Dubai — NVIDIA founder and CEO Jensen Huang used his GTC Taipei keynote to announce a wave of new products, platforms, and partnerships that the company says will power what he described as the largest infrastructure buildout in human history, with AI factories scaling from the cloud to enterprises, industries, sovereign infrastructure, and billions of edge devices.
Central to the announcements was a reinvention of the Windows PC for the age of personal AI. NVIDIA and Microsoft unveiled a joint initiative to bring AI agents natively to Windows PCs powered by GeForce RTX and DGX Spark, positioning the desktop as the next frontier for agentic computing.
The DGX Station expanded that vision further, with NVIDIA describing it as a trillion-parameter AI supercomputer on every enterprise desk. Running on Windows, the station is designed to let developers and enterprises prototype, fine-tune, and deploy advanced AI models locally without depending on cloud-scale data centres.
Inside the data centre, NVIDIA confirmed that the Vera Rubin platform has entered full production. Positioned as the successor architecture to Blackwell, Vera Rubin is engineered to power what NVIDIA calls agentic AI factories, large-scale facilities purpose-built to train and serve autonomous AI agents at scale.
Vera Rubin is paired with a new CPU named Vera, which NVIDIA described as the world’s first CPU built specifically for AI agents.
Complementing the chip is the Vera BlueField-4 STX, an agentic AI storage processor with in-silicon security designed to handle the unique data and trust requirements of agent workloads.
To help operators actually build these facilities, NVIDIA introduced DSX, described as a playbook for AI factory infrastructure. The blueprint covers reference architectures, networking, power, and cooling guidance for hyperscalers and sovereign operators racing to bring AI capacity online.
Semiconductor manufacturing also featured prominently. NVIDIA and TSMC announced an expanded collaboration to embed AI directly into fabs, applying generative and agentic models to chip design and manufacturing workflows to accelerate yield, defect detection, and process optimisation.
Physical AI was another major theme. NVIDIA launched Cosmos 3, the latest version of its open frontier foundation model for physical AI, designed to help robots and autonomous systems reason about the physical world before acting. A companion release brought a major collection of open-source agent tools and skills aimed specifically at physical AI developers.
Robotics took a further step with the Isaac GR00T reference humanoid robot, an open design targeted at academic researchers. Pairing the reference platform with Cosmos 3, NVIDIA hopes to accelerate humanoid robotics research by giving universities a standardised hardware and software stack.
In autonomous mobility, the company positioned DRIVE Hyperion as the global platform for a robotaxi-ready world, alongside Alpamayo 2, a new open reasoning model aimed at robotaxi developers. Together, the two are intended to give automakers and mobility operators a faster path to commercial autonomous fleets.
NVIDIA also moved deeper into enterprise software. Major enterprise software leaders are now building AI agents on NVIDIA’s stack, with the company outlining how its agent frameworks, microservices, and inference infrastructure are being adopted across business applications.
Healthcare in Taiwan emerged as a flagship use case. NVIDIA, Foxconn, and a group of Taiwanese medical centres unveiled the Healthy Taiwan initiative, which brings together agentic and physical AI to support hospitals, clinical workflows, and patient care across the island.
Beyond product launches, NVIDIA used the keynote to highlight the global expansion of its AI cloud ecosystem, with new and existing partners adding capacity to meet what the company described as runaway demand for AI compute. Taiwan’s own industry titans, spanning silicon, systems, and manufacturing, were positioned as central engines of the worldwide AI infrastructure buildout.
Factory operations received a dedicated blueprint of their own. NVIDIA introduced what it calls a new AI brain for factories, an operations blueprint combining simulation, perception, and agentic AI to optimise production lines, predictive maintenance, and supply chain decisions in real time.
On the developer side, the company detailed how it is levelling up local AI agents across RTX PCs and DGX Spark, with new performance gains for DGX Spark including roughly twofold acceleration for local AI agents and a simplified setup experience via NVIDIA NemoClaw. Developers around the world are also powering up DGX Station as a desk-side workbench for building agentic and physical AI applications.
Taken together, the announcements paint a picture of an AI stack stretching from individual PCs and workstations through data-centre-scale factories and out to robots, vehicles, and industrial systems. Huang’s message was that the agentic era is no longer a future scenario but an active buildout, with NVIDIA positioning itself as the connective tissue across cloud, enterprise, edge, and the physical world.
