CES 2026 was the stage for a major advancement in the world of autonomous driving, with Jensen Huang, the CEO of Nvidia, revealing Alpamayo, an artificial intelligence platform that promises to make cars think. Imagine for a moment vehicles capable of understanding the world around them, anticipating unexpected behaviors, and reacting like a human driver would. A true revolution that could redefine our relationship with the road.
Significant Progress in Autonomous Driving
Autonomous driving has made considerable advancements in recent years, but a major challenge remains to be overcome: managing rare and unpredictable situations. These moments where everything can change, whether it’s an unexpected behavior from a pedestrian or extreme weather conditions, continue to pose a puzzle even for the most sophisticated systems. At CES in Las Vegas 2026, Huang emphasized this reality, highlighting that current technologies must evolve to address these complex scenarios.
Cars Now “Think”
With Alpamayo, Nvidia introduces a radical change in the architecture of autonomous driving systems. Gone are the days when perception and planning were treated separately. Enter the Vision-Language-Action (VLA) models, which integrate reasoning and understanding of cause-and-effect relationships. These models do not just execute tasks; they can explain their decisions, a crucial aspect for improving transparency and user trust. This could very well be the necessary boost for the mass adoption of autonomous driving.
Huang referred to a true “ChatGPT moment” for physical AI, a period where machines begin to understand and act autonomously in the real world. According to Nvidia’s forecasts, robotaxis and level 4 autonomous vehicles will be among the first to benefit from this technological advancement, making our roads safer and smarter.
An Open Ecosystem for Next-Generation Autonomy
Alpamayo is not just a single model; it is a true open ecosystem based on three fundamental pillars. The first is Alpamayo 1, the first VLA model specifically designed for research in autonomous driving. With its 10 billion parameters, it uses video inputs to generate driving trajectories while documenting the reasoning behind each maneuver. This model is published as open source on Hugging Face, allowing developers to adapt it to their specific needs.
Toyota e-Palette 2025
The second pillar is AlpaSim, a fully open-source simulation environment that realistically reproduces sensors, traffic, and driving dynamics in various environments. Simulation remains a key tool for validating algorithms before they hit the road, thereby ensuring increased safety.
Finally, the last pillar is based on the Physical AI Open Datasets, one of the largest open datasets for autonomous driving. Composed of over 1,700 hours of real driving collected in various geographical contexts, this dataset particularly emphasizes rare and complex scenarios.
Support from the Automotive Industry
This innovative approach by Nvidia has already attracted the interest of key players in the automotive sector. Companies like Lucid, Jaguar Land Rover, and Uber, as well as academic institutions such as Berkeley DeepDrive, see in Alpamayo a true concrete accelerator for the development of AV (autonomous vehicle) stacks based on reasoning and aimed at level 4 autonomy. The openness of the models and datasets is seen as a crucial strategy to collectively tackle the challenges of autonomous driving.
Gallery: Toyota e-Palette 2025
