AMD bets $5B on Anthropic to build a 2-gigawatt AI empire by 2027

AMD is putting its money where its chips are, agreeing to invest up to $5 billion in Anthropic as part of a massive infrastructure deal. This isn’t just a cash injection; it’s a pact to build out tens of billions of dollars worth of AI systems, with Anthropic set to deploy up to two gigawatts of capacity using AMD’s next-gen Instinct MI450-series accelerators. The first gigawatt of that rollout is slated to begin in the first half of 2027.

The financial engineering here is notable. Unlike a straightforward purchase order, AMD’s investment is tied directly to Anthropic hitting specific, undisclosed deployment milestones. This structure mirrors other recent AMD deals but with a key difference: it’s a direct equity stake rather than the warrants offered to OpenAI and Meta in their own multi-gigawatt agreements. Lisa Su, AMD’s chief executive, framed it as a collaboration that brings together “Anthropic’s leadership in frontier AI with the full strength of AMD high-performance computing.”

For Anthropic, this adds a crucial second source of silicon to its compute network, which has been heavily reliant on Nvidia. The company already runs more than 220,000 Nvidia GPUs at a SpaceX facility in Memphis, but the planned AMD deployment represents over six times that power capacity. The new systems will be a full rack-scale platform, including EPYC “Venice” CPUs and Pensando networking, not just a box of GPUs. Some of this kit will land in Anthropic’s own data centers, with the rest hosted by cloud providers and specialist AI infra firms still being lined up.

This deal is as much about AMD’s ambitions as it is about Anthropic’s. By locking in a premier AI lab as a flagship customer for its MI450 platform, AMD is trying to prove its Instinct line can handle the brutal demands of frontier model training and inference at a scale that genuinely rivals Nvidia. The pressure is now on to deliver the silicon and make the software ecosystem, particularly ROCm, bulletproof. If they pull it off, it reshapes the competitive landscape. If they stumble, it’s a very expensive lesson in the gap between hardware promises and hyperscale reality.