Chip giant NVIDIA and AI giant OpenAI simultaneously announced a major strategic partnership late on September 23rd, declaring they have signed a strategic partnership agreement to deploy at least 10GW of NVIDIA systems over the next few years, with NVIDIA investing up to $100 billion in phased investments. This unprecedented investment in AI infrastructure marks the official escalation of the AI arms race into a global race to build "super-giant AI factories."
10GW of computing power: AI officially enters the gigawatt era
According to a press release from both parties, the massive 10GW facility can accommodate millions of GPUs operating simultaneously, far exceeding the maximum capacity of traditional cloud data centers. The announcement states that the first facilities are expected to be online in the second half of 2026 and will directly support OpenAI's next-generation models. The goal of both parties is clear: to advance towards superintelligence. OpenAI CEO Sam Altman commented:
"Computing infrastructure will be the foundation of the future economy."
NVIDIA invests $100 billion: from selling chips to building AI infrastructure
The announcement went on to state that NVIDIA will invest $100 billion at a pace of one GW at a time. This collaboration demonstrates that NVIDIA is moving beyond its role as a hardware supplier and is collaborating with OpenAI to form a capital and technology partnership, working together to optimize models, software, and hardware ecosystems.
OpenAI estimates that training costs alone could reach $19 billion by 2026, and could climb to hundreds of billions by 2030. However, this massive investment will also yield substantial returns: NVIDIA's funding and hardware have effectively built a dedicated AI cloud specifically for OpenAI.
Vera Rubin Platform: The Heart of the Next-Generation AI Factory
In terms of technology, the overall deployment of this system will adopt NVIDIA's Vera Rubin platform. This architecture combines Rubin GPUs, Vera CPUs, HBM4 high-bandwidth memory, and NVLink 6 networking, and will achieve FP4 performance of up to 50 PFLOPS per node.
NVIDIA likens these data centers to "AI factories," capable of scaling from hundreds of megawatts to gigawatts. Equipped with advanced power and cooling systems, they will provide long-term computing power for training and inference of extremely large models. NVIDIA CEO Jensen Huang also stated that this collaboration will "drive the next major leap forward in AI."
Global Layout: Stargate UK Becomes a Touchstone
It's worth noting that the collaboration has also expanded beyond the United States. Public information indicates that OpenAI and NVIDIA are advancing the Stargate UK project in the United Kingdom, with the goal of deploying 10,000 Blackwell GPUs there by 2026, with an investment of approximately $3 billion.
By collaborating with local partners such as Nscale, the two companies hope to establish a "sovereign computing power" network at the regional level, reducing dependence on a single cloud giant and transnational data paths, and further strengthening the security of the AI industry chain.
Nvidia shares rose 3.9%
Encouraged by this news, Nvidia's stock price rose in U.S. trading, reaching $183 and its market value rising to $4.45 trillion.
