Photo by Steve Johnson on Unsplash

OpenAI and Nvidia (NVDA) have signed a letter of intent for one of the largest AI infrastructure projects ever attempted. Nvidia plans to invest up to $100 billion to deploy at least 10 gigawatts of compute capacity for OpenAI, starting with a one-gigawatt installation scheduled for the second half of 2026 on its new Vera Rubin platform.

To understand the magnitude, a single large data center today typically consumes 50 to 100 megawatts. A few exceptional facilities approach one gigawatt. OpenAI’s plan is to build capacity equal to 10 nuclear reactors. Nvidia CEO Jensen Huang told CNBC that 10 gigawatts would be enough to power between four million and five million GPUs, roughly equal to the company’s total shipments this year. He called it a “giant project.”

The cost is equally staggering. Huang has told investors that one gigawatt of capacity costs between $50 and $60 billion, with about $35 billion of that going to Nvidia systems. That math suggests a total buildout could exceed half a trillion dollars. OpenAI’s scale-up is meant to serve its reported 700 million weekly active users and to secure the compute foundation for future growth.

Energy is the choke point. The International Energy Agency estimates that global data centers consumed about 1.5 percent of electricity worldwide in 2024 and could use nearly double that by 2030. Tech giants are already locking in nuclear capacity. Microsoft (MSFT) signed a 20-year agreement to restart part of Three Mile Island for 835 megawatts. Amazon Web Services bought a facility next to the Susquehanna nuclear plant and plans to draw nearly a gigawatt. In Wyoming, officials approved an AI campus designed to scale to 10 gigawatts, with its early phase already larger than the electricity use of every home in the state.

Compute and electricity are no longer background resources; they are board-level concerns. If you depend on AI to drive growth, you need to treat power supply and GPU hours as tier-one inputs. Prices will rise, scarcity will matter, and the companies that secure predictable access to both will have an advantage.

More By This Author: