OpenAI and Advanced Micro Devices said the companies have signed a multi-year agreement under which AMD will provide up to 6 gigawatts of computing power for OpenAI’s forthcoming AI infrastructure, marking one of the largest single vendor commitments to power for an AI developer to date. The first batch, equivalent to roughly 1 gigawatt of installed compute capacity, is slated to be deployed in the second half of 2026, the companies said in a joint statement.

"This agreement will help OpenAI scale the infrastructure needed for our next-generation models while expanding AMD’s footprint in data-center AI," the joint statement said. Both companies described the arrangement as a long-term strategic partnership but did not disclose pricing, the exact hardware models involved or how the chips will be distributed across OpenAI’s data centers and cloud partners.

Industry observers said the pact represents a notable shift in the competitive landscape for AI accelerators. For years, Nvidia has been the dominant supplier of GPUs for large-scale training and inference, but AMD has been aggressively developing its data-center accelerators and software stack to challenge that hegemony. For OpenAI, diversifying suppliers can reduce dependence on a single vendor and give it more leverage on costs and integration.

The scale of the agreement also highlights the vast electricity demands of modern AI. A gigawatt, engineers note, is equivalent to the output of a large power plant; six gigawatts of compute power would require substantial new energy provisioning and cooling capacity. "The scale of compute here raises important questions about energy use and grid impact," said an AI policy scholar at a major university. Local utilities, regulators and the companies themselves will need to coordinate on siting, energy sourcing and efficiency measures as deployment plans solidify.

Beyond energy concerns, the deal carries geopolitical and regulatory considerations. As U.S. policymakers have tightened export controls on advanced chips and semiconductor tools to limit transfers to certain countries, companies like AMD must navigate complex compliance regimes. Trade analysts warned that such restrictions could affect where OpenAI and AMD can place hardware and which international partners can access it.

OpenAI’s announcement comes amid a broader industry race to build computing capacity for increasingly large and capable models. The company’s recent moves to work with multiple cloud providers and hardware vendors reflect both the capital intensity of training state-of-the-art models and the strategic value of diversity in suppliers.

Analysts said the agreement could accelerate competition and innovation in data-center hardware, but it also underscores widening debates about who controls the massive computational resources that underpin advanced AI. "When enormous compute is concentrated, it changes the economics and governance of AI development," the AI policy scholar said. For policymakers, investors and the public, the AMD-OpenAI deal is a reminder that the technological choices made today—about chips, power and location—will shape the capabilities and consequences of AI for years to come.