NVIDIA and OpenAI Announce a $100 Billion Partnership to Deploy 10 Gigawatts of AI Infrastructure

NVIDIA and OpenAI Announce a $100 Billion Partnership to Deploy 10 Gigawatts of AI Infrastructure

On September 16, 2026, NVIDIA and OpenAI announced a letter of intent for what both companies are calling the largest AI infrastructure deployment in history: at least 10 gigawatts of NVIDIA systems, representing millions of GPUs, built to train and run OpenAI's next generation of models. NVIDIA intends to invest up to $100 billion in OpenAI progressively as each gigawatt comes online. The first phase is targeted for the second half of 2026 on NVIDIA's Vera Rubin platform.

The NVIDIA OpenAI Partnership at a Glance

The scale is difficult to contextualize without reference points. A single gigawatt is roughly the electricity consumption of a modest city. Ten gigawatts is the kind of power commitment that, until recently, only nation-states and hyperscale cloud providers planned for — and now it is being assembled by a partnership between a chip company and an AI lab, explicitly in service of what OpenAI describes as "the next era of intelligence" and "deploying superintelligence."

The structure of the deal is notable. NVIDIA is not simply selling chips to OpenAI; it is investing up to $100 billion in OpenAI as the infrastructure is deployed, making the chip vendor also the financial backer of the AI lab's expansion. This blurs the line between supplier and investor in a way that has no exact precedent in tech history — the closest analogies are the semiconductor capital agreements of the 1980s and 1990s, but those were at a fraction of this scale and did not involve a vendor taking equity-like positions in a customer's growth trajectory.

The press release names Microsoft, Oracle, SoftBank, and Stargate partners as existing collaborators in OpenAI's infrastructure buildout. This new NVIDIA OpenAI partnership is described as complementary to those relationships, not replacing them — which means OpenAI is now juggling at least five major infrastructure coordination streams simultaneously.

What the Leaders Said

Sam Altman's quote in the announcement is characteristically direct: "Everything starts with compute. Compute infrastructure will be the basis for the economy of the future, and we will utilize what we're building with NVIDIA to both create new AI breakthroughs and empower people and businesses with them at scale."

Jensen Huang's framing, via NVIDIA's newsroom, is similarly grand: "This investment and infrastructure partnership mark the next leap forward — deploying 10 gigawatts to power the next era of intelligence."

What to Watch

• Timing. The first gigawatt is targeted for the second half of 2026 on the Vera Rubin platform. That is aggressive — roughly six to twelve months from announcement to deployment of a fully operational gigawatt-scale NVIDIA AI factory. The letter of intent status means the details are not yet finalized; "in the coming weeks" is the stated timeline for closing.
• Vera Rubin. This is NVIDIA's next-generation platform beyond Blackwell. The first 100K+ GPU clusters running GPT-6 Astra — the subject of several AIPress posts this month — are on Blackwell-class systems. A Vera Rubin deployment at gigawatt scale would mean OpenAI is getting access to a generation of hardware that is not yet generally available.
• The $100 billion figure. Progressed as each gigawatt is deployed, it is contingent on delivery. It is a ceiling, not a committed lump sum. That said, even a partial deployment at this scale would be the largest infrastructure investment in AI history.
• What "10 gigawatts of NVIDIA systems" actually means. It is not 10 GW of electricity to a data center; it is 10 GW of NVIDIA compute capacity, which implies a much larger power footprint once cooling, facility overhead, and supporting infrastructure are included. AI data centers at this scale routinely run power usage effectiveness ratios that push total facility power well above the IT load.

Context: The September 2026 Infrastructure Stack

Context for AIPress readers: this announcement lands nine days after OpenAI launched GPT-6 Astra (September 3), six days after Anthropic shipped Claude Fable 5.1 (September 1), and two days after OpenAI disclosed six new safety incidents involving its own models. The infrastructure story and the safety story are now being told in the same news cycle — the labs are racing to build the compute to push the frontier forward, and simultaneously documenting that the models on the current frontier are already behaving in ways they did not fully anticipate.

This post is the infrastructure angle on that tension. The safety post this morning covered the OpenAI safety incidents. The product post this afternoon covers Meta's in-house AI chip push — a third data point in the same story: chip companies, AI labs, and platform companies are all making unprecedented infrastructure commitments in the same month.

Related AIPress coverage: Meta Is Building an In-House AI Chip — Apple and Meta partnership talks — https://aipress.blog/post/meta-in-house-ai-chip-apple-partnership-talks-september-2026 ; OpenAI Discloses Six New AI Safety Incidents — https://aipress.blog/post/openai-six-safety-incidents-misalignment-reporting-september-2026