LAS VEGAS, Jan 7, 2026, 04:55 (PST)
- Nvidia said its new Vera Rubin chips are in “full production” and would deliver a fivefold jump in AI computing over its prior silicon. Reuters
- The company introduced Alpamayo, a new set of open models and tools it is pitching for reasoning-based autonomous driving.
- Lidar maker Aeva said its 4D lidar has been picked as a reference sensor for Nvidia’s DRIVE Hyperion platform, which pairs sensors with Drive AGX Thor and DriveOS for Level 3 and Level 4 features.
Nvidia said its next-generation Vera Rubin chips are in full production and introduced Alpamayo, software it says can help self-driving cars pick a path and leave a paper trail for engineers, as CEO Jensen Huang opened the Consumer Electronics Show (CES) in Las Vegas. “Not only do we open-source the models, we also open-source the data that we use to train those models,” Huang said. Nvidia is facing rising pressure in AI chips from Advanced Micro Devices and from customers such as Alphabet’s Google, which designs some of its own silicon. Reuters
The pitch lands as the auto industry keeps running into the same wall: the “long tail” — rare, awkward situations that barely show up in training data but can break an automated system. Nvidia said Alpamayo uses “vision-language-action” models, which turn sensor video into a driving plan, and can output step-by-step “reasoning traces” for post-drive audits. “Handling long-tail and unpredictable driving scenarios is one of the defining challenges of autonomy,” Uber’s Sarfraz Maredia said. NVIDIA Newsroom
Nvidia is tying those models to DRIVE Hyperion, a reference hardware-and-sensor design aimed at Level 4 autonomy — driving without a human in defined areas. The company said Hyperion uses two Drive AGX Thor systems-on-a-chip built on its Blackwell architecture to fuse a 360-degree sensor view and run transformer-based perception and planning. “By unifying compute, sensors and safety into one open platform, we’re enabling automakers, robotaxi operators and researchers to bring full autonomy to market faster,” Nvidia automotive chief Ali Kani said. NVIDIA Blog
Rubin is a six-chip platform that combines new GPUs and CPUs with networking and security parts, Nvidia said. In its rack-scale Vera Rubin NVL72 system, 72 Rubin GPUs and 36 Vera CPUs sit in one unit, and Nvidia said the design aims to cut the cost of AI “tokens” — the small chunks of text models process — to about one-tenth of the current Blackwell generation. Rubin-based products will be available from partners in the second half of 2026, the company said. NVIDIA Newsroom
Alpamayo 1, the first model in the family, has 10 billion parameters and is positioned as a “teacher” model rather than software that runs directly in the car. Nvidia is releasing model weights and an open dataset of more than 1,700 hours of driving data, plus an open simulation framework called AlpaSim, TechCrunch reported. “It does this by breaking down problems into steps, reasoning through every possibility, and then selecting the safest path,” Kani said at a press briefing. TechCrunch
Nvidia said its DRIVE AV software will debut in the all-new Mercedes-Benz CLA, bringing enhanced Level 2 point-to-point driver assistance to U.S. roads by the end of this year. Level 2 systems can steer and brake, but the driver must stay alert and be ready to take over. “NVIDIA is the intelligence backbone that makes every vehicle programmable, updatable and perpetually improving,” Kani said in a company blog post. NVIDIA Blog
Automotive and robotics revenue was $592 million in Nvidia’s fiscal third quarter ended Oct. 26, up 32% from a year earlier, the company said. Nvidia also said it had partnered with Uber to scale a level 4-ready mobility network starting in 2027, targeting 100,000 vehicles.
But “full production” does not always mean customers get volume shipments right away, and new driving stacks still face long validation cycles and regulators who want proof, not demos. Austin Lyons, an analyst at Creative Strategists, said the CES message around Rubin was meant to reassure investors: “This CES announcement around Rubin is to tell investors, ‘We’re on track.’” Big cloud buyers are also investing in custom chips to hedge against Nvidia’s pricing power, Wired reported. WIRED
Nvidia started opening parts of its autonomous-driving work in December with Alpamayo-R1, a model that turns sensor input into plain-language text as it chooses what to do. Katie Washabaugh, a product marketing manager for autonomous vehicle simulation at Nvidia, said the aim was to let developers understand and evaluate models in a standard way, rather than trusting a black box.
