Travel value planned through Odessia
$60M+
Annualised trip value planned over the summer of 2026, per founder Francis Davidson (company-reported, unaudited)
Team OpenAI acquired
8
Odessia's team at its June 2026 launch in San Francisco, per Skift, Nov 10, 2026
ChatGPT weekly users
1B
OpenAI's announced weekly active user milestone, Aug 6, 2026
New cash raised by OpenAI
$11B
First close of a round valuing the company at $1.25 trillion, Nov 6, 2026
The deal
OpenAI Bought a Booking Machine and Switched It Off
On November 10, 2026, OpenAI bought Odessia, a San Francisco trip-planning agent that had been open to the public for less than a month, and switched its service off the same day. The purchase price was not disclosed. What OpenAI actually bought was a team of eight, an IATA accreditation that lets it sell travel on its own ticket, and a founder who had previously taken a hotel company public and then watched it collapse in 2025 — precisely the operational layer a model release cannot conjure.
| Item | Detail |
|---|---|
| Announced | November 10, 2026, by founder Francis Davidson |
| Purchase price | Not disclosed |
| Service status | Switched off during the transition; existing bookings and travel plans handled by Odessia support |
| Product | Plans and books flights, hotels and itineraries in one conversation, as a free text concierge |
| Launch history | Founded June 2026 with eight people in San Francisco; opened to the public Oct 14, 2026 |
| Funding before sale | About $6 million, led by Sequoia Capital |
| Credential | IATA-accredited travel agent, with access to private hotel rates and perks such as upgrades |
| Founder's next act | Davidson joins OpenAI to focus on the future of travel; no product or pricing announced |
The shutdown on announcement day is the tell: this is not a product roll-up but a capability purchase. Davidson claims Odessia planned more than $60 million of travel over a single summer from tens of thousands of early adopters, and that tasks a human travel designer would spend 40 hours on were compressed into two. Neither figure has been audited, and the service that produced them is offline. The asset is the plumbing and the people, and OpenAI has said only that it will \"have more to share on the shape of the future soon.\"
Closing share price change on Nov 10, 2026
Online travel agencies fell on the news; hotel chains rose. Expedia was down nearly 4% intraday and Tripadvisor more than 3% before both recovered into the close, per PYMNTS, Nov 10, 2026.
Unit: % change
Marriott International
Ask Bonvoy shipped in June 2026
2.7%
Booking Holdings
Closed at $159.26
-2.6%
Expedia Group
Closed at $253.77
-2.1%
Airbnb
Closed at $186.90
-1.8%
Sabre
Closed at $1.77
-1.7%
Tripadvisor
Closed at $7.09
-0.4%
The precedent
OpenAI Already Lost the Front-Door Fight in March
Eight months before the acquisition, OpenAI attempted the same move without buying anything: native travel checkout inside ChatGPT. By early March 2026 it had pulled the plug on native travel checkout entirely, and online travel agency shares rose on the news. The retreat is the reason the acquisition happened, and it defines what OpenAI now thinks it needs.
| Model | Who owns the front door | Who owns the cash register | Status |
|---|---|---|---|
| Instant Checkout, late 2025 to March 2026 | OpenAI | OpenAI | Withdrawn in early March 2026 |
| ChatGPT Apps, from October 2025 | OpenAI | Booking Holdings and Expedia Group | Live; OTA apps run inside the conversation |
| Odessia, from November 2026 | OpenAI | OpenAI | Service switched off; nothing announced |
The parallel structure with Booking Holdings' own case study is instructive. Booking integrated its systems with OpenAI models, shipped a conversational trip planner in ten weeks, and watched queries shift from \"Myrtle Beach\" to \"a quiet beach in September with my dog.\" That is a supplier renting its data and rates to the assistant. Owning the assistant's booking engine is a different trade.
The economics
The Target Is $21 Billion of Demand That the OTAs Currently Rent
Booking Holdings' own filing states where its demand comes from: performance marketing is spent \"primarily\" on online search engines, affiliate marketing, metasearch and social media. That is a rented front door, and the rent is the single biggest cost in the industry. Across the twelve months to September 2026, Booking Holdings and Expedia Group together reported about $21.3 billion of selling and marketing expense.
Cost of demand: selling and marketing as a share of revenue
Trailing twelve months through Sep 30, 2026, as each company reports it. Booking's line includes sales and other expenses, Expedia's includes partner commissions and brand marketing, and Airbnb's includes sales.
Unit: % of TTM revenue
Expedia Group
$8.70B on $16.23B of revenue
53.6%
Booking Holdings
$12.63B on $28.13B of revenue
44.9%
Airbnb
$3.08B on $13.96B of revenue
22.1%
Combined selling and marketing
$21.3B
Booking Holdings and Expedia Group, trailing twelve months through Sep 30, 2026
Marketing per revenue dollar at Booking
29.1¢
Q3 2026 marketing expense of $2.59B on $8.89B of revenue, up from 26.0¢ a year earlier
Booking revenue per $1 of bookings
16.3¢
Q3 2026, down from 18.1¢ in Q3 2025
The squeeze is already visible inside the numbers. Booking's gross bookings rose 9.6% to $54.46 billion in Q3 2026, yet revenue fell 1.3% to $8.89 billion while marketing expense climbed 10.6% to $2.59 billion. Expedia, by contrast, grew gross bookings 11% to $34.01 billion and revenue 12% to $4.94 billion, with adjusted EBITDA up 17% to $1.69 billion — and cut B2C marketing while absorbing higher B2B partner commissions.
| Q3 2026 metric | Booking Holdings | Expedia Group |
|---|---|---|
| Gross bookings | $54.46B, up 9.6% | $34.01B, up 11% |
| Revenue | $8.89B, down 1.3% | $4.94B, up 12% |
| Revenue per $1 of gross bookings | 16.3¢ (18.1¢ a year earlier) | 14.5¢ (14.4¢ a year earlier) |
| Marketing / selling and marketing | $2.59B, up 10.6% (29.1% of revenue) | $2.39B (48.5% of revenue); B2C spend down, B2B commissions up |
| B2B revenue | Not separately disclosed | $1.71B, up 23% |
| Room nights booked | Not disclosed | 115.7M, up 7% |
| Management view of agentic traffic | Not quantified in the quarterly filing | High intent but a small share of traffic; not denting direct bookings |
Supply chain
Who Keeps the Transaction When the Assistant Owns the Conversation
Buying the front end redistributes the travel chain, and the beneficiaries are not the OTAs. Upstream, an agent needs the model layer, an orchestration framework — Odessia was built with LangChain — and the rails that hold bookable inventory. Downstream, the hotel brands are already selling inside the assistant: Marriott International debuted Ask Bonvoy on June 17, 2026, IHG Hotels & Resorts shipped a ChatGPT app on June 4, 2026, and Wyndham Hotels & Resorts launched a native ChatGPT app in May 2026.
- LangChain — the agent framework Odessia was built on, per Boutique Hotel News, June 3, 2026; private, and plumbing rather than an asset
- Amadeus — the first infrastructure partner Google wired into its agentic hotel-booking protocol, per HospitalityNet, Aug 24, 2026; the inventory layer can win the transaction if agents book source-side
- Sabre — supplies airline and hotel inventory, pricing and availability to the distribution chain; $2.88B of trailing revenue against a $714M market cap
- OpenAI's model layer — the same partner stack Booking Holdings built its trip planner on, now owned by the party that sells the answers to everyone else
- Marriott International — Ask Bonvoy puts inventory and loyalty inside the assistant, and the stock rose 2.7% on the day of the Odessia deal
- Tripadvisor — the purest casualty: $436M of Q3 2026 revenue, a $337M operating loss, and a $831M market cap
- Airbnb — rolling AI into every step of a trip in the US and promising its own AI travel agent for the first half of 2027
- Expedia Group — B2B revenue up 23% to $1.71B is the hedge inside the house: supplying other brands is what an agent-disintermediated retail world looks like
So the correct reading is not \"AI is coming for travel agents.\" It is that the scarce asset in travel is becoming the conversation itself, and the party that owns the conversation no longer has to rent it. That does not require OpenAI to become a merchant of record: it is enough for the assistant to decide the trip, while the booking lands wherever the agent's plumbing points — a hotel chain, a distribution platform, or an OTA that has been paid to be the fulfilment layer.
Valuation
The Market Is Repricing Terminal Value, Not This Quarter's Bookings
| Company | Close | Market cap | TTM revenue | EV/sales | P/E | TTM net margin |
|---|---|---|---|---|---|---|
| Booking Holdings | $159.26 | $119.7B | $28.13B | 4.4x | 17.1x | 25.5% |
| Expedia Group | $253.77 | $29.1B | $16.23B | 1.8x | 14.5x | 13.0% |
| Tripadvisor | $7.09 | $0.83B | $1.67B | 0.6x | n/a (loss) | -21.6% |
| Airbnb | $186.90 | $111.9B | $13.96B | 7.7x | 35.7x | 23.0% |
| Marriott International | $428.26 | $110.3B | $7.58B | 4.7x | 41.6x | 9.7% |
| Sabre | $1.77 | $0.71B | $2.88B | 1.6x | n/a (loss) | -5.4% |
Booking Holdings' market capitalisation has fallen from $174.8 billion at the end of September 2025 to $119.7 billion, even as gross bookings grew 9.6%. That is not an earnings de-rating; the company still earns a 25.5% net margin. It is a terminal-value de-rating, and the same logic leaves Tripadvisor at 0.6 times sales with no earnings and a market cap of $831 million, roughly 48% below where it traded a year ago.
Horizons
Two Clocks: What to Watch, and What Would Break the Thesis
- OpenAI's travel roadmap — Davidson said more would be shared \"soon.\" The first disclosure of a product, pricing model or launch date is the catalyst, and none exists yet
- Booking Holdings' Q4 2026 and full-year marketing line — the single most informative number: double-digit marketing growth against shrinking revenue means the front door is being rented more expensively
- Expedia Group's next earnings commentary — management already calls agentic traffic minor and is testing OpenAI's advertising products even if returns are lower; a change in that language moves the multiple
- ChatGPT ad density — sponsored placements appeared in roughly 0.8% of 500 sampled prompts in February 2026, per Adthena analysis reported by Adweek; watch that ratio to see whether OpenAI monetises the front door with ads or with transactions
- Airbnb's AI travel agent, promised for the first half of 2027 — the fastest credible counter-move to an OpenAI-owned planner
- Whether OpenAI ships a first-party bookable flow — accreditation and a working travel-operations stack are the assets; the risk to the OTAs is a checkout that keeps the whole commission
- Whether the OTAs get paid for the conversation — a revenue share with the model platform replaces per-click economics, which is good for volume and bad for take rate
- Whether hotels and rails capture the disintermediated booking — every agent that books source-side through hotel systems and distribution networks is a commission an OTA never earns
- Whether Tripadvisor becomes a takeover candidate — at $831 million against a portfolio of review data and experiences inventory, it is cheap for either a model platform or an OTA
The thesis in one line: the OTAs rent the front door of travel from search engines and metasearch networks, and OpenAI now owns one. The near-term earnings risk is small, because agentic traffic is small, Odessia is switched off and no product exists. The long-term risk is structural, because a front door with a billion weekly users and no need for advertising revenue undercuts the auction prices that fund the incumbents' growth. Hold two things in view at once: the immediate numbers are not at risk, and the relationship with a counterparty that just demonstrated it will buy the front door rather than rent it is.
Investable read-through: who owns the front door after Nov 10, 2026
- Marketing expense rose 10.6% to $2.59B in Q3 2026 while revenue fell 1.3% — the demand bill an OpenAI-owned planner undercuts is already growing faster than sales
- Revenue per dollar of gross bookings fell from 18.1% to 16.3% year over year, so take-rate pressure arrives before any AI front end ships
- Near term, watch the Q4 2026 marketing line; long term, a model platform owning the front door caps the terminal take rate Booking is already valued on
- Booking Holdings also runs a ChatGPT app inside the assistant, so OpenAI is simultaneously its distribution partner and its most credible competitor
- CEO Ariane Gorin calls agentic traffic high intent but a small share of traffic and not denting direct bookings, so the near-term earnings hit from agentic search is limited
- But Expedia is testing OpenAI's advertising products while accepting potentially lower returns, funding the very channel that may eventually take its traffic
- B2B revenue of $1.71B, up 23% in Q3 2026, is the hedge: supplying other brands is what a retail world with an agent-owned front door looks like
- Horizon: the risk is terminal value in 2027 to 2029, and the offset is the B2B arm plus the ChatGPT app that keeps it in the conversation
- Q3 2026 revenue of $436M came with a $337M operating loss, and trailing operating income is negative $355M — the referral-and-reviews model has no earnings left to defend it
- At 0.6 times sales and an $831M market cap, the equity already discounts a business whose economics are eroding rather than one whose growth is at risk
- Answer engines take the review-and-recommendation step Tripadvisor monetises; owning experiences inventory such as Viator is what would make it a target rather than a casualty
- Airbnb spends 22% of revenue on selling and marketing against Booking's 45%, so it does not have to win the paid-traffic auction to keep growing
- The company is rolling AI into every step of a trip in the US and has promised its own AI travel agent for the first half of 2027, the fastest available counter-move
- With $13.96B of trailing revenue and a 23% net margin, Airbnb can absorb a shift in discovery, and its own agent could become the front door instead
- Horizon: 2027 is the decision year, and the Q3 2026 revenue growth of 18.9% buys time to get there
- Ask Bonvoy, launched June 17, 2026, places Marriott's inventory and loyalty inside the assistant — the defensive answer to an agent-owned front end is to become one
- The stock rose 2.7% on the day of the Odessia deal, the clearest market read that hotel brands gain when the front door changes hands
- At $428.26 the shares already pay 41.6 times trailing earnings, so the AI transition is expected to be smooth rather than to re-rate the multiple
- Watch for commission mechanics: if assistant-originated bookings arrive on a revenue-share basis, volume rises while the hotel's margin per booking falls
- Sabre supplies the inventory, pricing and availability any booking agent must shop, so source-side booking is the one route that keeps a rail inside the transaction
- Trailing revenue of $2.88B against a $714M market cap and net debt at 25.6 times EBITDA leaves the equity a leveraged option on where bookings route rather than a way to own the front door
- Catalyst to watch: hotel-distribution wins with model platforms, following Amadeus being wired into Google's agentic hotel-booking protocol in 2026
- Horizon: 2027 to 2029, and only if agentic volume becomes material rather than experimental
