AI Factories, Not Data Centers: NVIDIA's Vera Rubin Platform and the $1 Trillion Bet

Jensen's Trillion-Dollar Thesis

When Jensen Huang walks onstage at GTC, the leather jacket is a given. The ambition is not. At GTC 2026, held March 16-20 at San Jose's SAP Center, Huang didn't just announce products. He laid out an economic thesis: the world needs so much AI computing that it will generate at least $1 trillion in revenue for NVIDIA and its partners over the next two years. That's double what he projected twelve months ago, when the $500 billion figure already seemed aggressive. [1][3]

The logic starts with a simple observation. Every time an AI system thinks, reasons, or takes action, it consumes computing power. That computing power produces tokens. Tokens run on NVIDIA GPUs. And the demand for tokens — from enterprise automation to agentic AI to self-driving cars — isn't growing linearly. It's compounding. "I believe computing demand has increased by 1 million times over the last few years," Huang told the crowd. [1] That's the kind of statement that sounds like marketing until you look at the order books. NVIDIA's inference workloads have exploded as AI moves from training (teaching models) to inference (running them in production). Every chatbot response, every AI agent action, every autonomous vehicle decision is an inference workload. And unlike training, which happens once, inference runs continuously — forever. Huang had a phrase for this that kept coming back throughout the two-hour keynote: "Tokens are the new commodity." He even predicted that every engineer in the future will receive a yearly token budget alongside their salary — compute as compensation. [2]