Nvidia CEO Jensen Huang estimates $100B cost for 1 GW AI factory

Jensen Huang just put a number on the future of AI infrastructure, and it’s the kind of figure that makes even Big Tech CFOs pause. The Nvidia CEO estimates that a single 1 gigawatt AI factory, the massive data center complexes needed to train and run frontier AI models, could cost up to $100 billion.

That number is up from current estimates of around $55 billion per gigawatt. And here’s the part that should matter to anyone watching capital flows in tech: roughly half of that spend goes directly to Nvidia’s chips and systems.

The math behind the megawatt madness

Earlier estimates from mid-2025 pegged the cost of a 1 GW AI factory at $50 to $60 billion, with about $35 billion earmarked for Nvidia hardware. Huang’s updated projection suggests costs are trending upward as the complexity of AI workloads increases and the demand for more sophisticated silicon accelerates.

Reports from June 2026 noted that a fully operational 1 GW factory could generate $300 to $400 billion in annual intelligence value. That implies a payback period of just 2 to 3 years at full utilization.

Nvidia’s OpenAI partnership sets the scale

Nvidia and OpenAI announced a partnership on September 22, 2025, focused on deploying 10 GW worth of Nvidia systems. If each gigawatt eventually hits the $100 billion mark, we’re looking at a trillion-dollar infrastructure buildout between just two companies.

Nvidia committed to investing up to $100 billion over time as each gigawatt becomes operational. The first GW is targeted for deployment in the second half of 2026, running on Nvidia’s Vera Rubin platform, the successor architecture to Blackwell.

What this means for investors

Capturing roughly half the capex of every AI factory built means Nvidia’s revenue ceiling keeps rising as long as the buildout continues. The company’s market share in AI accelerators remains dominant, and architectures like Vera Rubin are designed to maintain that lead through the next generation of scaling.

The sheer magnitude of capital required creates concentration risk across the tech sector. A $100 billion factory doesn’t get funded by mid-cap companies. This is a game for Microsoft, Google, Amazon, and sovereign wealth funds.

The 2 to 3 year payback claim assumes full utilization of these facilities, which in turn assumes that demand for AI inference and training continues growing at its current trajectory. If enterprise adoption of AI plateaus, or if a major model architecture shift reduces compute requirements, the ROI math changes dramatically.