At Nvidia's GTC conference in San Jose, CEO Jensen Huang told an audience of thousands that the company has secured purchase orders for its Blackwell and upcoming Vera Rubin chip platforms expected to total $1 trillion through 2027. This figure doubles the revenue opportunity Nvidia projected just last year.

Demand, Huang noted, shows no sign of slowing. "If they could just get more capacity, they could generate more tokens, their revenues would go up," he said, referring to the broad spectrum of companies driving the surge. Nvidia's stock climbed about 2% following the announcement.

The company's rise to a $4.5 trillion valuation is fueled by its AI processors. As AI models evolve from simple chatbots to complex, multi-agent systems, the computational load—measured in tokens—has skyrocketed, demanding ever-faster and more efficient hardware.

Huang provided new details on the Vera Rubin platform, slated for release later this year. The company claims it will deliver a tenfold improvement in performance per watt over its predecessor, a vital gain as energy use becomes a primary constraint for AI expansion.

In a significant move, Huang also introduced the Groq 3 LPU, Nvidia's first chip stemming from its $20 billion acquisition of the startup Groq last December. A full rack of 256 Groq accelerators, designed to work alongside Vera Rubin systems, can boost token efficiency for Rubin GPUs by a factor of 35, Huang said.

Looking further ahead, Huang displayed a prototype of the Kyber architecture. This design, intended for the 2027 Vera Rubin Ultra system, stacks 144 GPUs vertically to increase density and reduce lag. The presentation underscored Nvidia's strategy: to meet trillion-dollar demand, it is pushing innovation on every front of chip and system design.

Source: CNBC