Nvidia and OpenAI announced a landmark letter of intent to build massive AI infrastructure together. Nvidia said it intends to invest up to $100B progressively as new systems are installed. The companies plan to deploy at least ten gigawatts of Nvidia systems to power OpenAI model training and serving.
Deal details now
The agreement is a letter of intent and not a final contract. Nvidia will scale its investment as each gigawatt of infrastructure is delivered. The first gigawatt is expected on Nvidia’s Vera Rubin platform in the second half of 2026. Reuters reports that the initial tranche could total ten billion dollars once the final terms are set.
OpenAI will name Nvidia as a preferred strategic compute and networking partner for its AI factory expansion. The wording in the announcements says Nvidia systems will represent millions of GPUs across the new data centers. The move is meant to let OpenAI run and train larger models with tighter hardware and network links.
Nvidia and OpenAI said this partnership will complement OpenAI’s existing arrangements with other providers and partners. OpenAI will continue to work with firms such as Microsoft, Oracle, and SoftBank. The companies characterized the deal as broad and cooperative rather than exclusive.
Why it matters
10 Gigawatts is a huge power consumption. Reporters write that it is approximately the capacity of several large power plants and enough to service millions of houses. That scale indicates the costliness of training and running frontier AI models on a large scale.
The partnership reduces part of OpenAI’s single supplier risk while also deepening Nvidia’s role in the AI supply chain. Microsoft remains a major partner of OpenAI, but OpenAI has been building additional infrastructure with other vendors since changes to its cloud arrangements earlier this year. The Nvidia deal gives OpenAI another path to secure the compute it needs.
Analysts and market watchers gave mixed reactions. Some see the move as a pragmatic way to scale compute needs quickly. Others warn that Nvidia is taking on large financial and operational risk by tying so much capital to one partner. Regulators and investors will watch execution and any effects on competition.
What comes next
The work will be a multi-product cycle in both companies. In the long run, Nvidia will launch its Vera Rubin and other platforms to the data center partners and to OpenAI locations. The new systems will be used to train and serve next-generation models with higher compute and reduced latency among chips, which OpenAI will use. Information regarding the way the investment will be structured was not fully made available. It is not made clear whether the support would be in the form of cash chip credits or any other means of support.
The deal can accelerate the emergence of closely integrated AI systems by customers and cloud operators alike. To the wider industry, it is an indication that the manufacturing of chips can also be a huge investor in the companies that consume their hardware. The combination of that might change the location and method of the construction of large AI factories and ownership of the stacks that operate the largest ones.
