The Robotaxi Race Became an AI Race — and It's Basically Settled. The Real Fight Is Whether You Own the Fleet, the Cars, or Just the Brain.

The robotaxi race stopped being about cars a while ago. Waymo, Tesla, and the British startup Wayve — which just raised $1.2 billion at an $8.6 billion valuation, stretched to about $1.5 billion — are all building the same thing under the hood: an end-to-end neural network that takes camera input and outputs driving, trained on internet-scale data. It's a vision-language-action model with a steering wheel. What separates these companies isn't the AI approach anymore; that question has quietly been settled. It's what each one does with the model, and the three answers are the actual race.

Here's the frame. Three business models are wrapped around one AI bet. Waymo owns the fleet — it builds the car, the sensor stack, and the service, vertically integrated and capital-heavy. Tesla owns the installed base — it put its driver software in six million cars people already bought. Wayve owns nothing and sells the brain: it licenses the AI Driver to Uber's fleet and to Mercedes, Nissan, and Stellantis, and lets them supply the metal. Asset-heavy, already-deployed, asset-light. Same underlying model, three incompatible theories of how you actually win with it.

That the AI itself has converged is the part worth dwelling on, because it reframes everything. Wayve's AI Driver and Waymo's EMMA — which is powered by Gemini — are both end-to-end neural nets, the camera-to-controls approach that has now beaten the old modular pipeline of separate perception, prediction, and planning modules. When the two leading players are running architecturally the same kind of system, the model stops being the moat. The moat becomes distribution, capital structure, and who's willing to take the liability — which is to say, the business model is the technology story now.

Wayve's version is the most interesting and the most exposed. Selling the driving intelligence across other people's fleets and other people's cars is genuinely asset-light: if it works, it scales faster than Waymo's owned-fleet model because Wayve never has to build a car. But it also means Wayve controls almost nothing that matters to whether it succeeds. It needs Uber to launch, it needs Mercedes and Nissan and Stellantis to ship vehicles with the right hardware, and it needs regulators to say yes. The upside of owning only the brain is you scale without capital; the downside is everyone else holds the steering wheel.

And then there's the detail that quietly reframes the whole "race": Uber is backing both Wayve and Waymo, and intends to run both in London. The distributor has deliberately refused to pick a winner, because it doesn't need one — it owns the demand, the app, the customer, and it will happily rent whichever brain performs. So the AI companies aren't really racing each other to beat Uber; they're racing to be Uber's supplier, and the supplier is by definition replaceable. The most comfortable seat in the robotaxi war belongs to the company that isn't building the AI at all.

Two reality checks the headlines skip, because the announcement runs ahead of the deployment. First, London hasn't launched. As of now the Uber-Wayve service is pre-launch, waiting on UK regulatory approval, with Wayve saying "the next couple of months" — and Waymo, also Uber-linked, has been testing there since spring. Both are pre-commercial in their lead market; the three-way showdown is imminent, not live. Second, and more important: the "AI Driver in consumer cars from 2027" everyone quotes is L2+ supervised driver-assist — steer, navigate, respond to traffic under driver supervision — not robotaxi autonomy. Don't conflate the L4 robotaxi, which is a handful of cars in one geofenced city with a safety case, with the L2+ feature that ships in a consumer Nissan. Wayve is selling both, and they are very different products. The $8.6 billion is priced on the first; the near-term revenue is the second, which is a better lane-keeper.

So watch three things: who actually launches London first once the regulator clears it, because robotaxi network effects are winner-take-most within a city; whether the 2027 L2+ consumer product ships as promised; and whether end-to-end AI driving demonstrably outperforms the modular stacks in a live commercial market, which remains unproven at scale. My read is that the architecture debate is over — end-to-end won — so this is now a business-model contest, and Wayve's sell-the-brain bet is simultaneously the most scalable and the most hostage to other people's execution. The one who has already won, quietly, is Uber, which converted the AI companies into interchangeable suppliers and kept the customer for itself. What would change my mind is concrete: Wayve launching London on schedule and its license model proving out across both a live Uber fleet and a shipping consumer vehicle. Until then, $8.6 billion is priced on a robotaxi waiting for a regulator and a consumer product that's a driver-assist feature — and on a bet that owning the brain beats owning the road.