What are the significant announcements made today by Nvidia CEO Jensen Huang at CES 2026

There was a lot of information provided to us during the keynote speech by Jensen Huang. At CES 2026, NVIDIA unveiled its most ambitious AI roadmap yet, headlined by the Rubin platform, a six-chip, extreme-codesigned AI system, and a suite of open models spanning healthcare, robotics, and autonomous driving. The company also showcased its vision for “physical AI” with real-world applications in vehicles, robots, and manufacturing.

Rubin is NVIDIA’s first co-designed, six-chip AI platform which succeeds the Blackwell architecture. Rubin GPUs produce 50 petaflops of NVFP4 inference.

Vera CPUs are optimized for data movement and agentic processing.

NVLink 6 and Spectrum-X provide High-speed networking.

ConnectX-9 SuperNICs and BlueField-4 DPUs for data center acceleration.

Inference Context Memory Storage is a new AI-native KV-cache tier offering 5x better performance per dollar and power efficiency.

This combined architecture delivers AI tokens at 1/10th the cost and accelerate time-to-market for frontier models.

NVIDIA emphasized its commitment to open, frontier AI models trained on its supercomputers across six domains:

1. Healthcare (Clara platform) for medical imaging, diagnostics.

2. Climate (Earth-2) for climate modeling and forecasting.

3. Reasoning (Nemotron) for multimodal and logical reasoning.

4. Robotics (Cosmos) for simulation and embodied AI.

5. Embodied AI (GR00T) for physical interaction and learning.

6. Autonomy (Alpamayo) for autonomous driving (Level 4-ready)

These models are open-source, enabling global collaboration and deployment.

Autonomous Driving with Alpamayo & Mercedes-Benz CLA

Alpamayo R1 is the first open reasoning Vision-Language-Action (VLA) model for AVs.

AlpaSim is an open simulation blueprint for high-fidelity AV testing.

Mercedes-Benz CLA is the first production car with Alpamayo, launching in the U.S. in 2026.

DRIVE Hyperion is NVIDIA’s modular, Level 4-ready AV platform which continues gaining traction with automakers and robotaxi firms.

DGX Spark & Personal Agents for AI on every desk.

DGX Spark is a desktop AI supercomputer with up to 2.6x performance for large models that supports local AI agents using Hugging Face models through a robot called Reachy Mini.

There will be new support for Lightricks LTX-12 and FLUX image models and NVIDIA AI Enterprise is coming to DGX Spark for broader deployment.

From Simulation to Real-World Robotics with Physical AI

NVIDIA Cosmos trains robots using video, simulation, and real-world data and is capable of generating realistic videos from single images, modeling edge-case driving scenarios, and performing physical reasoning and trajectory prediction.

Isaac Sim & Isaac Lab are used to train robots in photorealistic virtual environments.

There was also a partnership announced with Siemens that expands collaboration to integrate NVIDIA’s stack into industrial automation and digital twins.

NVIDIA now builds entire AI systems, from chips to software, to eliminate bottlenecks and accelerate innovation through open models.

CEO Jensen Huang said, “Our job is to create the entire stack so that all of you can create incredible applications for the rest of the world.”

https://blogs.nvidia.com/blog/2026-ces-special-presentation/