AMD and OpenAI today announced a strategic partnership to construct a 6 gigawatt AI infrastructure leveraging AMD's Instinct GPUs, with the initial deployment of 1 gigawatt of AMD Instinct MI450 accelerators scheduled for the latter half of 2026. This partnership not only underscores AMD's growing impact in AI acceleration but also signifies OpenAI's transition from a single-supplier model for future large-scale training to an era of multi-vendor parallel computing.

At the core of this agreement is AMD's forthcoming MI400 series accelerators, with the MI450 being the inaugural product geared towards ultra-large-scale AI computing. According to AMD, the MI450 boasts an AI computing capability of up to 40 PFLOPs, is equipped with 432 GB of HBM4 high-bandwidth memory, and achieves a total bandwidth of 19.6 TB/s, which nearly doubles the overall performance of the current MI350X. OpenAI will utilize these GPUs to establish next-generation compute clusters, providing the foundational computational power needed for GPT, Sora, and future multimodal models. Besides the MI450, OpenAI's servers will incorporate a blend of MI300 and MI350 series GPUs to optimize power efficiency and throughput across different computation stages.

The collaboration between AMD and OpenAI, initiated with the MI300X series and extended to the MI350X and MI400 series, encompasses both hardware and software co-development, as well as rack-level system optimization. The two entities will jointly innovate on deep learning frameworks and driver interfaces to fully leverage the ROCm software stack's performance for distributed training. This extensive collaboration goes beyond mere hardware procurement to involve architecture tuning and ecological integration, illustrating a deeper technological alliance between AI chip manufacturers and model developers.

Significantly, the partnership includes a financial component, as AMD has issued warrants for up to 160 million shares of common stock to OpenAI, serving as an incentive for long-term cooperation. The first tranche of warrants will be activated upon the completion of the 1 gigawatt MI450 deployment, with subsequent tranches vesting progressively as the procurement target scales to 6 gigawatts. Moreover, vesting conditions are linked to AMD achieving specific stock price targets and OpenAI's deployment milestones, reflecting the profound technological synergy and shared capital interests between the two firms.

"The partnership with OpenAI is a critical step in achieving the world's most ambitious AI expansion," said Lisa Su, Chairman and CEO of AMD. "By blending AMD's expertise in high-performance computing with OpenAI's pioneering research and applications, we aim to jointly spearhead the next wave of AI infrastructure," stated Sam Altman, CEO of OpenAI. Greg Brockman, President of OpenAI, highlighted that "the future of AI necessitates deep collaboration at all levels, and AMD's involvement will significantly aid in scaling global deployments."

From AMD's vantage point, this partnership promises not only fresh revenue streams but also an enhanced competitive edge in the AI chip market. Jean Hu, AMD's Executive Vice President and CFO, projected that the agreement could yield tens of billions of dollars in revenue over the coming years and markedly enhance earnings per share. Given the brisk expansion in data center and generative AI markets, this contract is expected to provide strong momentum for AMD's sustained growth.

Historically, NVIDIA has dominated the AI chip market, with its Blackwell architecture leading in inference and training segments, enjoying margins as high as 78 percent. AMD, however, is establishing a solid secondary position with its Instinct series and ongoing advancements in the ROCm software ecosystem. The alliance with OpenAI represents a pivotal breakthrough for AMD in infiltrating top-tier AI platforms and is poised to drive greater adoption of its solutions among cloud service providers and research bodies.

The realization of this 6 gigawatt partnership marks a significant milestone in the global expansion of AI infrastructure. With the MI450 slated for production by 2026, the joint efforts of AMD and OpenAI are set to usher in a new wave of high-performance computing upgrades, redefining the parameters of energy efficiency and scalability for upcoming generations of AI training clusters.