OpenAI And NVIDIA Launch $100B Partnership To Power AI Superintelligence

OpenAI and NVIDIA have signed a letter of intent for what may just be the largest single investment in AI compute infrastructure to date: a plan to deploy at least 10 gigawatts of NVIDIA systems dedicated to training and running OpenAI's future models. The partnership represents not just money, but a roadmap that ties OpenAI's quest for artificial general intelligence—and eventually "superintelligence"—directly to NVIDIA's hardware cadence.

Under the deal, NVIDIA will invest up to $100 billion into OpenAI, phased in alongside each gigawatt of compute that comes online. The first wave is scheduled for the second half of 2026, running on the Vera Rubin platform, NVIDIA's successor to the Blackwell line of GPUs. Rubin systems, particularly the NVL144 CPX rack configuration, are designed for massive-context inference workloads, with specialized GPUs for long-sequence reasoning.

For OpenAI, the partnership guarantees early access to Rubin, and potentially, a head start on competitors that are still filling datacenters with Blackwell GB200 and the brand-new Blackwell Ultra GB300 systems. GB200 NVL72 racks are reportedly rated for about 192 kW Electrical Design Power (EDP); scaling to 10 gigawatts could mean anywhere from ~27,000 Rubin racks at the upper end to as many as ~52,000 if NVIDIA leans on more modest Rubin configurations.


The numbers are staggering in their own right. Even at the conservative 27,000-rack estimate, this single OpenAI deployment would represent millions of GPUs and exascale-class AI compute capacity. This marks not only the scale of OpenAI's ambition but also NVIDIA's intent to keep its grip on the AI infrastructure market by embedding itself at the heart of the most prominent AI lab.

The deal is the latest in a string of outsized investments from NVIDIA. Just three days ago, the company committed £2 billion (about $2.6 billion US) to boost the UK's AI startup ecosystem, and at the same time, it pledged $5 billion into Intel Corp in a strategic partnership we've already written much about. Adding a potential $100 billion tie-up with OpenAI cements NVIDIA's dual strategy: push its datacenter platforms forward with Blackwell and Rubin, while also anchoring demand by bankrolling its biggest customers.
"Compute infrastructure will be the basis for the economy of the future," OpenAI CEO Sam Altman said in today's announcement. NVIDIA's Jensen Huang was blunter: this is about building "the next era of intelligence."

If nothing else, the partnership underscores how quickly the power requirements of frontier AI are accelerating. Ten gigawatts is an unbelievable amount of power that rivals the output of some national grids—a single nuclear reactor only puts out about one gigawatt. NVIDIA and OpenAI are betting that this scale of compute will be the foundation of not just AI models, but of a new technological economy.