Big Announcements From NVIDIA GTC Taipei 2026: AI Models, Robotics, Healthcare, And More!

AI is no longer limited to research labs or software tools. It’s gradually becoming part of real-world systems that move, build, and make decisions. From self-driving cars to hospitals and chip factories, companies are working to bring intelligence into physical environments in more practical ways. NVIDIA’s latest updates at GTC Taipei 2026 highlight this shift clearly.

 

TL;DR

 

• NVIDIA launches Alpamayo 2 Super with AlpaGym to boost robotaxi AI training and simulation.
• Cosmos 3 brings multimodal reasoning and world simulation for robotics and physical AI.
• Foxconn, TSMC, and NVIDIA expand AI into healthcare and chip manufacturing workflows.

 

Alpamayo 2 Super With AlpaGym And Simulation Tools

NVIDIA launched Alpamayo 2 Super, a 32-billion-parameter open AI model designed for robotaxis. The model can reason, plan, and act across the driving stack, supporting safer Level 4 autonomy. Alongside it, the company introduced AlpaGym, a closed-loop reinforcement learning framework, as well as OmniDreams and Omniverse NuRec, which help simulate realistic driving environments and convert real-world data into 3D training worlds.

Foxconn And NVIDIA Agentic AI For Taiwan Healthcare

In healthcare, NVIDIA and Foxconn are working together to bring agentic AI into Taiwan’s medical system under the “Healthy Taiwan” initiative. AI agents are now supporting doctors with diagnosis, reporting, and care coordination, while Nurabot robots assist with hospital operations. These systems are moving from pilot programs to real clinical environments across multiple medical centers in Taiwan.
 

Cosmos 3 With Multimodal Reasoning

In robotics and physical AI, NVIDIA introduced Cosmos 3, an open foundation model built for reasoning, simulation, and action generation across multiple formats, including text, image, video, sound, and motion. The Cosmos 3 lineup includes Nano for fast reasoning tasks and Edge for real-time applications. These models are designed to help developers build more capable physical AI systems that interact with the real world.

NVIDIA And TSMC AI For Semiconductor Fabs

For semiconductor manufacturing, NVIDIA and TSMC are using AI and accelerated computing to improve chip design and production. CUDA-X libraries help speed up processes like lithography, transistor simulation, and fab optimization. TSMC is also using NVIDIA Metropolis and TAO Toolkit for defect detection at nanometer scale. In addition, NVIDIA Omniverse is being used to build digital twins, including FabTwin, to improve planning and efficiency in fabs.

These updates show how NVIDIA’s ecosystem is expanding across industries where AI is not just supporting software, but actively shaping physical systems, industrial workflows, and real-world decision-making at scale.