Booking Holdings is up roughly 24% since I published “Why the AI Bear Case Has It Backwards” in February, as the market starts to shrug off the AI fears for now.
I wrote:
“The question isn’t ‘will travelers use AI agents to search for hotels?’ They probably will. The question is: when that AI agent finds you the perfect family-run guesthouse in Puglia, who actually processes the booking?
That guesthouse doesn’t have its own booking engine. It doesn’t accept Alipay, Pix or UPI. It can’t provide 24/7 customer service in over 40 languages. It has no app, no loyalty program, no cancellation management system.
Booking provides all of that. And it does so for 4.4 million properties across 220 countries.
The AI agent needs someone to actually process that booking. That someone is Booking Holdings.”
That figure is 4.7 million properties now.
A quick aside. When the AI fear was loudest in February, career risk forced most professionals to sell. You don’t carry that pressure. If you love travelling, Booking sits inside your circle of competence, and an hour over lunch is enough to study it. That’s the premise of The Lunch Break Investor, out next Tuesday: steadycompounding.com/book
Back to the quarter.
On the disintermediation question itself, Booking CFO Ewout Steenbergen put a number on it this quarter. Traffic from large language models, paid and unpaid combined, is “still significantly below 1% of our room nights. And that hasn’t moved so much recently.”
Yes… it only counts the LLM traffic Booking receives, so on its own it can’t rule out travellers asking ChatGPT for a hotel and then booking direct with the hotel. Booking never sees those people. What makes me doubt that’s happening at scale is that it would have to show up as missing demand, and it doesn’t. Room nights came in above guidance and direct bookings kept growing in absolute terms.
CEO Glenn Fogel’s read on the call was that travellers are using AI for discovery, then booking the way they always have, because “when they are actually ready to buy for real, that’s when they get serious.”
So I’m not changing my view that AI agents take over the booking. But there are two things in this quarter’s numbers I think are important to flag: the cost of the search transition, which is now visible in Booking’s marketing line, and what Airbnb is doing on the other side of it. Let’s go through both.
The AI transition is showing up as a cost
Marketing spend grew 10.8% this quarter while gross bookings grew 9%, which means marketing deleverage is taking place. The 10-Q tells you why:
“While we seek to adapt to evolving search engine dynamics, we expect SEO traffic to decline in the short to medium term, which may lead to increased spend in paid marketing channels.”
Free traffic from Google search is declining, and Booking is making up for it with paid traffic. Management also points to a shift in paid traffic mix and merchandising dollars moving into performance marketing, but SEO is the primary driver. That’s the AI transition showing up as a cost.
CEO Glenn Fogel on the call:
“I do believe that some of the changes that were made in the display at Google definitely put some pressure on SEO. Putting in that AI overview probably has done it.”
And he admitted he doesn’t actually know where that traffic went:
“They didn’t come to us from SEO, which would have been in that direct category. Where did they go? Did they come to us direct, or what happened? Because our direct number didn’t go down. I don’t know.”
I’m guessing most of that lost SEO traffic simply turned into paid clicks on the same Google results page, which would explain why the direct mix held while marketing costs crept up.
The direct numbers themselves are still healthy. B2C direct mix has been stable in the mid-60% range for four quarters while direct room nights kept growing in absolute terms. (The mid-60s figure excludes the B2B channel. Including B2B, direct mix was a mid-50s percentage over the trailing twelve months.)
App mix of room nights, trailing twelve months:
Two years ago: low 50s percent
A year ago: mid 50s percent
Now: high 50s percent
Steenbergen shared further, the direct ratio is stable because it is “a ratio of two positives.” Both the direct channel and the paid channel are growing in absolute terms. It’s just that paid traffic is now doing some of the work that free search used to do, and paid traffic costs money.
Honestly when I turned to Airbnb’s financial filings, I expected them to be more insulated. I assumed most of their bookings came direct: app-first, brand-led, little Google dependency. The kind of business that shouldn’t need to buy traffic.
Then I opened the 10-Q. Sales and marketing grew 27% in Q2 while revenue grew 17%. As a share of revenue it went from 22% to 24%. Brand and performance marketing alone jumped 30%.
The reason is different from Booking’s, though. Booking is spending more because SEO traffic is shrinking and paid search picks up the slack. That’s defense. Airbnb is spending more on paid growth marketing in emerging markets and on event partnerships like the World Cup. That’s offense. Their expansion markets are growing at roughly twice the pace of the core, and they’re paying to keep it that way.
How the quarter itself went
Room nights grew 5%, about a point above the high end of guidance, and this was despite the Middle East conflict pressuring long-haul travel for the entire quarter. Domestic room nights grew high single digits globally, which shows the demand didn’t disappear, it rerouted.
The less flattering part is that full-year guidance came down. Back in February, management guided for low double digit growth in gross bookings and revenue, with EBITDA growing faster than revenue. Now all three are pegged at high single digits, and EPS was trimmed from mid-teens to low-to-mid teens.
Management blames lower flight ticket growth and FX, and says constant currency remains in line with the original plan even with the conflict weighing on most of the year. I don’t think it changes the story, but you should know the full-year bar is lower now.
The first stall in alternative accommodations
Booking’s alternative accommodation room nights grew just 4% this quarter, below its own 5% overall growth, and the mix was flat at 37%. That’s the first stall after years of steady gains. Airbnb’s total nights and seats grew 10%, and accelerated from Q1.
Side by side this quarter:
- Booking room nights: +5%
- Airbnb nights and seats booked: +10%
- Booking gross bookings: +9% (about 8% constant currency)
- Airbnb GBV: +16% (15% ex-FX)
Management’s explanation on the call was that alternative accommodations were hit in part by the Middle East conflict, and that growth “was slightly lower than our overall 5% room night growth due to brands and regional mix, as we saw higher growth from Agoda and Priceline, and also in the U.S., where our alternative accommodation offering is relatively smaller.” In other words, growth came from the brands and regions where Booking’s homes offering is smallest.
Airbnb also benefited from being a World Cup official partnership and a supply campaign built around the tournament. Three weeks of it fell inside Q2. Airbnb was an official Tournament Supporter, hosted millions of guest arrivals including many first-timers, and more than 150,000 homes in host cities were listed on Airbnb for the first time.
That supply push started back in October 2025, targeting prospective hosts across the 16 host cities. So some of the outperformance came from a one-off boost, but I doubt all of it did.
One update to my own numbers. In October I compared Booking’s 8.4 million alternative accommodation listings to Airbnb’s 10 million. Booking’s count is now 9.1 million, up 8% year on year, and Airbnb hasn’t published a comparable figure in its Q2 materials, so I’ll drop that comparison until I can verify both sides.
Airbnb is coming after the hotel core
For a decade, Booking attacked Airbnb’s core. It stacked homes and apartments onto a hotel platform until alternative accommodations reached 37% of room nights, and the pitch to hosts was always the same: we bring you demand you’d never reach on your own, you pay us a fair commission, no upfront cost.
This quarter, Airbnb picked up that playbook and pointed it back at hotels.
They’ve added thousands of boutique and independent hotels across more than 20 destinations, including Singapore, with a price match guarantee and up to 15% credit toward the next booking. Hotels are still a single digit percentage of nights, but hotel nights are growing roughly three times faster than the homes business. And 35% of first-time hotel guests come back later to book a home.
Airbnb CEO Brian Chesky didn’t hide his surprise on the call:
“The hotel initiative is going significantly better than I expected, and I had high expectations… We now believe that we have the best hotel booking product online.”
And listen to his pitch to hotels. It’s Booking’s pitch to hosts with the logo swapped:
“What we are now seeing is a huge amount of influx of interest from hotels wanting to list on Airbnb… Not only do they like our product, but they like that we have a huge amount of traffic. We are one of the most trafficked travel sites in the world. We also have a young audience. We have a disproportionately American audience relative to some of our competitors, and we have an extremely favorable take rate.”
Hotels already pay commissions to OTAs, so a competitive cut plus a young, American-heavy audience could be an easy yes for them. Chesky says they are “absolutely going to be stepping on the gas.”
CFO Ellie Mertz on where they’re aiming first:
“We started in those markets that are regulatory constrained because it’s an obvious opportunity for us to bring supply to those markets where we’re not always able to fulfill demand that we already have. We’ve expanded our supply acquisition efforts to a top 20 number of cities… we’re looking to add inventory that can be accretive to the platform versus compete with homes.”
It took Booking about a decade to get alternative accommodations to 37% of room nights, and Airbnb’s hotels are still single digits, so this will take years to play out. Booking’s homes push also worked because Booking already had the travellers, especially in Europe. Whether Airbnb’s hotel push works will come down to the same thing: whether they have enough demand, especially in the US. Globally, first-time bookers grew 11%, the fastest in four years. North America, their most penetrated market, grew nights high single digits, the best in almost three years.
It’ll be interesting to watch whether they have the same success with hotels that Booking had with alternative accommodations.
The other demand levers Airbnb pulled
Two more things powered Airbnb’s quarter, and both are aimed at price.
First, Reserve Now, Pay Later. Over 20% of gross booking value was booked with no payment upfront, and Airbnb expanded eligibility further in July. It drives more bookings, longer lead times, and lets hosts lock in calendar share earlier.
Second, the single 15.5% service fee. About half of active listings are on it now, and Mertz expects the rollout to the majority of remaining hosts to be done by year end.
Mertz explained the intent:
“That single service fee allows us to provide more simplified pricing recommendations, and in aggregate, has a kind of downward pressure on pricing to allow us, one, to offer more value to guests, but also ensure that we are priced competitively relative to other platforms.”
That last phrase is aimed straight at Booking.
And Chesky says the fee change is the small lever. Asked about host pricing tools, he said pricing is “one of the biggest single levers for growth that we have… many multiples bigger than RNPL,” and that Airbnb is “building an entirely new pricing model.”
There’s a cost to all this, and it connects to my October piece on Booking’s float. RNPL pushes guest payments closer to the stay, so Airbnb collects the cash later and holds less of it. Their unearned fees were flat year on year at $2.8 billion, and the company says they would have grown if not for RNPL. Airbnb’s own 10-Q also notes that RNPL bookings “have experienced higher cancellation rates” than bookings paid upfront.
Booking offers payment flexibility too. Paying at the property has been a Booking.com staple for years, and on those bookings Booking never touches the cash at all. The guest pays the hotel, and Booking invoices for commission afterwards. That’s the old agency model, and it generates no float whatsoever.
The difference is which way each company is heading. Booking keeps shifting off agency and onto its own payments platform. Merchant bookings were 73% of gross bookings in Q2, up from 69% a year ago. You can see the result on the balance sheet: deferred merchant bookings, which the 10-Q defines as cash received from travellers before the stay happens, stood at $10.1 billion at the end of June. That’s up from the $9.1 billion I wrote about in October, and it’s the first time the balance has crossed $10 billion.
Connected Trip is the reason the shift keeps going. You can’t sell a flight, a hotel and a car in one transaction if the traveller pays each supplier separately, which is why Steenbergen described the merchant payments platform on the Q4 call as “a core enabler of the Connected Trip vision.”
So both companies let you book now and pay later. The difference is that Booking’s mix keeps moving towards collecting that cash itself, and Airbnb’s is moving away from it. Neither approach is wrong, but it means some of Airbnb’s growth is being bought with working capital.
What about AI?
Reading both calls side by side, AI is showing up in the cost lines before the revenue lines.
At Booking, AI spend is still a low single digit percentage of total technology spend, and Steenbergen says the ROI is already positive:
“We are already seeing an ROI that is positive on our AI investments today as a company… It’s still at a low single digit level of our overall technology spend.”
They’re disciplined about it too. Steenbergen described “cost aware model routing,” which means using cheap models for simple tasks and expensive models only where needed, and tracking AI cost per merge request, which is falling. On the customer side, voice AI now handles the majority of eligible inbound calls, and customer service cost per booking keeps falling at a double digit rate while satisfaction stays high.
At Airbnb, customer support cost per booking fell about 16% year on year, largely thanks to their AI agent, which works in over 50 languages and resolves nearly 45% of issues without a human. A voice version comes later this year. Their updated guidance absorbs a material increase in AI spend while margins still expand, and Mertz says headcount no longer needs to grow the way it used to.
Does the spend make sense? Chesky’s answer covers both companies, really:
“The inference cost of Airbnb are kind of de minimis relative to the ROI of our business model… Our transactions are very high dollar transactions and if AI can just increase our conversion rate just a little bit, the inference cost is so outweighed by the amount of money we make.”
When the transaction sizes are this large, a small conversion lift pays for a lot of tokens.
So is AI just a cost-savings story? At Booking, mostly, for now. The revenue side is early: personalisation ambitions, an OpenAI CPC test where Booking is in the pilot group, and a strategy of keeping travellers direct with its own AI tools.
At Airbnb, the contribution comes through speed. Time from concept to launch is down as much as 60%, and they shipped nearly 80% more features than in the same period last year. That speed is what built the conversion improvements across search, sign-up, checkout and payments, which management calls one of the biggest drivers of growth. It also built the hotel product. Groceries took eight to nine months to develop. Airport pickups took six weeks. An AI search toggle goes into testing this month.
Where this leaves my thesis
Room nights beat guidance through a regional conflict, and domestic demand filled the long-haul gap. The float and cash engine keeps compounding, and Airbnb sacrificing some of its float away makes the contrast even sharper. The AI symbiosis call is holding too, though the cost of the search transition is now visible in the marketing line, and I’ll be watching that line closely.
What changed is the competitive picture. For years, Booking was the attacker, grinding into Airbnb’s home turf. This is the first quarter where the attack clearly ran the other way. Airbnb outgrew Booking in Booking’s alternative accommodation business and opened a new fight in its oldest one.
I remain a shareholder in Booking. They’re still executing, the float keeps compounding, and nobody has disintermediated anything. But for the first time in years, the more interesting earnings report might be Airbnb’s.
One last thing.
This entire piece was built from quarterly filings and earnings calls. No terminal. No sell-side notes. No special access. Just documents anyone can pull up over lunch.
That’s what The Lunch Break Investor teaches. Six steps to analyse a business in about an hour a day. Booking and Airbnb make good practice because you already understand the product. You’ve probably used both.
The book launches next Tuesday, August 18. Pre-order now and you also get The Owner’s Manual, my three-lesson course on reading financial statements. The book arrives August 18. The course arrives today.
→ Pre-order: steadycompounding.com/book
Compound wisely,
Thomas
Disclaimer: These research reports constitute the author’s personal views only and are for educational purposes only. It is not to be construed as financial advice in any shape or form. From time to time, the author may hold positions in the below-mentioned stocks consistent with the views and opinions expressed in this article. Disclosure – I hold a position in Booking Holdings at the time of publishing this article (this is a disclosure and NOT A RECOMMENDATION).