Booking Holdings

Booking Holdings runs Booking.com, the largest of its brands by a wide margin, together with Priceline, Agoda, KAYAK and OpenTable. A ten-year-old could explain the business. A small hotel has a room it cannot fill. A traveller on the other side of the world wants a bed. Neither knows the other exists. Booking introduces them, and when the guest pays, Booking keeps a share, about 14 cents of each dollar across the whole group. The hotel pays nothing to be listed and nothing if the room stays empty. That one fact explains most of why hotels resent the commission and keep paying it.

In 2025 customers booked $186bn of travel through these brands: 1.24 billion room nights, 88 million rental-car days and 68 million airline tickets. Booking kept $26.9bn of that as revenue, needed only $322m of capital spending to do it, and produced $9.1bn of free cash flow. It owns no hotels. Booking.com alone lists 4.7 million properties, of which 4.2 million are apartments, homes and other alternatives, and over 500,000 are hotels.

This year the market decided all of that might be about to end. The argument is simple: if an AI agent can search every hotel, compare the prices and book the room, why would anyone need a middleman? The shares fell from above $210 in August to $146.28 at the low, then jumped about 10% on 4 November, the day after third-quarter results, along with the rest of the sector. This report asks three questions. What has actually happened to the business? What is the fear worth? And what are you paying?

Look at the middle cell before anything else. The great fortune in this stock was made long ago. Over the past decade an owner of one of the finest businesses on the internet has done worse than an index fund, because the price in 2016 already assumed a great deal, and because Covid took two years out of the middle. Wonderful businesses are not always wonderful stocks. The price paid decides that.

Two lessons for an owner. The first is that the great value came from a small thing bought cheaply, not from the big things bought dear. Priceline's original model was merchant and opaque: it bought rooms cheaply and hid the hotel's name. Bookings B.V. was the opposite. It let a hotel show its own rates, let the guest pay at the desk, and asked only for a commission afterwards. That friendlier model conquered Europe's independent hotels, and it is still the source of the moat. Eight years later, KAYAK and OpenTable were bought for $4.6bn at full prices, and they have cost owners about $2.5bn in write-downs.

The second is that the people who bought in the wreck made the money. On today's share basis the 1999 peak was about $39 and the 2002 low about 26 cents. From the peak, the stock has compounded at a modest 5.6% a year in price, which is respectable for the worst possible entry. From the low it has compounded at about 31%. Same business, same management, and a very different result, set almost entirely by the mood on the day of purchase.

The toll. Revenue as a share of bookings, the take rate, was 15.8% in 2016, 15.6% in 2019 and 14.5% in 2025. That slow slide looks like a weakening franchise, but we think it is in good part mix: flights and other lower-margin products have grown fast. Note the seasonal trap as well. Bookings are counted when made and revenue when the guest checks in, so the third quarter carries the summer and looks far richer than the first. Never annualise a Booking quarter.

The advertising bill. This is the number to keep in your head. In 2025 Booking spent $8.2bn on marketing, 30 cents of each revenue dollar and 4.4% of bookings, and it says plainly that paid channels are mostly search engines. In other words, Booking already pays a toll to a larger toll-keeper. About 45% of its room nights in total do not arrive directly: in the consumer business that share is in the mid-30s, and the rest of the gap is the business-to-business channel, where partners resell Booking's inventory. The strategy of the past decade, the app, Genius, the Connected Trip, has one purpose: to bring the customer back without paying Google a second time. It has worked up to a point. App bookings are in the high-50s percent of room nights, up from the mid-50s a year ago, and most of those are direct.

From matchmaker to paymaster. In 2016 Booking handled the money on 14% of bookings; now it handles 75%. This costs real money in card fees, fraud and chargebacks, up $221m in the first nine months alone, but the 10-Q says incremental payment revenue now exceeds those costs. More important, the party holding the card details is the merchant of record. Keep that phrase in mind for Part VII: it has turned out to be the thing the AI platforms did not want to take on.

The cash nobody mentions. Merchant bookings mean customers pay before they travel. At 30 September Booking held $6.45bn of their money as deferred merchant bookings, up $1.2bn in nine months, and owed partners another chunk of its $5.2bn of payables. So part of the $15.8bn of cash on the balance sheet is float, and part of the free cash flow is float growing. It costs nothing while the business grows, as Berkshire knows better than anyone, but in 2020 it ran the other way. We strip it out in Part XI.

Buried in note 13 of the 10-Q is the largest item in the quarter. In July, staff at the Federal Trade Commission told Priceline they intended to recommend a complaint against it and a business-to-business affiliate partner, over disclosures, fees, customer support and billing. Booking has reduced revenue by $900m and accrued a further $100m of loss. That is 19% of 2025 net income, in one line, for one of its smaller brands. No settlement has been agreed, and the filing warns that the FTC "may seek monetary penalties and restitution in excess of these amounts".

Three things about this bother us. First, the tone changed: the second-quarter 10-Q said Priceline did not agree with the FTC's position, and that sentence has gone. The third-quarter filing says a negotiated resolution "would be likely to include" payments, restitution and changed business practices. Second, on our reading of the available transcripts and the prepared remarks, management did not mention it on the call, although it took ten points off the growth rate. Third, a charge booked mainly as a reduction in revenue looks like money going back to customers, and it raises a fair question about how some of those customers were acquired. Reuters reported that the FTC was examining ads that seem to lead to a hotel's own website but route to a third-party site at higher prices. San Francisco's City Attorney has separately sued Booking Holdings over the same kind of scheme, alleging that Priceline's partner arm supplied the inventory.

Now the honest other side. Gross bookings are not affected, the charge is excluded from adjusted results with disclosure, and $1bn is about $1.34 a share before tax. It will not change the value of this company by more than 1%. But our own rule applies: an item that large belongs beside the headline, not in a footnote. Adjusted EPS +17% and GAAP EPS +7% are both true, and a reader deserves to see them together.

For the fourth quarter Booking expects room nights up 1% to 3%, gross bookings up 2% to 4%, adjusted revenue up 0% to 2% and adjusted EBITDA about flat. Demand "moderated somewhat" late in the quarter, and October suffered from disruption around the Middle East, dear airfares and what the release calls localized unrest in certain European cities. Part of the flat EBITDA is chosen: management is spending more on flights, attractions, dining and AI. The comparison is also the hardest of the year, and the guide assumes the conflict's effects last to December. Even so, it would be the slowest quarter by a wide margin in the nine-quarter table in the release, and the fear of AI had nothing to do with it. Travel is a cyclical business. Some of what looks like structural worry may simply be a soft patch.

Why, then, did the stock jump about 11% the next day? We could not establish a reason, and we would rather say so than invent one. It was flat before the open, and Goldman Sachs cut its target that morning. Expedia rose 12% and Airbnb 7% the same day, so it was a sector move. Our guess, and it is only a guess, is that investors braced for signs of AI damage and found none. Do not read much into a single day.

In his 1991 letter Buffett wrote that an economic franchise arises from a product or service that is needed or desired, is thought by its customers to have no close substitute, and is not subject to price regulation. Booking makes a very good test of that sentence, because on each of the three it has something new to tell us.

Inertia cuts both ways, and the whole investment case sits on it. A Buffett reader hears "customers stay out of habit" and thinks of a toll bridge. Regulators hear it and think of dominance. Short sellers hear it and think of the first thing an agent will remove, since an agent has no habits. All three are describing the same thing.

Our judgement is that the moat is wide, and changing shape. The consumer side, the part built on search arbitrage, habit and contract terms, is eroding through technology and law. The supply side, the part built on millions of small relationships, is widening. If that is right, Booking moves slowly from a brand consumers seek out towards a wholesale utility that everyone's agent must plug into. That would still be a good business. It would probably be a less profitable one, and the fall in the stock is an argument about how much less. We give it 8, not 9, and we would have given it 9 two years ago.

Start with why the stock fell, because the popular story is too simple. There were four legs, and AI is only the name of two of them.

On the reported evidence, disintermediation has not begun. CFO Ewout Steenbergen says LLM referrals have "accelerated" in recent weeks and "still account for significantly less than 1% of Booking.com's room nights". The direct share of consumer room nights has been stable in the mid-60s for four quarters. Consumer marketing grew more slowly than consumer bookings. And there is a structural point that matters more than any of those. Every agent launched so far has ended at a merchant of record. OpenAI retreated from taking the payment. In Google's AI Mode, which began booking US hotels in August, the hotel or agency stays the merchant, and Booking.com and Priceline are launch partners. Meta's Muse books flights through a direct connection, but for hotels it shops consumer websites the way a person would. Steenbergen's explanation is simple: "We get the vast majority of the traffic because they like to connect to aggregators", since no agent wants to connect to millions of small properties one at a time.

Two cautions. The under-1% figure is management's, unaudited, and it counts referrals, not influence: someone who plans a trip in ChatGPT and then opens the Booking app shows up as direct. And nobody publishes agent booking volumes, so the first hard number may arrive late. The first visible damage is further up the funnel: organic search. Booking expects it to decline "in the short to medium term". CEO Glenn Fogel has said AI has put pressure there.

Notice who is hedging. Booking has put two of its own experiments in the field. Lola, run by KAYAK's founders Steve Hafner and Paul English, replaces commission with membership tiers at $5 and $25 a month and passes commission back to the user. Vayo, with up to $100m planned, reads travel confirmations from any brand. A company willing to build the thing that could cannibalise its own take rate is taking the threat seriously. That is better than the alternative, and it is also an admission.

The bear's best line came from Fogel himself, who reportedly told a Citi conference in September: "There's almost no proprietary data of any kind." He meant it as a defence: the asset is relationships and trust with payments, not a database. We think he is mostly right. But this is the honest position: no one can know the answer, including the people who sold at $150 and the people who bought. An investor's job is not to predict it. It is to work out what the price already assumes, and we do that in Part XI.

The pattern across the rows. Booking's competitors are not really other travel agencies any more. It won that contest years ago, in Europe at least. Its competitors now are the owners of attention above it, and the owners of rooms below it. Both would like to push the middleman's slice down, and both have new tools. That is a different kind of pressure from a price war, and it moves more slowly.

Across the board. We have now written about the toll-keepers on the other side of this trade. In our October Alphabet X-Ray we found Search slowing to +12% as AI Mode earns less per query. It is odd to fear Google's AI on Booking's behalf while Google's own Search is the thing under strain: a Google that needs revenue is more likely to charge Booking more, and less likely to give hotels free direct links. And the same AI that threatens Booking's front door is cutting its back-office costs right now. Few of the companies on our board can say that as plainly.

Integrity, capital allocation, owner mentality. We see no reason to doubt the integrity of this management, but the FTC matter is the first thing in this report that gives us pause. Priceline is a small part of the group, and what is alleged concerns a partner, yet a billion dollars is a lot of restitution for a side-channel, and leaving it out of the prepared remarks is not how we would write to our partners. Buffett's habit is to give bad news early and plainly. Capital allocation since 2017 has been very good: no big deals completed, costs now under control, and a share count falling about 5% a year, and 7.5% lately. Owner mentality is mixed: pay is high (351 times the median employee), officers and directors together own under 1%, and we found nobody buying below $160 with their own money, while the company was buying with yours. That is why this dial reads 7.

At about 1% this is not an income stock, and nobody should own it as one. The owner's real return is the buyback: $12bn in twelve months is about 9% of the market value. Together with the dividend, Booking returned about 10% of its market value in a year. As Part IX says, that pace is above what the business earns, and a sensible pace is nearer 6% to 7%.

Why we subtract the "one-offs". Spain: about $485m accrued in 2024. Italy: $332m paid in 2024. Indirect taxes: $89m in 2025. The FTC: $1bn in 2026. That is about $1.9bn in three years, each excluded or described as unusual, and more are visibly on the way. A company with a one-off every year does not have one-offs. It has a cost of doing business in a world where it has become large enough to be noticed. We charge about $0.4bn a year after tax against owner earnings, and we also subtract the unpaid FTC and Spanish liabilities from value.

What is priced in? With owner earnings of $7.5bn, a 10% discount rate and net debt plus those liabilities of $5.7bn, today's price requires owner earnings to grow only about 4% a year to 2031 if the business then sells for 18 times, or about 7% a year if it sells for 15 times. Management's ambition is 8% or more for bookings and revenue, and 15% for adjusted EPS once buybacks are counted. Over the past decade room nights grew more than 9% a year, through a pandemic. So the market is not pricing a catastrophe. It is pricing a real slowdown and a lower multiple at the end.

The central value moves to about $195 at an 11% discount rate and $213 at 9%; to $176 at an exit multiple of 15× and $231 at 21×; and by about $18 for each two points of growth. The number that matters most is the multiple at the end, which is another way of saying the answer depends on what people believe about agents in 2031. We are not embarrassed by that, but it is the reason to demand a margin.

Multiples on the consensus. At the current price the shares trade on about 16× 2026 consensus EPS ($10.46), 14× 2027 ($12.35), 12× 2028 ($14.45), 10× 2029 ($17.30) and 9× 2030 ($20.09). Those forecasts assume about 18% growth a year, helped by buybacks, and we do not fully trust them. Even so, the contrast with the 19× 2027 earnings at which we priced Alphabet last week is worth noticing. Analysts' mean target is $219 (range $150–$270), with 47 buys, 25 holds and no sells.

Our own screen. The Buffett Take on the platform shows $168.62 by discounted cash flow, $200.74 on owner earnings and $79.66 on earnings power, with a "watch" rating. It is still working from 2025 data, and it scored pricing power on a gross margin of 100%, which is an artefact. Its owner-earnings figure is close to our central value, and the earnings-power figure assumes no growth at all. We side with the owner-earnings number.