TAIPEI (TVBS News) — Nvidia announced Monday (June 1) that its next-generation AI infrastructure platform Vera Rubin has entered full production. CEO Jensen Huang (黃仁勳) called it "the most ambitious endeavor in the history of our company" and confirmed the company has "orders in hand." Taiwan is central to the project: more than 1 million rack components for Vera Rubin are assembled across 25 factory sites on the island, according to Nvidia.
Huang made the announcement during his keynote at Nvidia GTC Taipei 2026. He said the supply chain built for Vera Rubin is twice the size of the previous Grace Blackwell generation, with assembly time for a single rack reduced from two hours to five minutes. The improvement came from a redesigned architecture that uses a central printed circuit board to connect components, replacing cables, tubing and fans, Nvidia said.
Huang praised Taiwan's role in the production ramp-up, drawing cheers from the crowd. "Taiwan's rich ecosystem, the richest ecosystem, the world's best supply chain ecosystem," he said. More than 150 Taiwan suppliers participated in the development, including Taiwan Semiconductor Manufacturing Co. (台積電), Hon Hai Precision Industry (鴻海) and Quanta Computer (廣達), according to Nvidia.
Vera Rubin is not a single chip but a multi-rack supercomputer system comprising seven core chip types. The platform includes the Rubin GPU for AI reasoning, the Vera CPU designed specifically for AI agents, BlueField-4 data processing units for security, and high-speed networking components. The system is manufactured using TSMC's 3-nanometer process with advanced packaging technology.
The announcement comes as AI infrastructure costs continue to climb. Huang said investment in 1-gigawatt AI factories has risen from US$20-30 billion (approximately NT$627-940 billion) to US$50-60 billion (approximately NT$1.57-1.88 trillion). Future investments could reach US$80-100 billion (approximately NT$2.51-3.13 trillion), he added.
"Compute is revenue now," Huang said, describing how AI tokens have become profitable units of output. He argued that throughput per watt and system reliability are now the key metrics for AI infrastructure competition. "When you put US$100 billion into an AI factory, it must work the first time," he said.
Huang positioned Vera Rubin as purpose-built for the emerging "agentic AI" era. Unlike earlier systems designed for training large language models, Vera Rubin is optimized for AI agents that must reason, use tools, manage memory and interact with databases in real time, he said.
A complete Vera Rubin NVL72 system contains 18 compute trays, nine NVLink switch trays and liquid cooling infrastructure capable of handling more than 5,000 amperes. The system comprises 1.3 million components, Nvidia said. To accommodate that complexity, Nvidia said millions of square feet of manufacturing capacity have been added to support Grace Blackwell and Vera Rubin production. The company did not disclose specific customer names or order volumes.
Huang's remarks about Taiwan's economic growth aligned with official data. Taiwan's Directorate-General of Budget, Accounting and Statistics (DGBAS), the government's statistics agency, on May 29 forecast GDP growth of 9.64% for 2026 — a 16-year high driven by AI-related exports. "AI demand is strong, and it has also caused Taiwan's GDP and related AI supply chain companies to surge," Huang said.
Huang said Vera Rubin would be "NVIDIA's most successful new product ever" and promised more announcements next year. Specific shipment dates and deployment timelines were not disclosed. ◼ (At time of reporting, US$1 equals approximately NT$31.34)