How Marketplaces Tip

Key takeaways

• Multi-homing is the brake on tipping. Uber and Lyft both grew in 2025, Uber to $193.5B in gross bookings and Lyft to a record $18.5B, because switching apps costs a rider one tap.
• Take rate is a design choice, not a market rate: Airbnb's implied take was about 13.4% of gross booking value in 2025, Etsy's was 24.2%, and Uber's Mobility and Delivery segments ran roughly 30% and 19% in the same quarter.
• A tipped platform monetizes twice. Amazon's third-party seller services brought in $172.2B in 2025 and its advertising business another $68.6B. The second toll is charged to sellers already paying the first.
• Regulators now define tipping numerically: the EU's Digital Markets Act presumes gatekeeper status at 45 million monthly EU end users plus 10,000 yearly business users and €7.5B EU turnover.

What tipping actually means

A market tips when one platform's lead stops being reversible. Not when it gets big, but when it gets self-reinforcing. The distinction matters because most people use the phrase network effects to mean the first thing and then reason as if they had proved the second.

Here is the mechanism in plain terms. A marketplace has two sides that each want the other. Buyers go where the sellers are; sellers go where the buyers are. If that loop is tight enough, an early lead in one side pulls the other side across, which widens the lead in the first side, and so on. Past some point the smaller platform cannot win a seller by offering better terms, because the seller is not buying terms. The seller is buying access to buyers, and the smaller platform does not have them. That is tipping: the moment when a competitor's product quality stops being the binding variable.

The formal treatment of this is older and more careful than the popular version. Rochet and Tirole's 2003 paper in the Journal of the European Economic Association established that in a two-sided market the price structure matters independently of the price level: which side you charge, and how much, changes the outcome even when total revenue per transaction is identical. That is why marketplaces subsidize one side (free listings, driver bonuses, zero-commission introductory periods) and tax the other. The subsidy is not generosity; it is the purchase of the scarce side.

What the popular version gets wrong is the direction of the inference. Network effects are a reason a market could tip. They are not evidence that it will, and they are certainly not evidence that a specific platform is the one it will tip toward. Plenty of markets have textbook two-sided structure and have stayed split for a decade. The next section is about why.

The four conditions — and the one that usually fails

Tipping requires four things at once. Miss any of them and the market stays contested more or less indefinitely.

One: the cross-side effect has to be strong. Adding a seller must make the platform meaningfully better for buyers, not marginally better. A marketplace with 900,000 listings does not get much better at 950,000, because the buyer already had more choice than they could evaluate. This is why network effects tend to saturate: the first thousand sellers are transformative, the hundred-thousandth is a rounding error. Saturation is the reason a challenger can be a tenth the size and still deliver a perfectly good buyer experience.

Two: the effects have to be global rather than local. A rideshare network in Denver does nothing for a rider in Lisbon. Ride-hailing, food delivery, home services, and most labor marketplaces are stacks of independent city-level networks wearing one brand. That is why they are won city by city and why a well-capitalized local entrant is a real threat. Contrast a global goods marketplace, where a seller in Guangzhou genuinely improves the catalog for a buyer in Ohio.

Three: users must single-home. If participants can cheaply use several platforms at once, the winner never gets exclusivity over the scarce side, and the loop never closes. This is the condition that fails most often, and it fails hardest in exactly the markets people confidently predict will tip.

Four: the market must not be differentiable into defensible niches. If a segment of buyers wants something the giant is structurally bad at (handmade goods, vintage instruments, industrial surplus, regulated categories) that segment supports its own network, and the giant's scale does not reach into it.

Run those four against any marketplace you are evaluating and you will usually find the argument for inevitability resting entirely on condition one, with two through four quietly assumed.

Multi-homing is the brake, and the numbers show it

US ride-hailing is one of the cleanest live experiments in platform economics, because the theory said it should have tipped years ago and it has not.

Both platforms are growing. Uber reported full-year 2025 gross bookings of $193.5 billion, up 19%, on revenue of $52.0 billion and 13.6 billion trips. Lyft reported full-year 2025 gross bookings of $18.5 billion, up 15%, revenue of $6.3 billion, and 945.5 million rides, its own record year, with $528.8 million of adjusted EBITDA and $1.12 billion of free cash flow. That is not a dying competitor being slowly starved by a network effect. That is a second network compounding at a healthy rate inside a market the first one supposedly owns.

The reason is that both sides multi-home and it costs them almost nothing to do so. A rider comparing two prices switches apps in seconds. A driver runs both apps on the same windshield mount and takes whichever ping pays better. Neither side has an asset stranded on one platform. Because drivers multi-home, the effective supply available to each platform is far larger than each platform's exclusive supply, which means the smaller platform can offer a competitive wait time even at a fraction of the leader's scale. Once the challenger's service quality is good enough, the network advantage stops converting into a user-visible advantage, and price and habit take over.

Compare that to a market where one side is locked. A seller who has spent three years accumulating reviews, search rank, and fulfillment integration on a goods marketplace cannot replicate that elsewhere in a weekend. Their history is the switching cost, and it is held by the platform. Where that asymmetry exists, tipping is live. Where it does not, it is not.

The practical test is one question: what does the scarce side lose by adding a second platform tomorrow? If the answer is nothing, stop modeling winner-take-most. If the answer is reputation, ranking, financing, integration, or physical inventory sitting in the platform's warehouse, start.

Reading a take rate

Take rate, platform revenue divided by the gross value transacted across it, is the single most informative number a marketplace publishes, and it is almost always misread as a measure of greed. It is better read as a measure of how much of the transaction the platform actually performs, and how little the participants can do about it.

The 2025 spread across four public marketplaces makes the point. Airbnb took roughly 13.4% of $91.3 billion in gross booking value ($12.24 billion of revenue). Etsy reported a consolidated take rate of 24.2% on $11.9 billion of gross merchandise sales, up 190 basis points from 22.3% the prior year. Uber's fourth-quarter Mobility segment implied about 29.9% of gross bookings as revenue ($8.2 billion on $27.4 billion), while its Delivery segment implied about 19.2% ($4.9 billion on $25.4 billion) in the very same quarter. One company, one quarter, an eleven-point spread between two segments.

A caution the headlines skip: these are implied take rates, computed as reported revenue over reported gross value. They are not commission rates. Uber and Lyft record revenue on a gross basis in some markets and arrangements, which inflates the implied figure relative to what any individual driver experiences. Amazon, meanwhile, discloses $172.2 billion of third-party seller services revenue for 2025 but does not disclose marketplace gross merchandise value in the 10-K at all, so its take rate cannot be computed from the filing, only estimated, and you should distrust anyone who quotes it to one decimal place.

What drives the spread is not market power alone. It is three things: how much work the platform does (Uber dispatches, prices, insures, and processes payment; Airbnb largely matches and holds funds), how contested the scarce side is (an Uber driver's alternative is Lyft; an Etsy seller's alternative is building their own store from zero), and how much of the transaction happens off-platform (a host cleans the house, a driver supplies the car).

Illustrative only, and the arithmetic is deliberately round: imagine a marketplace facilitating $10 million of annual gross value at a 15% take, so $1.5 million of revenue. Raising the take to 18% adds $300,000 of revenue if nothing else moves. Suppose it instead drives 8% of gross value off the platform, with sellers steering repeat buyers to direct channels. Volume falls to $9.2 million, revenue becomes $1.656 million, and the increase is $156,000, roughly half the naive projection. Now suppose the leakage is 20%: $8 million at 18% is $1.44 million, which is less than the original. The take-rate curve has a peak, the peak moves as competitive alternatives improve, and platforms discover its location by overshooting it.

What tipping does to everyone standing on the platform

A tipped marketplace does not stop monetizing at the commission. It monetizes a second time, by selling back the visibility the first payment used to include.

Amazon's advertising services revenue was $68.6 billion in 2025, against $56.2 billion in 2024 and $46.9 billion in 2023, a 46% increase in two years, disclosed as a separate line in the 10-K alongside the $172.2 billion of third-party seller services. Those are, in large part, the same sellers paying twice: once for the right to transact, and again for the right to be found. Nothing about that is unlawful or even surprising; it is what a platform does when it controls a scarce resource (position in the result set) that participants cannot obtain any other way. But it changes the arithmetic of being a seller in a way the commission schedule alone will not show you.

The drift is measurable if you watch it. Etsy's take rate rose from 22.3% to 24.2% in a single year, and that year consolidated gross merchandise sales fell 5.3% to $11.9 billion and active buyers fell 2.0% to 93.5 million, while active sellers rose 7.7% to 8.8 million. Read those four numbers together and you get a precise picture of a mature marketplace: more sellers competing for fewer buyers, with the platform's share of each transaction rising. That is the equilibrium every long-lived marketplace converges toward, and it is why a seller's cohort year matters so much. The person who joined when the platform was buying supply had a different business than the person who joins when it is harvesting.

The defensive posture is not to avoid tipped platforms, since they hold real demand you cannot cheaply replicate, but to hold a hard ceiling on dependence and to know your number. Track platform revenue concentration the way you track customer concentration. Track your all-in cost of participation, commission plus advertising plus fulfillment, as one percentage of gross sales, and watch its trend rather than its level. A take rate that rises 190 basis points a year compounds into something structural in five years.

When the market answers back

Tipping is now a regulated condition with numeric thresholds attached, which is a genuinely new fact about the world and one operators should build into their planning.

The EU's Digital Markets Act does not ask whether a platform is dominant in the traditional sense. Article 3(2) presumes gatekeeper status when an undertaking provides a core platform service in at least three Member States with at least 45 million monthly active end users in the Union and at least 10,000 yearly active business users, and either has annual Union turnover of at least €7.5 billion in each of the last three financial years or an average market capitalisation of at least €75 billion. Meet the arithmetic and the presumption applies; the undertaking may rebut it, and the Commission may designate a platform that misses the thresholds but satisfies the qualitative criteria. The design intent is explicit: replace years of case-by-case market-definition litigation with a bright line that catches tipped positions early.

On the enforcement side, the US Department of Justice announced on 2 September 2025 that the District Court for the District of Columbia had ordered remedies in its monopolization case against Google in online search: a prohibition on exclusive contracts for the distribution of Google Search, Chrome, Google Assistant and the Gemini app, and orders to make certain search index and user-interaction data available to rivals and to offer search and search text ads syndication. The Department described the market it was trying to open as having been “frozen in place for over a decade.” That phrase is the cleanest available description of what a tipped market looks like from the outside: not a market where the leader keeps winning, but one where the question stopped being asked.

For an operator, the strategic reading is threefold. First, tipping risk is asymmetric: it is catastrophic if you are the challenger and merely expensive if you are a participant, so size the bet accordingly. Second, if you are building a marketplace, the four conditions above tell you whether you are racing toward a defensible position or funding a permanent subsidy war; global scarce-side lock-in is the only version worth the capital. Third, if you are selling on someone else's tipped platform, regulatory intervention is a real but slow tailwind. It arrives on the timescale of appeals, not quarters, and it is not a substitute for owning a demand channel nobody can reprice.

Put it to work

Answer four questions in writing about the platform you depend on: is the cross-side effect still strong at your scale, is it global or city-by-city, can the scarce side multi-home cheaply, and is there a niche the leader is structurally bad at. Then track your all-in cost of participation (commission, ads and fulfillment as one percentage of gross sales) over four quarters and watch the slope, not the level.

Sources & references

Linked entries open the named source directly. Entries without a link say exactly what kind of reference they are — and how to check them yourself.

• Etsy, Inc. — Q4 and Full Year 2025 results (GMS, revenue, 24.2% take rate, buyers and sellers)
• Uber Technologies — Q4 and Full Year 2025 results, filed with the SEC (gross bookings, revenue, segment detail)
• Lyft, Inc. — Q4 and Full Year 2025 results, filed with the SEC (gross bookings, revenue, rides)
• Amazon.com, Inc. — Form 10-K for fiscal year 2025 (third-party seller services and advertising services revenue)
• Airbnb, Inc. — Q4 2025 Shareholder Letter (gross booking value, revenue, implied take rate)
• Digital Markets Act, Regulation (EU) 2022/1925 — Article 3, designation of gatekeepers (full Article text)
• European Commission — Digital Markets Act: Commission designates six gatekeepers, 6 September 2023
• U.S. Department of Justice — remedies ordered in United States et al. v. Google LLC, 2 September 2025
• Rochet & Tirole, "Platform Competition in Two-Sided Markets," Journal of the European Economic Association 1(4): 990–1029 (2003) — Toulouse School of Economics record
• U.S. Census Bureau — Quarterly Retail E-Commerce Sales, 4th Quarter 2025 (e-commerce was 16.4% of total US retail sales in 2025)
• Implied take rates are derived — Take rates quoted here are computed as reported revenue divided by reported gross bookings or gross merchandise value, except Etsy’s 24.2%, which the company reports directly. Uber and Lyft record revenue gross in some markets, so implied take rates are not commission rates. Amazon does not disclose marketplace GMV in its 10-K, so no Amazon take rate is stated.