A strategic partnership to deploy at least 10 gigawatts of NVIDIA systems for OpenAI’s next‑generation AI infrastructure has been announced by the two companies in San Francisco and Santa Clara on 22 September 2025. The agreement, formalised through a letter of intent, will see NVIDIA invest up to $100 billion in OpenAI as the new hardware is rolled out, with the first phase slated to go live in the second half of 2026 using NVIDIA’s Vera Rubin platform. The collaboration follows a decade of joint development, from the original DGX supercomputer to the launch of ChatGPT, and aims to provide the compute capacity required to train and run increasingly sophisticated models on the path to artificial general intelligence.
NVIDIA Invests Up to $100 Billion to Power OpenAI’s Next‑Generation Models. NVIDIA’s pledge of up to $100 billion to OpenAI, announced from San Francisco and Santa Clara on 22 September 2025, will fund the deployment of at least 10 gigawatts of NVIDIA‑powered compute. The scale of the commitment eclipses the combined capacity of the world’s leading AI data centres. It represents the largest infusion of capital into artificial‑intelligence infrastructure since the advent of the first commercial supercomputers.
How Vera Rubin Architecture Maximizes Compute Throughput
Vera Rubin architecture maximizes compute throughput
The partnership will be executed through NVIDIA’s Vera Rubin platform, a next‑generation architecture designed to maximise throughput while minimising power consumption. The first phase of the deployment is targeted to come online in the second half of 2026.
Jensen Huang, founder and chief executive of NVIDIA,
Jensen Huang, founder and chief executive of NVIDIA, underscored the strategic significance of the deal, noting that the collaboration “marks the next leap forward—deploying 10 gigawatts to power the next era of intelligence.” Sam Altman, co-founder and CEO of OpenAI, echoed this sentiment, describing compute infrastructure as the foundation for the future economy and emphasising that the partnership will enable both breakthrough research and scalable commercial applications. Greg Brockman, co‑founder and president of OpenAI, highlighted the long‑standing relationship between the two organisations, citing the use of NVIDIA’s platform in the development of ChatGPT and other widely deployed AI systems.
Strategic Alignment Optimizes Software and Hardware Roadmaps
Strategic alignment optimizes software and hardware roadmaps
The letter of intent positions NVIDIA as OpenAI’s preferred strategic compute and networking partner, with both parties agreeing to co‑optimise their roadmaps for model and infrastructure software alongside NVIDIA’s hardware and software. By aligning hardware, software, and power delivery, the alliance aims to accelerate the training of increasingly sophisticated models while maintaining energy efficiency. The collaboration is expected to reinforce OpenAI’s mission to build artificial general intelligence that benefits all of humanity, leveraging the unprecedented scale of AI supercomputing to push the frontier of intelligence further than any previous endeavour.
First Phase Deployment Using Vera Rubin Platform
First Phase to Go Live in Second Half of 2026 Using Vera Rubin Platform. The initial rollout of the joint effort is slated to commence in the latter half of 2026, with the Vera Rubin architecture serving as the backbone for the new compute infrastructure. Under the agreement signed in San Francisco and Santa Clara on 22 September 2025, NVIDIA has pledged up to $100 billion to support the deployment of a minimum of 10 gigawatts of GPU‑based processing power.
Jensen Huang described the collaboration as a pivotal
Powering the next generation of artificial intelligence
Jensen Huang described the collaboration as a pivotal step toward harnessing the next generation of intelligence, while Sam Altman stressed that the scale of compute will underpin the future economy. By delivering 10 gigawatts of sustained performance, the partnership aims to elevate AI supercomputing to a level that could accelerate the development of general‑intelligence systems and broaden the reach of AI applications across industries.
Achieving 10 GW of dedicated compute power necessitates significant advancements in data center thermal management, moving beyond traditional liquid cooling models. The sheer density of next-generation AI accelerators generates unprecedented heat loads, requiring innovative direct-to-chip liquid cooling loops integrated into the server infrastructure. This shift in power delivery architecture is critical, as efficient heat extraction is as important to maintaining operational uptime and maximizing the effective computational throughput of the Vera Rubin systems as the raw transistor count itself.
A foundational requirement for training massive models, such as the trillion-parameter scale anticipated for AGI, is an extremely high-bandwidth, low-latency interconnect fabric. The architecture relies heavily on advanced networking solutions, ensuring that data transfer between thousands of specialized chips does not become the compute bottleneck. These proprietary networking layers allow distributed model training to operate coherently, enabling the massive parallelization necessary to handle the petabytes of data required for modern foundation models.
Furthermore, the development pathway is constrained by the evolving complexity of the software stack. OpenAI and NVIDIA are therefore co-optimizing the machine learning compilers and operating systems to fully exploit the parallel processing capabilities of the new hardware. This joint optimization addresses the ‘full-stack’ challenge, ensuring that the algorithms and models can efficiently map their compute requirements onto the novel architecture, maximizing the utilization rate of every deployed compute cycle.
From a broader industry perspective, this commitment underscores the shift from isolated supercomputing clusters to integrated, utility-scale AI infrastructure. The deployment requires managing complex power grids and addressing geopolitical supply chain risks related to advanced semiconductor manufacturing. The sheer scale of the capital outlay also solidifies the compute capacity gap, making access to next-generation accelerators a central determinant of future technological leadership.
