Nvidia CEO sees US$1 trillion AI demand by 2027

San Jose/Taipei, March 16 (CNA) Nvidia CEO Jensen Huang (黃仁勳) predicted Monday that demand for AI infrastructure could reach at least US$1 trillion by 2027 amid a shift in how computing is done.

"The inference inflection point has arrived," Huang said at Nvidia's annual GTC 2026 in San Jose, California, adding that the shift toward inference-driven computing is pushing demand for AI infrastructure beyond US$1 trillion in the coming years.

The projection more than doubles Huang's earlier estimate that demand for Nvidia's Blackwell and Vera Rubin systems could reach about US$500 billion by 2026.

"Computing used to be retrieval-based. Now it's generative," he said, describing a fundamental transformation in how machines process information.

He reiterated that Moore's Law "has run out of steam" and predicted that "every single software company of the future will be agentic, and they will be token manufacturers."

Huang expected Taiwan to be a key part of the supply chain delivering Nvidia's Vera Rubin architecture.

Three slides in his presentation showed that of the more than 60 global partners for Nvidia's Vera Rubin architecture, many were Taiwanese companies, such as Asustek Computer Inc., Foxconn, Quanta Cloud Technology, Wistron Corp., and Wiwynn Corp.

Huang also unveiled further details of Nvidia's next-generation AI systems, including its Vera Rubin platform, designed for agentic AI workloads.

The system features 100 percent liquid cooling and significantly reduced deployment time, which he said would cut installation from two days to two hours.

Nvidia is also working with AI chip startup Groq Inc. to optimize inference performance, with Groq chips to be manufactured by Samsung Electronics Co., Huang said.

Looking further ahead, Huang introduced the company's Feynman architecture as the next step in its roadmap, featuring new processing and networking technologies, including co-packaged optics.

On the software front, Huang highlighted the rise of "agentic AI," describing a shift in which software systems autonomously perform tasks and generate outputs.

He pointed to the rapid emergence of OpenClaw, an open-source agentic platform, and announced Nvidia's enterprise-focused NemoClaw system, developed with enhanced security and privacy capabilities for corporate use.

Huang also outlined Nvidia's ambitions beyond planet Earth, revealing plans to develop space-based data centers powered by its Vera Rubin Space-1 system.

Such efforts face significant technical challenges due to the absence of conventional cooling methods in space, with no conduction or convection and only radiation available for heat dissipation.

"We have to figure out how to cool these systems out in space, but we've got lots of great engineers working on it," he said.