OpenAI and Nvidia are preparing for an AI infrastructure buildout that puts electricity, not just chips, at the center of the story. The companies announced a letter of intent for a strategic partnership to deploy at least 10 gigawatts of Nvidia systems for OpenAI, with Nvidia planning to invest up to $100 billion as the systems are deployed.
The first gigawatt is expected to come online in the second half of 2026 using Nvidia’s Vera Rubin platform. The companies said they expect to finalize details in the coming weeks.
A compute deal at unusual scale
The agreement would make Nvidia OpenAI’s preferred strategic compute and networking partner. It also sits alongside OpenAI’s existing relationships with Microsoft, Oracle, SoftBank, and the recently announced Stargate project partners.
OpenAI CEO Sam Altman framed the partnership around the central role of infrastructure in the next phase of AI. “Everything starts with compute,” Altman said in the announcement. “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.”
The headline number is 10 gigawatts. Nvidia CEO Jensen Huang told CNBC that this amount of power corresponds to between 4 million and 5 million graphics processing units. He said that figure matches Nvidia’s total GPU shipments for this year and is double last year’s volume.
Huang described the effort plainly: “This is a giant project.” That scale is why the deal is about more than a supplier relationship. It is a test of whether AI companies, chipmakers, utilities and data center developers can move in sync at a level far beyond most current facilities.
Why 10 gigawatts changes the conversation
The source article compares 10 gigawatts with the output of roughly 10 nuclear reactors, because a nuclear facility typically produces about 1 gigawatt. It also notes that current data center energy consumption ranges from 10 megawatts to 1 gigawatt, with most large facilities using between 50 and 100 megawatts.
That comparison helps explain the practical challenge. A conventional large data center may already be an important local power user. A 10-gigawatt AI infrastructure plan would move the discussion into the territory of major power generation and grid planning.
The source also says OpenAI’s planned infrastructure would require as much electricity as multiple major cities. That does not mean the companies have disclosed where the electricity will come from. They have not specified power sources in the announcement.
Still, the energy requirement is large enough that power supply becomes a core part of the AI roadmap. For AI infrastructure, chips and networking equipment only matter if the sites can be built, connected and supplied with electricity.
The investment structure draws attention
The Nvidia investment plan is up to $100 billion as systems roll out. After the announcement, Nvidia’s stock rose nearly 4 percent on Monday, adding roughly $170 billion to its market capitalization.
The arrangement also drew scrutiny because the money and equipment flow are closely linked. Bryn Talkington, managing partner at Requisite Capital Management, described the circular nature of the plan to CNBC: “Nvidia invests $100 billion in OpenAI, which then OpenAI turns back and gives it back to Nvidia,” Talkington told CNBC. “I feel like this is going to be very virtuous for Jensen.”
The announcement comes after rapid growth for OpenAI, which the source says has reached 700 million weekly active users. It also follows Nvidia’s separate disclosure, a week earlier, of a $5 billion investment in Intel. Nvidia took a 4 percent stake in Intel as the companies plan to co-develop custom data center and PC products.
Huang has previously given investors a sense of what this kind of capacity can cost. In an August earnings call, he said one gigawatt of data center capacity costs between $50 billion and $60 billion, with about $35 billion going toward Nvidia chips and systems. At that rate, the 10 gigawatt project could require total investment exceeding $500 billion.
Nuclear power and grid limits loom over AI data centers
Although OpenAI and Nvidia did not name power sources for this plan, the source places it within a broader industry move toward major energy arrangements. Microsoft signed a 20-year agreement in September 2024 to restart a Three Mile Island reactor for 835 megawatts. In May of this year, Amazon Web Services purchased a data center next to Pennsylvania’s Susquehanna nuclear plant with plans to use up to 960 megawatts.
Other AI infrastructure projects are also appearing across the US. In July, officials in Cheyenne, Wyoming, announced plans for an AI data center that would eventually scale to 10 gigawatts. The source says the project would consume more electricity than all homes in the state combined, even in its earliest 1.8 gigawatt phase. Whether it is connected to OpenAI’s plans remains unclear.
Altman’s interest in very large data center deals has been discussed for more than a year. In September of last year, Constellation Energy CEO Joe Dominguez told Bloomberg he had heard Altman wanted five to seven data centers of 5 gigawatts each. Alex de Vries of Digiconomist told Fortune that seven 5-gigawatt units would have “twice the power consumption of New York State combined.”
The International Energy Agency estimates that global data centers already consumed roughly 1.5 percent of global electricity in 2024. It also says global data center electricity demand could reach 945 terawatt hours by 2030. Against that backdrop, the OpenAI and Nvidia plan shows why AI growth is increasingly tied to questions of power generation, transmission and environmental impact.
What has to happen next
The companies still need to finalize details. Huang told CNBC that the $100 billion investment is in addition to Nvidia’s existing commitments and was not included in the company’s recent financial forecasts to investors.
The next questions are practical. Where will the systems be deployed? How will grid connections be secured in power-constrained markets? Which energy sources will support a project of this size?
The source makes clear that existing power grid connections are already bottlenecks in some markets, while utilities are struggling to keep pace with rapid AI expansion. For OpenAI and Nvidia, the challenge is no longer only whether enough GPUs can be assembled. It is whether the physical infrastructure around those GPUs can be built at the same pace.