The Robotaxi Doesn't Beat the Platform It Was Built to Replace

Based on the research ofNoh, Tunca, and Xu, "Evolution of Ride Services: From Ride Hailing to Autonomous Vehicles," Management Science, 2026

The pitch everyone in the robotaxi race believes

The story Waymo and Tesla tell investors is a story about control. Own the vehicle, own the sensor stack, own the software, and you own the unit economics: no driver to pay a majority of the fare to, no independent contractor who can log off during a snowstorm, no principal-agent problem at all. Uber and Lyft, by this telling, are just software wrapped around someone else's car and someone else's labor, a thinner business that a fully integrated AV operator should eventually steamroll. It is a clean story, and it is the story behind tens of billions of dollars of capital allocated to Waymo, Tesla, Zoox, and their peers.

A new game-theoretic paper in Management Science, by Daehoon Noh, Tunay Tunca, and Yi Xu, takes that pitch seriously enough to model it formally, and finds a sharper, stranger answer. The AV firm's vertical integration is a genuine advantage: it lets the firm commit to a fleet size before the platform can react, the classic first-mover capacity play from industrial organization. But a commitment is a two-edged asset. It is a promise the AV firm cannot revise once demand arrives, while the ride-hailing platform's driver supply is a promise it renegotiates constantly, hour by hour, through pay and pricing. In small markets, or with impatient customers who won't wait for a car, that rigid commitment can indeed scare the platform into softer competition and let the AV firm win. But in larger markets with sufficiently patient customers, the paper's equilibrium flips: the platform's flexibility becomes the dominant asset, and the incumbent ride-hailing firm keeps the upper hand even against a fully autonomous, vertically integrated rival.

What "capacity commitment" costs in practice

You do not need the algebra to see this playing out on the street. Waymo's own numbers show the mechanism at work. The company was completing more than 250,000 paid trips a week in the spring of 2025 across roughly 1,500 vehicles in San Francisco, Los Angeles, Phoenix, and Austin; by late March 2026 it had doubled that, to about 500,000 weekly rides across ten cities. But as Waymo pushed its US footprint further, expanding from ten cities to fifteen, the analytics firm Business Model Analyst found the weekly ride count barely moved off that 500,000 mark, while trips per vehicle fell roughly 25 percent, from about 167 a week in May 2025 to about 125. Every new metro carries the same fixed costs on day one, tow contracts, depot leases, charging queues, fire-department relationships, regardless of how much demand shows up, and each new city just spreads the same finite fleet thinner. That is a capacity commitment behaving exactly as the model predicts: expandable in principle, but each expansion is a slow, capital-heavy, discrete decision, not a dial you turn with a price.

Contrast that with what happens inside a single city when demand spikes. A ride-hailing data firm, Obi, analyzed nearly 100,000 San Francisco rides between Thanksgiving and New Year's Day and found Waymo's average wait time climbing to 5.74 minutes, against 3.15 minutes for Uber and 5.14 for Lyft, attributing the gap specifically to a Waymo capacity crunch during the 4 to 6 p.m. rush. Uber and Lyft close that same gap by paying more per ride during the surge, which pulls idle drivers back onto the road within minutes; Waymo can only reposition the cars it already owns. It cannot conjure a 501st vehicle for the dinner rush, and it cannot quietly retire idle ones during a Tuesday-afternoon lull, because both the buying and the parking of an AV are capital decisions, not shift decisions.

A fleet you own is a promise you can't take back when demand disappoints, and can't extend when demand spikes.

Tesla's robotaxi effort makes the same point at a smaller, rawer scale. The company began carrying riders in Austin in mid-2025 with a safety monitor in the passenger seat, and Elon Musk announced in January 2026 that some vehicles had begun operating without one, though reporting at the time found no verified rides carrying a paying customer with nobody in the car providing oversight, and the unsupervised group stayed a small slice of the fleet. The scale has stayed correspondingly tiny: even after a Cybercab launch event on September 3, 2026, briefly pushed the number of Tesla robotaxis actually seen carrying riders in Austin past 100, Electrek reported that the actively-trip-carrying fleet fell back to just 8 vehicles by September 22, out of 476 vehicles registered with the Texas DMV under the state's new AV rules. A vertically integrated AV company does not get to fake capacity it hasn't built; the fleet is either on the road or it isn't, and building it is slow.

Uber doesn't want to own the robots either

The clearest evidence that platforms understand this dynamic is what Uber has actually done with Waymo, rather than what either company says about the future of driving. Uber and Waymo expanded their partnership in September 2024 to bring Waymo's cars onto the Uber app in Austin and Atlanta, building on an earlier Phoenix arrangement that Uber CEO Dara Khosrowshahi said had already carried "tens of thousands of riders." That deal put Uber, not Waymo, in charge of fleet management: vehicle cleaning, repair, and depot operations, exactly the unglamorous capital-intensive work a platform prefers to keep flexible rather than own.

Then, in the summer of 2026, the arrangement started coming apart in a telling way. Waymo and Uber ended their Phoenix collaboration that June, with Waymo folding those vehicles into its own app rather than Uber's, and Uber said it was lining up a different autonomous-vehicle partner for the city instead of naming Waymo as its exclusive robotaxi supplier there. By July, the two companies had agreed to end exclusivity in Atlanta and Austin as well. Read individually, each headline sounds like friction. Read together, they describe a platform acting exactly like a platform: Uber does not want to be locked into any single AV supplier's fixed fleet any more than it wants to own the fleet itself. It wants Waymo's cars, Tesla's cars, and human drivers all interchangeable, addable and droppable supply nodes it can route demand across, which is precisely the flexibility the paper identifies as the platform's structural weapon. Waymo, meanwhile, keeps pulling toward exclusivity and its own branded app, because a company that has sunk capital into owned vehicles wants to control how that capital gets used. The two firms are not fighting over a robotaxi feature. They are fighting over who gets to hold the flexible position in the deal.

The welfare warning inside the equilibrium

The paper's least comfortable finding is not about which firm wins. It is about what happens to everyone else when the AV firm does win a market away from an incumbent ride-hailing monopolist. Because the AV entrant competes by committing a fixed, capital-constrained fleet rather than by inviting in a flexible pool of independent drivers, its entry can shrink total vehicle supply in that market rather than grow it, and lengthen customer wait times in the process, even though the whole premise of the robotaxi pitch is that automation is the efficiency upgrade. A platform can always summon a marginal driver by raising the price a little; an AV firm can only summon a marginal car if it already built one and parked it nearby. Swap out the flexible fleet for the rigid one, and you can end up with a market that looks more "advanced" and serves its customers worse.

That is the sentence regulators and city transportation departments should sit with as they wave robotaxi pilots through: the vertically integrated, seemingly more sophisticated entrant is not the automatic upgrade the marketing suggests, and displacing a platform with an owned fleet can be, in the paper's own terms, a welfare loss dressed up as progress. For operators, the lesson is narrower but just as sharp: before you commit capital to owning the road, ask whether the market you're entering is one where your customers will wait for you, and whether it is big enough to reward a commitment you cannot undo. If the answer is no on either count, the platform you're trying to replace may simply out-flex you.