OpenAI and NVIDIA plan 10 gigawatts for AI

OpenAI and NVIDIA have announced a partnership to deploy at least 10 gigawatts of artificial intelligence infrastructure, a scale that would involve millions of chips and new data centers to train and run OpenAI’s next models.

The agreement is still a letter of intent, not a final contract. Even so, it sets a clear direction: OpenAI wants to secure massive computing capacity, and NVIDIA will be its preferred partner for chips, networking and data center systems.

The first phase is planned for the second half of 2026 and will use NVIDIA’s Vera Rubin platform, a generation of systems designed for AI workloads.

Up to $100 billion in investment

As part of the partnership, NVIDIA plans to invest up to $100 billion in OpenAI. The investment would be made progressively as each gigawatt of capacity is deployed.

A gigawatt measures power, not the number of chips. To put the figure into context, 10 gigawatts represents energy demand comparable to that of a large urban region. In this case, that energy would power data centers filled with AI accelerators, along with the cooling, storage and networking systems they require.

The goal is for OpenAI to train larger models and make them available to more people, businesses and developers. The company says it already has more than 700 million weekly active users.

Why the partnership matters

Training and running AI models requires increasing amounts of computing power. Computing power is the ability of chips to process operations, generate responses and learn from large volumes of data. Without enough infrastructure, even a capable model can take too long to respond or become too expensive to use.

The partnership will allow OpenAI and NVIDIA to coordinate their roadmaps. OpenAI will adapt its software and models to NVIDIA’s systems, while NVIDIA will be able to design hardware and tools around OpenAI’s specific needs.

For you, this does not mean ChatGPT will automatically change tomorrow. If the plan moves forward, the most direct effect would be greater capacity to:

• handle more queries without congestion;
• offer more advanced models and more complex features;
• reduce response times on certain services;
• expand the use of AI in businesses and applications.

It also brings costs and challenges. Building these data centers requires electricity, land, networks and cooling systems. Energy availability could become one of the main limits on AI’s growth, alongside chip prices and manufacturing capacity.

OpenAI is already working on infrastructure with Microsoft, Oracle, SoftBank and other Stargate partners. The collaboration with NVIDIA adds to that network rather than replacing it.

The important thing now is to distinguish the announcement from its execution: the 10 gigawatts and the investment of up to $100 billion remain targets subject to the final details of the agreement. The first concrete sign will be whether the initial Vera Rubin phase begins operating in the second half of 2026 and whether deployment continues afterward at the announced scale.