AMD to supply 6GW of compute capacity to OpenAI in chip deal worth tens of billions (techcrunch.com)

AMD has struck a multi-year deal to supply OpenAI with 6 gigawatts of compute capacity across multiple generations of its Instinct GPUs — starting with the upcoming MI450 — in a pact that could generate “tens of billions” in revenue for the chipmaker. OpenAI will get the first gigawatt in H2 2026; the total capacity equals roughly the power draw of 4.5 million homes. AMD says MI450 will outperform Nvidia’s Rubin CPX through combined hardware and software improvements developed with OpenAI, building on the company’s existing MI300X/MI355X designs that are already favored for LLM inference due to large on‑chip memory and high bandwidth. The deal is significant because it diversifies OpenAI’s supplier base in a market long dominated by Nvidia and materially accelerates AMD’s push into hyperscale AI. Financially, AMD granted OpenAI the option to buy up to 160 million shares (about 10% of AMD), with tranches vesting as capacity is deployed and a final tranche tied to a lofty $600 share-price milestone; AMD stock jumped ~35% on the news. Strategically, the agreement complements OpenAI’s broader compute sourcing (Nvidia, Broadcom, DRAM deals with Samsung/SK Hynix and its Stargate datacenters), and could shift competitive dynamics around GPU architecture choices, memory-centric inference designs, and co‑optimized hardware–software stacks for large models.