CES 2025 was just as hectic and eventful as every other year, with announcements from automotive to smart home, AI to autonomy and much more. Counterpoint’s Associate Director Greg Basich and Senior Analyst Murtuza Ali sat down with NVIDIA’s VP of Automotive Ali Kani to discuss how the company’s new foundation AI model Cosmos will accelerate autonomous vehicle development. Ali also discussed how NVIDIA’s partnership with Toyota will lead to the mass-market adoption of autonomy and in-cabin AI features.

The Interview

Key Takeaways from the Discussion

• NVIDIA, through Omniverse, is helping its partners create synthetic data to test their AVs through in-car computer Drive AGX. This enables a development loop from car to cloud to simulation.

• NVIDIA has launched generative AI “world foundation” model Cosmos, which enables users to create complete 3D, visual and physics-based scenarios that are rare in the real world and apply physics to them, what the company describes as “physical AI”.

• Ride-sharing giant Uber has partnered with NVIDIA to integrate the Cosmos platform and DGX Cloud.

• Toyota, one of the world’s largest automakers, has also announced plans to utilize NVIDIA’s DRIVE AGX Orin chipset, running on the NVIDIA DriveOS operating system.

Analyst Takeaways

• NVIDIA has launched Cosmos as a “world foundation model”. This is a generative AI model that can create realistic 3D scenes like other video generation models, and apply physics to these scenes.

• By generating many different scenes, like those that simulate rare, real-world driving scenarios that are otherwise difficult to find (like a construction zone with a worker holding a sign), autonomous vehicle testing can be done by using them as synthetic data.

• Cosmos also enables testing of a model by creating behaviors specific to a car — how it accelerates, turns and brakes, such that the AV now has traits (the automotive “DNA”, so to speak) of the automaker developing the vehicle.

• For Uber, Cosmos helps generate data that it can share with its partners that are developing AV software, helping them optimize it.

• Cosmos can help generate synthetic data that is very close to the real world, thus reducing the need for physical AV vehicle fleets and cutting testing and validation time and related costs, especially as companies test vehicles at higher levels of autonomy.

• The Toyota partnership involves the use of high-compute NVIDIA chips across multiple car lines, working with Tier 1 suppliers and Toyota’s Woven organization. This will provide a significant volume uplift for NVIDIA.