Uber + Nvidia robotaxi roadmap
Uber revealed plans to scale a robotaxi fleet using Nvidia DRIVE Hyperion and the Alpamayo LLM, targeting LA/SF in 2027 and 28 cities by 2028—signaling how travel companies are marrying perception stacks with LLM orchestration for mobility services. The timeline underscores the operational scale required for travel‑grade autonomous services. (x.com)
NVIDIA and Uber set an explicit scale target of roughly 100,000 Level‑4 vehicles over time and said they will back that scale with a joint AI data factory built on NVIDIA’s Cosmos world‑foundation model platform. (nvidianews.nvidia.com) The companies described a three‑phase city rollout that begins with instrumented data‑collection vehicles, shifts to operator‑led service, and then transitions to fully driverless Level‑4 operations. (investor.uber.com) NVIDIA framed Alpamayo as an open family of vision‑language‑action (VLA) teacher models that use chain‑of‑thought reasoning to tackle long‑tail driving scenarios and are intended to be fine‑tuned and distilled into in‑vehicle backbones. (nvidianews.nvidia.com) Alpamayo’s rollout is paired with simulation and dataset tools—AlpaSim and Physical AI datasets—designed for city‑specific fine‑tuning and closed‑loop validation before in‑vehicle deployment. (nvidianews.nvidia.com) NVIDIA is positioning its Halos safety certification program and Omniverse/Cosmos simulation stack as the production governance and curated data pipeline to validate physical‑AI behavior and support regulatory inspection. (nvidianews.nvidia.com) Uber described the operating model as a unified ride‑hailing network that will mix human drivers and autonomous vehicles on a single dispatch and payments platform, implying changes to matching, telemetry, and operator oversight practices. (nvidianews.nvidia.com) NVIDIA listed a broad OEM and AV ecosystem—Stellantis, Lucid, Mercedes‑Benz, Avride, May Mobility, Momenta, Nuro, Pony.ai, Wayve and WeRide—indicating Uber’s robotaxi deployments will rely on multi‑vendor hardware and software integrations built atop DRIVE Hyperion. (nvidianews.nvidia.com)
