Airbnb's AI is cheap per booking — the everything-app is where the premium lives

On September 8, at Goldman Sachs' Communacopia conference, Brian Chesky told investors that AI is a "cognitive revolution," a thousand-year journey, and that the companies that win it will be the ones that "surf the wave" rather than get swallowed by it. A few days later, Airbnb stock sat about 13% below its all-time high of $193. Here is the tension every investor in this name has to hold: the CEO is right that something real is happening, and the market is still trimming positions.

The reason both things can be true is that Chesky's AI pitch splits cleanly into a piece that is verifiable in the financials and a piece that is marketing until it materializes. Separate them and the stock's argument, and its price, come into focus.

The part that shows up in per-booking economics

The provable part is Airbnb's second quarter. Revenue grew 17% to $3.6 billion, gross booking value hit $27.2 billion, and adjusted EBITDA rose 21% to $1.3 billion with the margin at 35%. Underneath that, the AI claims have direct operating fingerprints: customer-support cost per booking fell about 16% year over year, and the AI assistant now resolves roughly 45% of support issues without a human agent. Feature output is up nearly 80% year over year while headcount sits roughly flat.

What makes this worth parsing rather than shrugging past: this is AI that pays for itself, which is rare. Most of the AI economy right now is a cost center — training and inference eat capex that may or may not come back. Airbnb's version is different because of the unit economics, and Chesky has actually described them correctly. The per-booking take rate is roughly 13% of gross booking value, so every transaction drops real revenue into the business, while the inference cost of resolving a support ticket is pennies against that. Cheap inference, big per-transaction revenue. That is why he can call AI "the best thing to have happened to Airbnb" and spend "a lot more" on tokens without blowing up a 35% margin — the efficiency gains are bigger than the bill.

There is a confession embedded here worth noting. A year ago the internal question was whether AI was "good or bad" for Airbnb. Chesky has said he "so underestimated" the effect. The productivity numbers are the independent evidence that the answer turned out to be good, and they are the load-bearing reason to believe the AI wave is more than keynote decoration.

The part that is still a promise

The expensive, unproven half is the "everything" platform. Chesky is re-positioning Airbnb from a one-hit home-rental wonder into travel's version of everything — hotels, experiences, services, car rentals, resort passes, flights eventually, with the core gross booking value approaching $100 billion and, in his telling, able to double.

Some of this is real but small. Hotel nights are growing about three times faster than home nights — but from a single-digit share of total bookings. A third of first-time hotel guests later return to book a home, which is a genuine cross-sell, but it is a story in early innings, not a mature business line. And it deserves the incumbent-incompetence caveat: Airbnb is scooping up independent and boutique hotels, roughly half the world's inventory, precisely because the incumbent online agencies have under-served that fragmented supply for years. Before crediting Airbnb's strength in convincing hotels, recognize that it is partly walking into an opening the OTAs left unattended.

What the price already assumes

Now the gap between the wave and the stock price. This is the uncomfortable part that a kernel of customer-service savings does not resolve.

After a roughly 40% surge in three months, Airbnb trades at about 38 times trailing earnings and 34 times EV/EBITDA — roughly double where Booking Holdings and Expedia trade on both measures. Booking, the profitable OTA incumbent with a lower-growth but far cheaper multiple, is the value counter-argument. The AI acceleration, the hotels expansion, the everything-app optionality, all of it is already in the multiple. That is why the stock eased off its high even as the CEO delivered the most bullish presentation of the season: the marginal buyer is no longer paying for news, they are asking whether a 38x multiple needs durable cash-flow proof that isn't priced in yet.

The attribution that matters

So which piece are you paying for when you buy today — the confirmed, margin-visible AI economics, or the everything-app vision that is still a promise? The first is cheap and real; the second is where the premium lives and where the execution risk sits. Chesky says to surf the wave. Fine — but surfing means you are already on the wave, and the price reflects it. The test that would resolve the premium is not another keynote, "cognitive revolution," or feature count. It is whether hotel nights, first-time bookers, and the take rate keep compounding at the accelerated pace that got this multiple repriced in the first place — and whether the "everything" segments actually reach the economics Chesky is selling, one reported quarter at a time. Treat the wave talk as the marketing asset it is, and the per-booking numbers as the only thing that proves it.