OpenAI and NVIDIA announced a strategic collaboration to deploy at least 10GW of NVIDIA systems, comprising millions of GPUs, for OpenAI's next-generation AI infrastructure, which will be used for training and running subsequent models. To support this, NVIDIA intends to gradually invest up to $100 billion in OpenAI as computing power becomes available at each stage. The first 1GW cluster is expected to launch on the Vera Rubin platform in the second half of 2026. Both parties claim this will be one of the largest AI infrastructure projects known to date.
The goal of the collaboration is to leverage more efficient computing power and network architecture to support cross-industry generative AI and intelligent agent applications, accelerating development from R&D to production. OpenAI CEO Sam Altman stated that the collaboration will meet the scale and speed requirements of their roadmap. NVIDIA will provide end-to-end capabilities, from chips to systems and software stacks. Data center site selection and detailed commercial arrangements are still underway, and official updates will be provided.
Frequently Asked Questions
Q: What is the core content of this cooperation?
A: Jointly deploy at least 10GW of NVIDIA systems to provide training and inference infrastructure for OpenAI's next-generation models, and provide supporting phased funding.
Q: What is the investment and timeline?
A: NVIDIA plans to invest up to $100 billion, which will be invested in phases according to the implementation progress of each GW; the first 1GW is expected to be put into production in the second half of 2026.
Q: What impact will this have on OpenAI’s existing collaborative ecosystem?
A: While maintaining existing cooperation, we have added NVIDIA as an important partner in computing power and funding to diversify risks and expand production capacity.
Q: Why emphasize “scale and speed”?
A: Very large models and multi-agent systems have extremely high requirements for throughput and connectivity; large-scale clusters can reduce unit training/inference costs and shorten iteration cycles.
Q: Have the location and energy consumption details been published?
A: Not yet fully disclosed. Officials have only provided information on the total scale, milestones, and platforms. Targeted cities and energy support will be announced later.
