At NVIDIA GTC Taipei at COMPUTEX, the world’s developers, researchers and industry leaders converged to dive into the breakthroughs shaping every industry — from AI factories and scaling infrastructure to agentic and physical AI.
The Keynote
“AI is now a profit generator. AI is now a GDP generator,” NVIDIA founder and CEO Jensen Huang told the audience at the Taipei Music Center, gathered in person and across 70 watch parties throughout Taiwan. Huang argued that useful AI has arrived, with agents becoming the computing pattern of the next decade.
His framing: tokens are now profitable units of revenue, and customers don’t want to buy computers — they want to build AI factories. “The more you buy, the more you make,” Huang said.
Vera Rubin in Full Production
Huang announced that NVIDIA Vera Rubin is ramping into full production. The five-rack platform — Vera Rubin NVL72 systems, Vera CPU, Groq 3 LPX, Spectrum-6 SPX Ethernet and Vera BlueField-4 STX storage — is being ramped by hundreds of partners across 350+ factories in 30 countries, including 150 in Taiwan.
The supply chain created for Vera Rubin is twice the size of Grace Blackwell. Also in production: Spectrum-X Ethernet Photonics, the first 200Gb/s SerDes Ethernet switch with co-packaged optics, adopted early by CoreWeave, Lambda and Oracle Cloud Infrastructure, delivering 5x better power efficiency than transceiver networks.
Nemotron 3 Ultra
NVIDIA released Nemotron 3 Ultra, a 550-billion-parameter mixture-of-experts open model built for long-running agents. It delivers up to 5x faster inference and lowers the cost of complex agentic tasks by up to 30 percent. Early adopters include Perplexity, Palantir, ServiceNow, Aible, Glean, Greptile and Harvey, with support across agent harnesses including Cline, OpenHands, LangChain Deep Agents and OpenCode.
- Open weights, datasets and recipes, deployable anywhere models and data reside
- New Nemotron Speech streaming ASR covering 40 language locales
- Nemotron 3.5 Content Safety, a 4-billion-parameter multimodal model covering 23 safety categories
The AI Factory Ecosystem
Huang positioned AI factories as the largest infrastructure build-out in human history. NVIDIA expanded MGX, the open modular reference architecture, with third-generation rack designs for Vera Rubin and 800 VDC power infrastructure upgrades. New reference architectures for Secure Agent Workspaces and confidential computing target runtime-enforced guardrails for autonomous agents.
Agents, Robotics and Media
Updates rippled across the stack: Isaac GR00T accelerated humanoid robot development from data to deployment, with GR00T models past 274,000 downloads. The Build-a-Claw experience brought secure long-running agent development to APAC developers using NemoClaw blueprints and the OpenShell runtime. And NVIDIA AI for Media added real-time LipSync, Active Speaker Detection, RTX Video Super Resolution and a Synthetic Video Detector for live and post-production pipelines.
What This Means
GTC Taipei at COMPUTEX set the agenda for the rest of 2026: Vera Rubin in full production, open Nemotron models powering the agent economy, and a hardware-plus-runtime stack built for AI factories. The agentic era now has the full-stack infrastructure to back it up.