Introduction
Welcome back to Laboratory. This week we look at OpenAI’s new deal with AMD: a 6 gigawatt partnership that locks in GPU supply for years, starting with 1 gigawatt of MI450 chips in 2026. The agreement also includes a warrant for up to 160 million AMD shares, tying OpenAI’s scale-up to AMD’s success. The goal is clear: secure enough power and hardware to keep training frontier models without waiting on Nvidia.
In this briefing, we’ll chart:
Supply as Strategy: How working with multiple chipmakers gives OpenAI more flexibility and leverage in a tight GPU market.
Power as a Moat: What 6 GW of compute really means for data center buildouts, energy sourcing, and grid planning.
Software as the Decider: Why AMD’s ROCm platform – short for Radeon Open Compute, the software that lets AI models run on AMD chips – will decide whether this partnership can truly rival Nvidia’s CUDA ecosystem.