NVIDIA and OpenAI have signed a letter of intent for a partnership that puts a staggering number on the table: at least 10 gigawatts of AI computing capacity, representing millions of GPUs, all dedicated to training and running OpenAI’s models. To make it happen, NVIDIA plans to invest up to $100 billion in OpenAI, tied to deployment milestones. The first gigawatt comes online in the second half of 2026, built on NVIDIA’s upcoming Vera Rubin platform. That’s a tight timeline for a project of this scale, and it signals just how urgently both companies view the race toward what they call superintelligence.
Jensen Huang framed the deal as the natural next step after a decade of mutual dependence — from the first DGX supercomputer to the explosion of ChatGPT. Sam Altman went even further, calling compute infrastructure “the basis for the economy of the future.” The language here isn’t about incremental gains. It’s about building the physical substrate for a post-AGI world. Greg Brockman reinforced that ambition, pointing to the hundreds of millions of daily users already on OpenAI’s platform and the desire to “push back the frontier of intelligence” with this new horsepower.
The partnership also formalizes a co-optimization pact. OpenAI will treat NVIDIA as its preferred compute and networking partner, and the two companies will align their hardware and software roadmaps. That’s a significant technical commitment. It means NVIDIA’s chip designs and OpenAI’s model architectures will evolve in tandem, potentially squeezing out efficiency gains that a generic cloud deployment would miss. The deal doesn’t replace OpenAI’s existing relationships with Microsoft, Oracle, or the Stargate consortium — it layers on top of them.
Here’s what I’m watching: $100 billion in staged investment is not guaranteed cash. It’s contingent on actual gigawatts deployed, which means physical construction, power purchase agreements, and regulatory green lights. The first Vera Rubin systems in 2026 will be a test of whether this timeline holds. If it slips, the financial scaffolding shifts. And while 700 million weekly active users gives OpenAI leverage, the underlying question is whether anyone — even a partnership of this size — can actually build 10 gigawatts of AI-specific datacenter capacity without hitting a wall on energy, chips, or both.
