OpenAI Signs AMD Chips Deal to Support Build Out

Partnership & Strategy - AMD and OpenAI are embarking on a massive buildout of 6 gigawatts of AI compute, representing a significant milestone for both companies and the AI industry [3] - The partnership between AMD and OpenAI is an "all-in" commitment, involving hardware, software, and supply chain collaboration to meet OpenAI's AI compute needs [9] - The initial target is 1 gigawatt of compute in the second half of 2026 using the next-generation MI450 chip [4][8] - AMD views this deal as a win for AMD, OpenAI, and its shareholders, expecting it to be accretive to AMD's earnings from day one [19][21] Market Demand & Compute Landscape - The world is underestimating the demand for AI compute, with an explosion of demand driven by products like ChatGPT, which has 800 million weekly active users [5] - The industry is heading towards a "compute desert" due to the exponential improvement of AI models and the increasing reliance of the economy on AI [6] - OpenAI sees compute as a limiting factor for launching new features and products due to a lack of computational power [5] - The AI accelerator total addressable market (TAM) is estimated to be over $500 billion in the next few years [39] Supply Chain & Manufacturing - The industry requires an industry-wide effort to increase compute capacity, involving energy, cloud providers, and the entire supply chain [11] - AMD is prioritizing building in the United States to create a strong US AI stack and is working with global partners to meet the expected demand [25][26] - OpenAI is deploying AMD in its own data centers and with cloud providers, including Oracle, emphasizing the need for as much compute as possible [12][13] Technology & Competition - AMD's MI450 series is viewed as an incredible chip with a good balance of memory and computational power, suitable for inference workloads [28] - The deal with AMD is incremental to what OpenAI is already doing with NVIDIA, with AMD technology offering specific benefits for inference [27] - There is a diversity of workloads across customers, models, and use cases, positioning AMD well in the market [31]