Nvidia CEO Jensen Huang sees $1T AI chip demand by 2027 at GTC

Nvidia expects purchase orders for its next-generation artificial intelligence chips to reach $1 trillion through 2027, underscoring the massive computing demand created by the rapid expansion of AI technologies.

Speaking at Nvidia’s annual GTC developer conference in San Jose, California, CEO Jensen Huang said demand for the company’s upcoming Blackwell and Vera Rubin chip systems is accelerating as startups and large technology firms scale up their AI infrastructure.

“If they could just get more capacity, they could generate more tokens, their revenues would go up,” Huang said during his keynote presentation.

Nvidia shares closed 1.63% above on Monday following the announcements.

The company previously estimated a $500 billion revenue opportunity from the two chip architectures, but Nvidia executives now believe demand will exceed those earlier projections.

Chief Financial Officer Colette Kress said last month that the company expects growth this year to surpass the earlier estimate.

Nvidia forecasts massive AI infrastructure demand

Huang’s comments highlight the enormous demand for computing power driven by artificial intelligence.

Nvidia’s graphics processing units (GPUs) have become the backbone of modern AI systems, helping power everything from large language models to advanced autonomous software.

As the AI industry evolves from chatbot-style applications to agentic systems capable of performing complex tasks through multiple software agents, the computational requirements are expanding rapidly.

“The inference inflection has arrived,” Huang said at the conference.

He also noted that computing demand has surged dramatically.

According to Huang, demand for AI computing has increased one million times over the past two years.

This surge is driving an unprecedented buildout of AI infrastructure, including data centers equipped with Nvidia’s high-performance chips.

The company said earlier this year that its quarterly revenue is expected to surge about 77% year over year to roughly $78 billion, extending a remarkable streak of rapid growth.

Nvidia has now reported 11 consecutive quarters with revenue growth above 55%.

New chips and AI systems unveiled at GTC

Alongside the demand projections, Nvidia unveiled several new technologies designed to support the next phase of AI development.

One of the highlights was Vera, a new CPU designed specifically for agentic artificial intelligence workloads.

Nvidia said the processor is twice as efficient and 50% faster than traditional rack-scale CPUs.

The company also introduced a Vera CPU rack, which integrates 256 liquid-cooled Vera CPUs and can support more than 22,500 concurrent CPU environments.

Several major hyperscalers are already collaborating with Nvidia on the system.

“Vera is arriving at a turning point for AI. As intelligence becomes agentic — capable of reasoning and acting — the importance of the systems orchestrating that work is elevated,” Huang said.

“With breakthrough performance and energy efficiency, Vera unlocks AI systems that think faster and scale further.”

The company also unveiled the Groq 3 Language Processing Unit (LPU), the first chip produced following Nvidia’s $20 billion asset purchase of startup Groq in December.

The chip is designed to enhance AI processing by improving memory capacity and accelerating GPU workloads.

Nvidia plans to ship the Groq 3 LPU in the third quarter.

Next-generation AI infrastructure takes shape

Nvidia is also preparing to launch its Vera Rubin rack-scale system later this year, which the company says delivers ten times more performance per watt compared with the previous Grace Blackwell system.

Energy efficiency has become a major concern as AI infrastructure expands globally, with data centers requiring enormous amounts of electricity.

To further improve performance, Nvidia introduced a Groq LPX rack capable of housing 256 LPUs designed to operate alongside the Vera Rubin system.

Huang said the new rack configuration can increase the tokens-per-watt performance of Rubin GPUs by 35 times.

“We united, unified two processors of extreme differences, one for high throughput, one for low latency. It still doesn’t change the fact that we need a lot of memory,” Huang said.

“And so we’re just going to add a whole bunch of Groq chips, which expands the amount of memory it has.”

Looking ahead, Nvidia also previewed Kyber, a prototype architecture that will form the foundation of its next-generation rack-scale computing systems.

The design integrates 144 GPUs arranged vertically in compute trays to increase density and reduce latency.

Kyber will be incorporated into Vera Rubin Ultra, Nvidia’s next major AI infrastructure system expected to ship in 2027.

The announcements reinforce Nvidia’s position at the center of the global AI infrastructure race as companies compete to deploy increasingly powerful computing systems to support advanced artificial intelligence applications.