Uber Doesn't Own
the Cars. It Owns
the Roads.
Uber has signed partnerships with more than 25 AV companies in 18 months — Nvidia, Rivian, Waymo, Zoox, Wayve, Avride, WeRide and more. This is not diversification. It is a deliberate architectural strategy to become the operating system every self-driving car runs on.
In the same week — March 2026 — Uber announced partnerships with Nvidia for Level 4 robotaxis in LA and San Francisco, signed a $1.25 billion deal with Rivian for 50,000 autonomous R2s, confirmed a Tokyo launch with Wayve and Nissan, and extended its existing arrangements with Zoox. Four major AV partnerships. Seven days.
Most coverage treated each announcement separately. But the pattern they form together tells a more important story than any individual deal — and it is a story about platform strategy, not vehicle technology.
Uber is not competing with the companies building self-driving AI. It is collecting a toll from every single one of them. That is a fundamentally different position — and potentially a more durable one.
The architecture in plain terms
To understand what Uber is building, it helps to think in layers. The AV ecosystem has four distinct layers, and Uber has strategically positioned itself to own the one that all the others depend on.
Demand generation, dispatch, pricing, payments, safety monitoring, and the data factory that feeds back into every AV partner's training pipeline. This is what Uber owns — and what no AV company has been able to replicate independently.
The critical insight: every AV company that signs with Uber is acknowledging that building their own consumer-facing distribution across dozens of cities simultaneously is a harder problem than building the self-driving technology itself.
Nvidia's Drive Hyperion platform provides the shared compute infrastructure that multiple AV software stacks run on. The Alpamayo 1.5 model adds language-prompted scenario generation — developers describe edge cases in plain language and the model generates appropriate vehicle responses with full reasoning traces.
Uber, Lyft, Bolt, and Grab are all building on this shared Nvidia backbone, announced at GTC San Jose in March 2026.
This is where the actual self-driving AI lives — and where Uber's partners are all competing with each other. Waymo uses LiDAR-heavy sensor fusion and city-specific HD maps. Wayve uses camera-only embodied AI that generalises across environments without expensive mapping. Rivian builds its own end-to-end stack around the RAP1 chip. Zoox uses a purpose-built bidirectional vehicle with no steering wheel.
The key point: these companies are all competing at Layer 2 while simultaneously cooperating at Layer 4. Uber benefits from that competition — more capable AV technology means a better product on the Uber platform — while taking no technology risk itself.
The vehicles themselves — and Uber has deals at this layer too. Rivian R2 (50,000 vehicles), Nissan Leaf (Tokyo fleet), Lucid Gravity (20,000 target), Stellantis vans (European cities). In most cases Uber is not buying the vehicles. In the Rivian deal it is — a meaningful shift that signals Uber is willing to take asset risk when the exclusivity terms justify it.
Why every AV company needs Uber more than Uber needs them
The asymmetry in these partnerships is easy to miss but important to understand. Each individual AV company is replaceable on Uber's platform — if Waymo underperforms, Zoox picks up the rides. If Rivian's autonomy timeline slips, Wayve fills the gap in Tokyo. Uber's platform is indifferent to which technology wins.
The reverse is not true. No AV company has 170 million monthly active users. None of them have Uber's city-level regulatory relationships, its insurance frameworks, its customer support infrastructure, or its brand recognition in the markets where autonomous vehicles are launching. Building all of that from scratch, while simultaneously developing the underlying AI, is an almost impossibly capital-intensive proposition.
"The companies racing to build the best self-driving AI are all competing with each other. Uber is not competing with any of them. It is collecting tolls from all of them."
ITquette Autonomous Lab analysis — March 2026This is the same dynamic that made Microsoft dominant in personal computers, Google dominant in mobile, and Amazon dominant in cloud infrastructure. Own the platform, not the product. Let others compete at the product layer while you capture value from every transaction.
The flywheel that makes it self-reinforcing
The most durable aspect of Uber's position is that it compounds over time. More AV partners bring more vehicle supply, which lowers prices for riders, which drives more demand, which generates more ride data, which makes Uber's platform more valuable to the next AV partner that needs distribution. Each partnership makes the next partnership easier to close.
The data advantage
Every ride on the Uber platform generates training data that feeds back into AV development. Uber has accumulated over 3 million hours of autonomous vehicle data across its partner fleet. That data asset grows with every additional partner and every additional ride — and it sits on Uber's platform, not the individual AV company's servers.
The only risk to the model
Waymo has begun launching markets without Uber — Dallas, London, Denver, Miami, Washington DC — and separately partnered with Lyft for Nashville deployment. If the best AV companies decide they do not need Uber's distribution, the moat shrinks. But right now, no AV company has demonstrated it can replicate Uber's consumer reach independently at commercial scale.
What this means for incident management
There is a question that receives almost no coverage in the partnership announcements but will become increasingly important as deployments scale: when a Rivian R2 running Rivian's autonomous software on the Uber platform is involved in an incident in San Francisco in 2028, who owns the incident response?
Is it a Rivian incident because the vehicle and the software are Rivian's? Is it an Uber incident because the ride was booked through Uber and Uber took a fee? Is it a shared incident requiring coordinated response across both companies' operations, legal, communications, and regulatory affairs teams simultaneously?
The answer matters not just legally but operationally. The regulator notification goes to whom? The evidence is packaged by whom? The public statement is issued by whom? The corrective action is owned by whom? As the AV industry organises itself around platform models with multiple parties involved in each ride, the incident lifecycle frameworks that sit underneath the commercial agreements will need to be as carefully designed as the technology itself.
The partnership model Uber is building is commercially elegant. The operational and governance infrastructure that needs to sit underneath it is still being designed — and in most cases has not yet been built.
Related insights
Uber × Rivian — $1.25B and the exclusive clause
Why the exclusivity clause matters more than the investment figure. Read the full article →
Will AVs replace every car on the road?
The case for optimism, the structural barriers, and the five milestones that will tell us whether the optimists are right. →
Incidents are sequences, not moments
As AV partnerships multiply, incident ownership becomes distributed. Trust improves when the lifecycle is consistent and auditable.
Follow the research. Reach out if it resonates.
Published openly through this site and LinkedIn. If something connects with a challenge you are working through, we welcome the conversation.