NVIDIA rolls out new AI models and infrastructure at CES 2026

• At CES 2026, NVIDIA introduced new AI systems for real-world decision-making.
• The focus is on reasoning and rare scenarios.

The latest wave of AI development is moving beyond chat interfaces and into systems that must act in the real world. Cars, robots, and industrial machines face situations that cannot be solved by pattern matching aloneespecially when rare or unexpected events occur.

A set of new models and platforms from NVIDIA highlights how the industry is responding to that challenge. The releases span reasoning-based autonomous vehicle models, robotics tools, and new compute and storage designs, all aimed at supporting AI that can plan, adapt, and explain its actions in the real world.

Reasoning enters autonomous driving

One of the hardest tests for physical AI is autonomous driving. Vehicles must operate safely across countless conditions, including rare and unpredictable events often described as the “long tail.” Traditional designs separate perception from planning, which can limit how well systems respond when something unusual happens.

New approaches focus on end-to-end learning, but these still face limits when scenarios fall outside training data. Addressing that gap requires models that can reason step by step, rather than react based only on past examples.

The Alpamayo family of open models and tools is designed around this idea. Instead of running directly inside vehicles, the models act as large teacher systems. Developers can fine-tune and distil them into smaller components that fit within existing autonomous driving stacks.

Alpamayo 1, released to the research community, uses video input to generate both driving trajectories and reasoning traces that show how decisions are made. By exposing the logic behind each action, the model is meant to help teams test behaviour in rare situations and improve explainability, which remains a key concern for safety and trust.

“The ChatGPT moment for physical AI is here — when machines begin to understand, reason and act in the real world,” said Jensen Huang, founder and CEO of NVIDIA. “Robotaxis are among the first to benefit. Alpamayo brings reasoning to autonomous vehicles, allowing them to think through rare scenarios, drive safely in complex environments and explain their driving decisions — it’s the foundation for safe, scalable autonomy.”

Alongside the models, Alpamayo includes an open simulation framework and large-scale datasets drawn from thousands of hours of driving across different regions and conditions. Together, these pieces are intended to form a loop where models are trained, tested in simulation, refined, and then distilled for real-world use.

Industry interest reflects the broader challenge autonomy still faces. “Handling long-tail and unpredictable driving scenarios is one of the defining challenges of autonomy,” said Sarfraz Maredia, global head of autonomous mobility and delivery at Uber. “Alpamayo creates exciting new opportunities for the industry to accelerate physical AI, improve transparency and increase safe level 4 deployments.”

From single-task machines to reasoning robots

The same limits appear in robotics. Many machines today are built for one task and require heavy engineering to change their behaviour. Moving beyond that model means teaching robots to understand their surroundings, plan actions, and adapt when tasks or conditions shift.

New open models for robotics aim to reduce the cost and effort needed to reach that point. Instead of starting from scratch, developers can build on pretrained systems that already capture how the physical world behaves.

Recent releases include world models that generate synthetic data and test robot behaviour in simulation, as well as reasoning vision-language models that link what a machine sees with what it should do next. For humanoid robots, this extends to full-body control, allowing systems to coordinate movement based on context rather than fixed scripts.

“The ChatGPT moment for robotics is here. Breakthroughs in physical AI — models that understand the real world, reason and plan actions — are unlocking entirely new applications,” said Jensen Huang. “NVIDIA’s full stack of Jetson robotics processors, CUDA, Omniverse and open physical AI models empowers our global ecosystem of partners to transform industries with AI-driven robotics.”

Developers in manufacturing, logistics, and healthcare are already testing these approaches. In surgical settings, for example, autonomous systems are being trained in simulation to guide instruments using real-time analysis. In industrial environments, robots are learning new behaviours without being reprogrammed line by line.

The common thread is a shift from rigid automation toward systems that can generalise across tasks, even when conditions differ from what they were trained on.

Infrastructure catches up with reasoning AI

As models become more complex, the demands placed on infrastructure increase. Large reasoning systems rely on massive amounts of context, especially for multistep decision-making. This context is often stored as key-value caches that help models remain consistent across long interactions.

Keeping that data on GPUs is not practical at scale. It creates bottlenecks and limits how many systems can run at once. To support long-context, multi-agent AI, storage itself must change.

New approaches focus on treating context as a shared resource that can be accessed quickly across clusters. By extending memory beyond individual GPUs and managing it at the system level, AI workloads can maintain responsiveness while scaling to larger models.

“AI is revolutionising the entire computing stack — and now, storage,” said Jensen Huang. “AI is no longer about one-shot chatbots but intelligent collaborators that understand the physical world, reason over long horizons, stay grounded in facts, use tools to do real work, and retain both short-and long-term memory.”

At the same time, compute platforms are being redesigned to support these workloads end to end. New chip families bring together CPUs, GPUs, networking, and data processing units as a single system, rather than separate components stitched together later.

“Rubin arrives at exactly the right moment, as AI computing demand for both training and inference is going through the roof,” Huang said. “With our annual cadence of delivering a new generation of AI supercomputers — and extreme codesign across six new chips — Rubin takes a giant leap toward the next frontier of AI.”

For developers and cloud providers, the appeal lies in efficiency. Training and inference costs matter as much as raw performance, especially as models grow larger and more specialised. Reducing the number of chips needed for complex workloads lowers both energy use and operational overhead.

“Intelligence scales with compute. When we add more compute, models get more capable, solve harder problems and make a bigger impact for people,” said Sam Altman, CEO of OpenAI. “The NVIDIA Rubin platform helps us keep scaling this progress so advanced intelligence benefits everyone.”

A shared direction for physical AI

Across vehicles, robots, and infrastructure, the direction is consistent. Physical AI is moving away from narrow optimisation and toward systems that can reason, adapt, and explain their actions. Open models, simulation tools, and shared datasets are lowering barriers to experimentation, while new hardware and storage designs aim to make long-context reasoning practical at scale.

What remains unresolved is how quickly these systems can move from controlled environments into everyday use. Rare events, safety concerns, and real-world complexity continue to test even the most advanced models. But the focus has shifted. The challenge is no longer whether machines can act, but whether they can think through what they are doing when the world does not behave as expected.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology eventsclick here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.