[News] OpenAI Ramps Up Global Compute Power with NVIDIA, AMD, and Broadcom — Securing 26 GW of AI Infrastructure

In a rapid series of announcements over the past month, OpenAI has unveiled three landmark hardware partnerships that together amount to a staggering 26 gigawatts (GW) of AI compute capacity. The company has struck deals to:

• Co-develop 10 GW of custom AI accelerators with Broadcom
• Invest up to USD 100 billion with NVIDIA to deploy at least 10 GW of NVIDIA-powered systems;
• Roll out at least 6 GW of AMD Instinct GPU clusters over the next several years.

OpenAI × Broadcom: 10 GW of Custom AI Accelerator

On October 13, OpenAI and Broadcom announced a partnership to jointly develop 10 GW of custom AI accelerator.

Under the agreement, OpenAI will lead the AI accelerator and system architecture design, while Broadcom will provide Ethernet-based interconnects and networking solutions to build and scale the rack systems. These accelerators will be deployed across OpenAI’s facilities and partner data centers worldwide to meet surging global demand for AI computation.

Deployment is expected to begin in late 2026, with full rollout targeted for completion by 2029. Previously, OpenAI CEO Sam Altman revealed that the two companies have already been collaborating for 18 months, “reimagining the technology stack from the transistor level all the way to the ChatGPT user query.”

OpenAI × AMD: At Least 6 GW of AMD Compute Power

On October 6, OpenAI announced a multi-year partnership with AMD to deploy no less than 6 GW of AMD GPU compute power.

The collaboration combines technology and equity incentives. AMD has granted OpenAI warrants for up to 160 million AMD common shares, which will vest in stages based on deployment milestones — beginning with 1 GW in the first phase and gradually scaling up to the full 6 GW. Should OpenAI exercise all warrants, it could hold approximately 10% of AMD’s equity.

The initial 1 GW phase is scheduled to start in late 2026, powered by AMD’s next-generation Instinct MI450 GPU and paired with integrated rack-level AI systems. Each subsequent expansion phase will trigger additional warrant unlocks as compute capacity scales up.

OpenAI × NVIDIA: USD 100 Billion for 10 GW of AI Systems

On September 22, NVIDIA and OpenAI signed a memorandum of understanding outlining plans for NVIDIA to invest up to USD 100 billion in OpenAI’s infrastructure, supporting the deployment of at least 10 GW of NVIDIA AI systems.

The initiative aims to provide massive compute capacity for training and running OpenAI’s future large-scale models through the deployment of at least 10 gigawatts (GW) of NVIDIA-powered AI systems. According to industry estimates, 10 GW of compute roughly equates to the combined performance of 4 to 5 million high-end GPUs — a scale several to dozens of times larger than any existing single AI data center in operation today. Notably, the USD 100 billion investment will not be delivered in a single lump sum. Instead, it follows an innovative, progress-linked funding model, under which NVIDIA’s capital will be released incrementally as each 1 GW of new data center capacity comes online.

According to sources familiar with the matter, the first tranche of funding — potentially around USD 10 billion — is expected to be released upon the signing of the final agreement or the launch of the first project. The initial investment will primarily support infrastructure development and hardware procurement, with each funding round valued based on OpenAI’s market valuation at the time of investment.

Under the partnership roadmap, the first 1-gigawatt-scale AI system is scheduled to come online in the second half of 2026. This initial deployment will be built on NVIDIA’s next-generation Vera Rubin platform, which is expected to serve as the technological foundation for OpenAI’s forthcoming large-scale AI infrastructure.

(Photo credit: OpenAI)