovers (better maps, safer AV/ADAS, faster product QA). A functioning, QA-verified workforce inside the Driver app is also a defensible data-ops moat as AV mixing ramps and idle windows shift. Execution caveats apply like task quality, fair pay, and not distracting from rides, but as a flex-work adjacency it’s on brand and potentially durable.

Remembering A Friend, Dr. Nolan Williams

I wanted to share a personal note: my friend Dr. Nolan Williams passed away earlier this month. Nolan’s research and humanity left a lasting mark on the field of psychiatry and on everyone who knew him. A fund has been created to support his wife, Dr. Kristin Raj, and their two young children. You can learn more or contribute here:

Donate on GoFundMe

Performance & Valuation Snapshot

As a reminder these snapshots below are interactive (filter/order/search). Unfortunately substack emails present them as a stale image that you need to click through to view them so I highly recommend clicking or viewing on Platform Aeronaut if you’re coming from email.

What I Read This Week

• Visa unveils “Trusted Agent Protocol”: A standard to identify/verify shopping agents for safer agentic commerce this holiday season.

• Sabre debuts AI-native “classless” pricing engine (CRO): ontinuous pricing with claimed ~3.5% revenue lift; Riyadh Air named as partner

• Waymo targets London launch in 2026: First international robotaxi market announced; supervised testing to start soon.

• DoorDash + Waymo pilot driverless deliveries in Phoenix: Starts with DashMart; customers unlock the trunk to retrieve orders.

• Amazon opens new same-day grocery depot in Minnesota: 185k-sq-ft site extends same-day perishables radius ~800 sq mi.

• U.S. online grocery hits a record $12.5B in Sept: Up 31% YoY; momentum continues into fall.

• American Airlines Unveils New AI Trip Planner: Shows early promise but needs significant improvement

• Agentic Commerce Deep Dive III from Freda Duan: Implications of agentic commerce on merchants/platforms, payment stack, and ads players

Poll of the Week

From My Recent Work

Transcript Highlights (Exec Signals)

1. “Our loyalty program is set to double the EBITDA by the end of the decade, signifying expansive growth.” - Andrew Nocella

2. “Our major cost focus at United is to drive real cost efficiency through our use of technology,” - CEO Scott Kirby.

3. “Starlink could be the biggest game-changer for United since it’s the fastest and most reliable for our MileagePlus members,” - Brett Hart.

1. “The U.S. economy remains on solid footing, and our customer base is financially strong.” - Ed Bastian

2. “The bifurcation you’re seeing in the industry is going to continue.” - Ed Bastian

3. “Revenue grew 4%, led by premium, corporate, and loyalty, reflecting the power of Delta’s brand.” - Ed Bastian

4. “Delta’s loyalty ecosystem continues to be a powerful driver of enterprise value.” - Glenn Hauenstein

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

It appears the first run at AI travel distribution is landing squarely in the “app store” model I previewed before: OTAs as enabled apps/connectors (MCP-style plugins) inside ChatGPT that you can query natively. That’s the right shape of the future, just not the feel yet

What’s working

• Supply pipes are finally reachable from the assistant. Being able to call Expedia and Booking from within ChatGPT is a meaningful step toward real-time pricing/availability without forcing users to context-switch. This is the connective tissue I expected as LLMs move from informational to agent.

• Partnering buys time. In the short term, partnering with the incumbents prevents data shut-offs and accelerates coverage while the ecosystem sorts out standards (MCP, tool schemas, settlement).

What’s not (yet)

1. Latency kills conversion. In my testing, Expedia took ~2 minutes from ask → think → connect → show results. That’s a non-starter at checkout. Voice/agent UX wants sub-second first paint and progressive refinement. The GDS and many direct connects still live in 300–3000ms land per call. This is exactly why I’ve flagged latency as a first-order constraint for agents.

2. The flow isn’t end-to-end. You still click “Book on Expedia.” Every extra hop is drop-off. The magic trick of an agentic future is discovery → selection → payment → servicing in one continuous conversation, not a referral link with an off-ramp. I’ve argued the assistant needs to “own” the transaction loop or the value leaks.

3. Brand minimization is real. Search used to hand users to a branded surface (Expedia.com) where loyalty can compound via UX. In a connector world, ChatGPT owns the information layer and the UI. If OTAs are reduced to a “book” button, they’re functionally payment/servicing rails which is an expensive role for a 10–20% take rate. I’ve been clear: the OTA’s true value has always been discovery/information/comparison; if the assistant captures discovery, the OTA’s bargaining power erodes.

“But app stores and search had guidelines too…”

True the iOS App Store and Google search both constrained brand expression. The difference here is who owns the intermediary step. In mobile or web, OTAs still controlled room selection pages, image galleries, maps, and merchandising. With connectors, the OTA doesn’t even fully own that step; it’s increasingly just fulfillment. That’s a worse strategic position over time.

What consumers actually want

Not a referral click. They want the assistant to finish the job: select the suite, apply loyalty, pay, then be on the hook for changes, re-shops, and service disruptions. That requires:

• Low-latency inventory and pricing (caching, prefetch, smarter fan-out).

• Native payments and identity (cards, passports, loyalty IDs bound to the agent).

• Integrated servicing (modify/cancel, IRROPS handling, on-stay messaging) without leaving the chat. All of this is consistent with where I’ve said voice/agents are heading—and why half-measures feel clunky today

So who wins if we keep marching down this path?

Near term (next 12 months):

• OTAs benefit as the default pipes inside assistants. They’ll see incremental lead-gen without paying Google as much at the top of funnel. But economics will compress if agents intermediate discovery.

• Google loses hotel/flight meta surface area if users start the journey in chat. “10 blue links” are structurally challenged for travel discovery in an LLM world.

Longer term (2–4 years):

• UX owners win. The platform with the fastest, most trusted, end-to-end agent wins share of consumer (and therefore supplier) mind. That could be ChatGPT, iOS/Siri, or WhatsApp, whoever pairs latency + breadth + servicing with a delightful loop.

• Direct suppliers regain leverage if they expose data right. Hotels that publish AI-readable inventory, policies, perks, and transact via open standards (think MCP as the “USB-C for agents”) can route around OTA dependency and let assistants book direct, with richer post-booking service. I’ve laid out this playbook for hoteliers explicitly.

The uncomfortable truth

This first inning feels like Custom GPTs with better branding: useful, but not the right answer yet. I don’t want a referral link; I just want it done. And when assistants actually own discovery and execution end-to-end, the rent paid to intermediaries that don’t control discovery will compress.

Subscribe now

Performance & Valuation Snapshot

I’ve made a few adjustments to this. I’ve combined the two data sets into one, and I’ve also added a column “Chg in Multiple.” For this what I’ve done is calculated the average NTM EV / EBITDA consensus multiple that the business has traded at over the trailing 18 months and calculated the difference between that and the multiple it’s trading at today.

So taking Roblox for example, the stock up 116% YTD and is trading at 50x EV / EBITDA, but that’s only 18% higher than the average multiple it’s traded at in recent history.

Separately I’m experimenting with a chart below showing YTD price change vs the delta between today’s multiple and last 18mo average multiple. It’s kind of interesting how it shows both the implicit increase in fundamentals as well as multiple expansion (or contraction).

So DoorDash is up 67% for the year but the multiple is only 24% higher than the trailing 18mo average, implying considerable expansion in underlying fundamentals driving the majority of the stock performance this year.

What I Read This Week

• ChatGPT adds “apps” via MCP: Wiring ChatGPT into travel tools using Model Context Protocol in a big step forward for agentic trip planning.

• Instacart Partners with TikTok Retail Media: New integration tightens end-to-end ad-to-cart flow on TikTok

• Tesla rolls out FSD v14.1: Adds robotaxi-style drop-offs for consumers where you can select where FSD should park: lot, street, driveway, etc

• NHTSA Investigation into Tesla FSD Traffic Violations: They’re looking into incidents where FSD blew through red lights or traveled the wrong way.

• DoorDash Ads Feature Expansion: Sponsored products, Ghost Ads, and category-level share reporting among the changes

• Waymo Lobbyists Pushing for AVs in Minnesota: Due to state laws, big changes are required before AVs can function in the state.

• Amazon Fresh Closing Stores in the US: Despite a push into fresh grocery delivery, Amazon Fresh is closing 5 of their US stores.

Poll of the Week

From My Recent Work

Transcript Highlights (Exec Signals)

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

What Will AI Travel Distribution Look Like?

I continue to think on this and try to explore the plausible routes that travel distribution looks like in an AI and agentic world. Here’s my rough thinking of the possibilities of where things go but I’m open to feedback and comments:

Branded Apps & Agents
Consumers keep starting inside big brands surfaces whether it’s Booking, Expedia, Marriott, Delta, or emerging white-glove agents like Otto. The apps embed agents that search, hold inventory, resolve changes, and upsell across the brand’s owned graph. With >50% of OTA traffic direct today there’s a tangible possibility things break this direction and that’s where most of the internal efforts at incumbents have been focused.

App-Store Model
MCP-style “plugins” compete inside of general AI frontends (think Chrome-extension like economics). User’s “install” an OTA, metasearch, or a niche concierge into ChatGPT or Claude and the horizontal AI is able to monetize through revenue sharing, reviews, and sponsored listings much like Apple does today. The supply and enablement companies participate in a robust advertising ecosystem like SEM does today to get user installations.

Open-Internet Model
Horizontal agents (ChatGPT, Claude, Gemini) keep the full funnel of discovery while vertical search and aggregators only win if they’re demonstratably faster, cheaper, and more accurate. They compete much like the early days of the internet on Google with SEO. If your travel company has the best, quickest, and most accurate data out there you will naturally rise to the top when a user queries “Find me a family-friendly resort in Cancun.”

OS/Wallet/Identity Agent
The travel discovery and agent lives at the OS layer (Apple, Google, Microsoft) and can read calendar, email, and wallet details. It auto-rebooks when flights slip and brokers supply via existing rails. Distribution here is default-placement plus ecosystem trust. Think of the billions of dollars that Google pays Apple for default search engine placement.

Electronic Arts to be Acquired by Saudi PIF and Silver Lake for $55B

It was a rumor last week and confirmed on Monday, but EA is being taken private in a deal valuing the business at $55B.

Subscribe now

Performance & Valuation Snapshot

What I Read This Week

• ChatGPT Launches Agentic Commerce Protocol (ACP): An open standard that enables a conversation between buyers → AI Agents → Businesses to complete purchases

• Booking.com Smart Flex Is Expanding: Booking is auto opting-in hotels to Smart Flex which allows Booking to widely offer “flexible cancellation for a fee.” Booking takes the inventory risk for Smart Flex cancellations, captures the cancellation fee, and has the obligation of back-filling or paying the hotel for the room.

• Uber Grocery+Retail at $12.5B Run-Rate by YE25: Up from $10B run-rate in May, Uber expects to exit 2025 at a $12.5B GBV run-rate

• Only 10% of Business Travel Buyers trust AI Recommendations: In the business travel segment, just 1 in 10 buyers express confidence in AI-based bookings or itinerary tools, trust remains a huge barrier.

• Kroger and DoorDash Expanding Relationship: Bringing grocery delivery to 2,700 Kroger stores on DoorDash

• GetYourGuide AI Product Push: The fall product release from GetYourGuide includes AI-powered review management, expansion into multi-day tours and theme parks, and AI-powered filters for consumers enhancing traveler discovery

• Expedia Research on how fragmented hotel distribution leads to rate leakage: 98% of hoteliers lose an average of 6% of revenue annually due to rate leakage (essentially wholesalers, bed banks, or redistributors sniping rates).

• TikTok reveals Smart+ Travel Ads: Enabling catalogue, event, and direct bookings within Tiktok

Poll of the Week

From My Recent Work

Transcript Highlights (Exec Signals)

Added some additional transcripts but generally just ready for earnings season to begin to test this out more thoroughly:

1. “Take Airbnb from a pre-AI app to the first or one of the first truly native AI applications.”

2. “Shift Airbnb from a marketplace for vacation rentals to a global community where you can travel and live anywhere.”

3. “Evolve from vacation rentals to becoming this entire ecosystem like Amazon, where we can sell homes, services, experiences, and many things we haven’t yet announced.”

• “FREENOW carries revenue margins in the low teens compared to North America’s mid-30%.”

• “Our DoorDash partnership is only ~10% penetrated, but growth continues to outpace the overall business.”

• Product innovation (Price Lock, Women+ Connect, Lyft Silver) expanding customer base. Lyft Silver targeting seniors (>65 demographic, now 18% of U.S. population, growing to 20%+).

• “Underpenetrated markets still represent two-thirds of the TAM across North America.”

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.

At the event the company revealed a bunch of new additions to the platform including restaurant bookings, DoorDash Dot autonomous delivery robot, DashMart Fulfillment Services, expanded grocery access, and a slew of app updates both for consumers and dashers.

Among competitors DoorDash has continued to accelerate the second derivative of absolute $ growth in GTV Y/Y. How they’ve done it is a deliberate focus on TAM expansion outside of core restaurant delivery.

I did a deep dive on the philosophical clarity that DoorDash has around TAM expansion and reinvestment back in 2023 and it’s hard to argue they haven’t been perfectly aligned with that vision throughout their entire time in the public markets.

Subscribe now

DoorDash TAM

For a business that started with a hyper focus on the core restaurant delivery market in the United States, we’ve come a long way. Back when DoorDash filed for an IPO they identified a rough TAM of $302B based on total off-premise food consumption in the United States. Fast forward 5 years and by my estimate that TAM has more than doubled to $676B once you add in all the other segments, focuses, and opportunities that they’re focused on addressing:

Core Local Delivery

The biggest component of local delivery started and remains within restaurant delivery. It’s why consumers originally download the app and why they continue to use it more than 1x per week to order food to their home or office.

And this global delivery TAM and all of these numbers can get supercharged by continued internationally expansion with Wolt and the Deliveroo acquisiton. Beyond restaurant delivery though, when you look under the hood at those various components separating out restaurant delivery you can see over time the growth in grocery and retail (albeit off small bases):

While grocery has reached an over $7B GTV run-rate, the big question is about ability to continue scaling and the use case here. While the death of exclusive inventory and the rise of grocery cross listing across every platform has been here and continues to grow, does DoorDash have the technology, picker/driver quality, and consumer intent to drive big basket sizes? I explored unit economics and market models in grocery delivery last month:

The challenge for DoorDash here is that the consumer intent is split into 1) weekly big basket orders, and 2) smaller basket top-up orders. Most of the profitability and the big CPG advertising dollars are in the big basket sizes (>$100/order) where Instacart tends to dominate, while DoorDash and Uber Eats have seen extensive growth in total GTV but average basket sizes much more in-line with the top-up orders (<$75). That’s not to say that DoorDash can’t continue to build a really big business in top-up smaller orders, that’s clearly the strategy behind Amazon’s expansion into fresh same-day grocery and it’s a big market.

Subscribe now

Fulfillment & Logistics

Underappreciated by the market is the white-label delivery portion of the business. DoorDash’s Drive/Dispatch is the “use our drivers, keep your customer” use-case and typical customers are multi-unit restaurants, regional grocers, DTC brands, and specialty retailers.

Retailers use it for coverage, control, and time-windows and plugs into the existing e-comm stacks so the merchant owns the session, loyalty, promos, and support policies while DoorDash handles courier selection, routing, ETAs, live tracking, proof-of-delivery, and exception handling. Beyond the pure last-mile delivery some partners use DoorDash for in-store picking (staff augmentation during peak times) and recently announced was micro-fulfillment via DashMart.

DashMart Fulfillment Services

Think of this as DoorDash’s venture into the Amazon model. Given the significant investment in DashMart’s, DoorDash has a large selection, geographical coverage, and optimized driver picking and delivery. They’ve already started working together with CVS Pharmacy and Party City and it enables retailer brands to extent same day delivery in an efficient and flexible manner. DoorDash benefits from incremental demand liquidity to optimize their network.

The entire goal behind fulfillment and logistics is making DoorDash the #1 in mindshare for delivery of anything from both a consumer and retailer perspective.

Forward-placing SKUs raises on-time rate and lowers cost per drop and enables merchants to fulfill same-day delivery even after stores close.

Dot Delivery Robot

There are plenty of competitors and Uber Eats has been testing and utilizing delivery robots for awhile, but Dot is DoorDash’s foray into 1P autonomous delivery robots.

It’s hard to put a TAM on this as it’s all pretty experimental, but the key competitive advantage from Dot appears to be speed; it can go up to 20mph for rapid neighborhood trips and delivery. It’s still an pilot-scale but could potentially be a game-changer, the value is cost/ETA improvement on shorter delivery routes.

Marketing & Commerce

The biggest component of this with the most traction so far is media and in-app advertising. Whether that’s sponsored listings, meal deals, or brand/CPG advertising, it’s already at a more than $1B annual revenue run-rate:

The more interesting components here that have an increased focus in the Dash Forward 2025 event is the continuing compilation of marketing, loyalty, credit cards, and branded apps and websites.

A large piece of this is as a result of the Symbiosys ad-tech/retail-media platform that DoorDash acquired in Jun 2025 for $175m. It’s essentially an off-platform amplification and attribution layer that allows DoorDash to go to a restaurant and capture not only the in-app advertising monetization, but the greater marketing budget a restaurant may have.

The CRM and guest experience component has been bolstered by the acquisition of SevenRooms for $1.2B in Jun 2025. It brought guest profiles/CRM, marketing automation, and some other components I’ll mention later that are more restaurant management.

Restaurant Management

Beyond online ordering which is sort of consistent with the white-label enablement layer and mobile app/website construction, the big existing components here both came from acquisitions.

Reservations and Waitlist were part of the SevenRooms acquisition and is a real opportunity to compete vs OpenTable, Resy, Tock, and Toast Tables. It’ll be integrated into the app as a new “Going Out” tab.

QR Order & Pay came from the Bbot acquisition back in Mar 2022 for $88m and has been a bit orphaned ever since as most of the years after have been hyper focused on delivery as opposed to in-restaurant.

But there are several clues which lead me to putting a lot of pieces together on DoorDash building a restaurant and retail POS system.

DoorDash Restaurant POS?

Remember that Uber was working on a POS system awhile back that eventually got cancelled, and it wouldn’t surprise me if DoorDash is still doing so. A full POS system would enable DoorDash to push deeper into the stack and capture some of the $12-20k per restaurant economics that other POS systems capture.

One big complaint from restauranteurs about DoorDash is that in essence they are being double charged by their POS for facilitating a delivery and then again by DoorDash. Selling a POS system would help collapse and minimize some of this double charging.

Evidence of DoorDash Building a Restaurant POS called “Carte”

• Dec 2021: Bbot pre-acquisition filed trademark for a Restaurant POS called “Carte”

• Mar 2022: Acquisition of Bbot to Bolster In-Store Digital Ordering

• Jun 2023: Report about DoorDash building a POS called “Carte POS”

• Mar 2025: DoorDash patent for NFC Payment Tap-to-Pay

• Mar 2025: Continued rumors about “Carte POS” in a 2025 Industry Blog

• May 2025: Acquisiton of SevenRooms building out Reservations, Marketing, Loyalty, and Table Management

• Jun 2025: DoorDash hiring Global Sourcing Manager for “custom-built private label tablets and smart devices”

• July 2025: Rumors on Reddit of a DoorDash POS live in San Francisco from a Restauranteur

• Today: Job opening in Merchant Platform Innovation for an “exciting new confidential initiative”

Challenges for DoorDash

The biggest challenge I see is a lot of hesitation from restaurant owners to have their POS and system-of-record being run by their biggest demand generation platform. For the same reason hoteliers wouldn’t use an OTA for their PMS, a restauranteur wants to keep separation from DoorDash on the back end of the house.

Assume you run a restaurant that is 50% in-house, 35% DoorDash delivery, and 15% Uber Eats delivery today. If you utilize Toast or SkyTab today they are independent of the economics and competition within delivery. If you adopt a DoorDash POS then suddenly you may not own the data or feel independence of customer ownership or be able to negotiate take rates between Uber Eats and DoorDash as freely. Certainly you’ll save some $ as DoorDash bundles and packages to get customers in the door, but it’s not a strategy without significant hurdles and adoption challenges.

There’s also switching costs and complexity, data ownership, and rate-parity complications. Can merchants negotiate marketplace take-rates if the POS is bundled? How can I trust my main demand funnel to not screw me over once I’m locked into a POS for the next 6-8 years?

The size of the TAM is big in restaurant and retail POS but it’s not without challenges to overcome if DoorDash is actually working towards the product.

Tickers Mentioned:

The information presented in this newsletter is the opinion of the author and does not reflect the view of any other person or entity, including Altimeter Capital Management, LP (”Altimeter”). The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies. This is for informational purposes and should not be construed as investment advice or an investment recommendation. Past performance is no guarantee of future performance. Altimeter is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Altimeter and its clients trade in public securities and have made and/or may make investments in or investment decisions relating to the companies referenced herein. The views expressed herein are those of the author and not of Altimeter or its clients, which reserve the right to make investment decisions or engage in trading activity that would be (or could be construed as) consistent and/or inconsistent with the views expressed herein.

This post and the information presented are intended for informational purposes only. The views expressed herein are the author’s alone and do not constitute an offer to sell, or a recommendation to purchase, or a solicitation of an offer to buy, any security, nor a recommendation for any investment product or service. While certain information contained herein has been obtained from sources believed to be reliable, neither the author nor any of his employers or their affiliates have independently verified this information, and its accuracy and completeness cannot be guaranteed. Accordingly, no representation or warranty, express or implied, is made as to, and no reliance should be placed on, the fairness, accuracy, timeliness or completeness of this information. The author and all employers and their affiliated persons assume no liability for this information and no obligation to update the information or analysis contained herein in the future.