Paying for AI tokens no longer says much about a company. Nearly every large engineering organization now runs its developers on AI coding tools and treats the bill roughly the way it treats cloud hosting. What is rare is a chief executive telling investors that the bill explains the results. Airbnb's did on Friday, and the stock rose 17.4% to close at $178.07. Brian Chesky told CNBC in an exclusive interview after Thursday's second-quarter results that the company will spend "a lot more" on tokens this year than it forecast, because the cost of inference "pales in comparison" to the return.
Tokens are the billing unit for AI models, each one roughly three quarters of a word. Inference is what it costs to run a finished model every time it answers something, as distinct from the far larger cost of building the model in the first place. Chesky's argument is that Airbnb's economics make that cost easy to absorb: a booking is a high-dollar transaction, so a small lift in the share of visitors who complete one pays for a lot of tokens. Airbnb's second-quarter shareholder letter put revenue at $3.6 billion, up 17%, net income at $816 million, and raised full-year revenue growth guidance to at least the mid-teens.
What moves this past a routine budget update is the causal claim. A company that overshoots its AI spending usually frames it as an investment whose payoff is still ahead. Chesky framed it as the payoff. Airbnb, he said, is becoming an "AI-native company," which he called "probably the number one explanation for our results." That is the difference between a cost line management hopes to justify later and one it is using to explain revenue today.
What Airbnb says its AI use produced
The gains are mostly internal. Airbnb says it has cut the time from concept to launch by as much as 60% on key projects and shipped nearly 80% more features and improvements in the first half of 2026 than a year earlier, with headcount roughly flat. Customer support cost per booking fell about 16% year over year, and about 45% of guests who reach the company's AI agent never need a human. Airbnb uses more than a dozen models internally, including Anthropic's Claude Code and OpenAI's Codex, and restricts the slower, pricier ones when a task does not require them. On the earnings call, Chesky said the company is not "tokenmaxxing," his word for treating token consumption as a scoreboard.
Much of the shift dates to January, when Airbnb named Ahmad Al-Dahle as chief technology officer. He ran generative AI at Meta and the team behind its Llama models. Chesky has said Airbnb was middle of the pack on AI before he arrived.
Booking Holdings and Expedia decline the same attribution
Both rivals reported within the same week, and both use AI heavily without crediting it for the quarter. On its earnings call, Booking Holdings said AI chatbot referrals remain under 1% of room nights even as internal use pushed customer-service cost per booking down at a double-digit rate, and finance chief Ewout Steenbergen said the return on its AI investment is already positive. Expedia told investors that its natural-language trip planning does not yet convert bookings but captures far more about traveler intent. All three describe cheaper operations and faster engineering. Only Airbnb makes that the headline explanation for growth.
The filings do not show the AI bill
Nothing requires Airbnb to publish it. Its quarterly report filed with the SEC splits spending into five buckets by function, not by vendor or input. Token costs land inside cost of revenue, which covers payment processing, third-party data centers and system software, or inside product development, depending on the use.
What the filing does show cuts slightly against the story. Product development rose $62 million, or 10%, to $672 million, and Airbnb attributed the entire increase to payroll from a higher average headcount, which also sits awkwardly beside the claim that headcount stayed flat. Cost of revenue rose 16%, roughly in line with revenue. Companies do name these costs when they matter: Duolingo's filing for the same quarter credits part of its research increase to $3.3 million in software and third-party AI costs.
So the spending Chesky says will overshoot his budget is not yet large enough to surface anywhere in the accounts, while the 60% and 80% figures behind the payoff are internal measurements no outsider has audited. Revenue grew 17% as reported and 13% stripping out currency. The World Cup lifted the quarter. On the call, Chesky himself called the results a culmination of several years of changes. The claim that AI explains them may well be right. For now, the only place it appears is in what he says.