NVIDIA and OpenAI Partner on 10GW AI Infrastructure
NVIDIA and OpenAI have announced a partnership to deploy 10 gigawatts of NVIDIA systems for next-generation AI development. The infrastructure is intended to power OpenAI's pursuit of 'superintelligence' and large-scale embodied AI models, signaling that future robotics and autonomy will be built on hyperscale compute.
- The total 10 GW of infrastructure will involve millions of GPUs, with NVIDIA investing up to $100 billion in OpenAI as each gigawatt is deployed. The first 1 GW phase, built on NVIDIA's Vera Rubin platform, is scheduled to come online in the second half of 2026. - To put the 10 GW figure in perspective, global AI data centers were projected to require an additional 10 GW of power capacity in 2025. Some estimates suggest a single gigawatt of AI data center capacity can cost between $35 billion and $60 billion to build. - This massive compute power is aimed at achieving "superintelligence," which OpenAI CEO Sam Altman has suggested could arrive in early forms within a few years, potentially by 2028. Superintelligence refers to AI systems that surpass human cognitive abilities across virtually all domains. - OpenAI is significantly ramping up its robotics division, which it had previously shut down, to focus on embodied AI. The company is hiring researchers with expertise in humanoid robots and teleoperation, suggesting a strategy to ground its AI models in real-world physical interaction. - Job postings for the revived robotics team show a focus on simulation tools like NVIDIA Isaac Sim, which is used for training humanoid systems in virtual environments before real-world deployment. This aligns with NVIDIA's broader strategy of creating an integrated hardware and software ecosystem for robotics development. - For on-device robotics applications, NVIDIA offers the Jetson platform, including the new Jetson Thor computer. This system incorporates a Blackwell GPU to run generative AI models locally at the edge, a crucial capability for autonomous robots that require real-time decision-making without relying on the cloud. - The power required for such large-scale AI is a significant challenge; a 1 GW data center can have an annual electricity bill of around $1.3 billion. To meet these energy demands, the companies are reportedly considering investments in their own on-site clean energy sources, including nuclear reactors.
