techOpenAI’s Two-Track Chip Play: AMD GPUs now, Broadcom custom silicon next

Core Insights - OpenAI is making significant hardware commitments to enhance its AI infrastructure, focusing on increased capacity, control, and supplier diversity [1] Group 1: Hardware Commitments - OpenAI will deploy 6 gigawatts of AMD Instinct GPUs, starting with an initial 1 gigawatt of MI450 in the second half of 2026, as part of a multi-year, multi-generation roadmap [2] - In addition, OpenAI will co-design custom AI accelerators with Broadcom, planning to deploy 10 gigawatts of these systems between 2026 and 2029 [2] - Overall, analysts estimate that OpenAI is aligning approximately 26 gigawatts of AI infrastructure across various vendors, indicating a clear direction for future growth [2]