NVIDIA brings Foxconn and Uber into its Hyperion robotaxi tent

Building a working autonomous car is hard enough, but turning it into a commercially viable robotaxi business usually requires a small army of partners to handle chips, manufacturing, and passenger dispatch.

At GTC Taipei, NVIDIA made its move to become the operating backbone for that exact army, announcing a major expansion of its Hyperion platform across Asia, Europe, and the Middle East.

Jensen Huang declared that vehicles are rapidly becoming physical robots and autonomous mobility is finally hitting its industrial scaling moment. His pitch is simple: instead of reinventing the stack, automakers and fleet operators can build on NVIDIA's pre-integrated foundation.

Foxconn, Uber, and a global manufacturing push

The partner roster reads like an assembly line of specialized giants carving up geographic territories.

In hardware manufacturing, Foxconn is expanding its partnership with NVIDIA to design and deploy level 4 electric robotaxis. The rollout kicks off in Kaohsiung, Taiwan, in 2028, initially covering airport-to-city transit before linking up with high-speed rail corridors and expanding throughout Asia.

Over in Southeast Asia, Vietnamese automaker VinFast is pairing its factory floors with software maker Autobrains to tackle the region's notoriously chaotic traffic environments at lower vehicle costs.

Europe gets a deployment through Uber, which will launch a robotaxi pilot in Munich later this year using Hyperion alongside Autobrains' Agentic AI driving stack. Uber plans to reveal the vehicle manufacturing partner for that project in the coming months.

Meanwhile, AI venture HUMAIN signed on to bring Hyperion-powered fleets to Saudi Arabia as part of the country's broader smart mobility infrastructure push.

The Android playbook for physical AI

At the technical core of this push is NVIDIA Hyperion, which packages in-vehicle compute with safety-certified software.

The system bundles NVIDIA DRIVE AGX compute, a suite of multimodal sensors, and the NVIDIA Halos full-stack safety system, anchored by the DriveOS foundation. This architecture allows developers to skip building low-level safety plumbing and focus on driving logic.

As Autobrains CEO Igal Raichelgauz pointed out, scaling self-driving tech cannot rely on a single brittle model to handle edge cases; it requires agentic systems that reason through uncertainty on standardized automotive silicon.

Uber's autonomous mobility chief Sarfraz Maredia echoed that sentiment, noting that the real commercial bottleneck has never been proof-of-concept prototypes, but integrating vehicle-agnostic autonomy into a ride-hailing network that riders actually use.

Why it matters

We think NVIDIA is aggressively executing the classic Android playbook before anyone else can establish a dominant platform.

Right now, the robotaxi landscape is starkly divided. On one side are vertically integrated giants like Waymo, which control their software, sensors, and operations with incredible precision but carry staggering capital burdens. On the other side are legacy automakers and software startups drowning in integration costs.

By staying strictly in the infrastructure tier, NVIDIA avoids the liabilities of fleet ownership while locking the entire ecosystem into its silicon and safety standards. Letting Foxconn handle metal, Autobrains handle autonomy, and Uber handle passenger demand is an undeniably clever division of labor.

For everyday commuters, this alliance could finally drag robotaxis out of limited test bubbles and into widespread urban transit by 2028. You won't have to live in Phoenix or San Francisco to hail an autonomous ride.

Still, multi-vendor consortiums carry obvious operational friction. When a vehicle runs on NVIDIA chips, with third-party reasoning code, manufactured by an electronics contractor, and dispatched on Uber, figuring out system liability during an inevitable edge-case failure will be a massive regulatory headache.