AMD CEO Su Talks About Huge Chip Deal With OpenAI

Key Concepts

• AI Compute: The computational power required to run artificial intelligence models, encompassing both training and inference.
• Inference: The process of using a trained AI model to make predictions, decisions, or generate outputs in real-world applications. This is distinct from "training," which is the process of building the model.
• Gigawatt (GW): A unit of power, equivalent to one billion watts. In this context, it refers to the scale of electrical power capacity dedicated to AI compute.
• MI450 Chip: AMD's next-generation AI accelerator chip, central to the compute build-out.
• Compute Desert: A term used to describe a future scenario where the demand for AI computational power far outstrips the available supply.
• Total Addressable Market (TAM): The total revenue opportunity that is available for a product or service, in this case, for AI accelerators.
• Supply Chain (AI): The entire network of entities involved in producing and delivering AI compute, from energy generation to chip manufacturing and data center deployment.
• Accretive (financial): A financial term indicating that an acquisition or investment is expected to increase the value per share of the acquiring company.
• OpenAI: A leading AI research and deployment company, partner in this initiative.
• AMD: Advanced Micro Devices, a semiconductor company, providing the compute hardware.

---

The Landmark AMD-OpenAI Partnership and its Strategic Importance

The collaboration between AMD and OpenAI marks a significant milestone for both companies and the broader AI industry. Lisa Su of AMD emphasized that this partnership represents a "huge moment for the industry because... you need more compute," highlighting compute as the fundamental foundation for all AI intelligence. AMD, as a compute provider, has dedicated years to its roadmap and working with OpenAI. The partnership is embarking on a "massive build out of six gigawatts of AI compute," which is deemed a substantial deal for AMD, its shareholders, teams, and the overall ecosystem.

Addressing the Exploding Demand for AI Compute: The "Compute Desert"

Greg Brockman of OpenAI underscored the world's underestimation of AI compute demand, describing the current trajectory as heading towards a "compute desert." He cited the explosive growth of AI applications, such as chatbots, which garnered 800 million weekly active users within three years of non-existence. OpenAI currently faces limitations in launching new features and products due to a lack of computational power, even as AI models continue to improve exponentially. The core objective of this build-out is to provide the necessary compute for inference, which is crucial for driving an AI-powered economy. The initial target is one gigawatt, with an eventual goal of six gigawatts, built as quickly as possible to bring computational power to the economy and the world.

AMD's Commitment, Technology, and Deployment Plan

This partnership represents AMD's largest announced deployment to date. The first gigawatt of compute is slated for deployment in the second half of 2026, utilizing AMD's new, next-generation MI450 Chip. Lisa Su clarified that such "all-in partnerships" take years to develop, involving extensive engineering work across hardware and software, and meticulous supply chain management to ensure delivery on this massive commitment. The MI450 series is anticipated to be a "really incredible chip," offering a diversity of memory and computational power balances suitable for varied AI workloads, particularly for inference.

Industry-Wide Collaboration and Supply Chain Considerations

The initiative is viewed as an industry-wide effort, requiring the entire supply chain to scale up significantly. This includes securing far more energy, with nuclear power being highlighted as a potentially important source. OpenAI plans to deploy AMD chips in its own data centers and in collaboration with various cloud providers, including Oracle. Greg Brockman stressed that the goal is to acquire "as much compute as possible" for the economy, the nation, and humanity, necessitating collaboration with "everyone in this whole industry." The deployment will occur across multiple locations and with multiple providers, rather than a single site, to achieve rapid online availability.

Financing the Massive AI Compute Infrastructure

OpenAI's financing strategy is underpinned by its rapidly growing AI revenue, which Greg Brockman stated is "growing faster than I think almost any product in history." The company is actively exploring various financing avenues, including equity, debt, and other creative solutions, to fund the substantial compute infrastructure required to transform the economy into an AI-powered one. From AMD's perspective, the deal is a "win for AMD, a win for OpenAI, and a win for our shareholders." As OpenAI purchases chips, AMD's revenue and earnings are expected to increase, making the deal "very, very accretive" to AMD's shareholders from day one, fostering a "virtuous, positive cycle."

Manufacturing Strategy and Global/National Security Implications

AMD is prioritizing manufacturing in the United States, aligning with the "USAID stack" vision and citing over $500 billion in US investments. Lisa Su confirmed a deep partnership with TSMC for the supply chain. However, Greg Brockman also acknowledged the global need for compute, viewing it as a "national security, strategic resource" that every country will require. While prioritizing US leadership in this technology, AMD is also working with global partners to meet anticipated demand in the coming years, not limiting its sites for building.

AMD's Role in the AI Accelerator Market and Workload Diversity

OpenAI's decision to partner with AMD for inference workloads is strategic. While OpenAI has historically focused its AI training efforts on NVIDIA platforms due to the "huge fixed cost" involved, the "easier barrier to entry" for inference makes AMD's technology a compelling alternative. Greg Brockman noted that the MI450 series is proving to be an "incredible chip," and that having a "diversity of chips" accelerates capabilities across a "diversity of workloads." This partnership validates AMD's technology and capability, positioning it strongly in the AI accelerator market. Lisa Su reiterated her belief in a "huge market" for AI accelerators, estimating the Total Addressable Market (TAM) to be over $500 billion in the next few years, suggesting her previous estimates might have been conservative given the immense demand.

Pathway to Deployment and Future Outlook

The pathway to the first gigawatt deployment is set for the second half of next year, with a commitment to build as fast as possible. This involves not only technological milestones but also commercial and adoption milestones. Both companies foresee increasing financial ties between AI users and chip makers as the world transitions to an AI-powered economy. Greg Brockman emphasized that access to better AI tools directly correlates with productivity, stating that "if you can have ten times as much air power behind you, you will probably be ten times more productive." The entire industry, from power generation to silicon, must "rise to meet the occasion" of this unprecedented demand explosion. The partnership serves as a significant catalyst for the industry to accelerate its build-out of AI compute infrastructure.