AMD is putting up to $5 billion behind Anthropic and committing to expand the AI lab’s computing power, according to an announcement made Wednesday and reported by The Verge AI. The deal pairs cash with silicon: Anthropic will deploy up to 2 gigawatts of AMD’s Instinct MI450 AI GPUs, running on the chipmaker’s new Helios rack-scale system. This is significant because it hands AMD its highest-profile AI customer yet and gives Anthropic another pillar under its fast-growing compute needs.
What was announced
The headline numbers are big, and the timeline is concrete. Here’s what the two companies committed to, per The Verge AI:
- Up to $5 billion invested by AMD into Anthropic.
- Up to 2 gigawatts of Instinct MI450 GPU capacity for Anthropic.
- First gigawatt goes live in the first half of 2027.
- A multi-year engineering collaboration, with AMD adopting Anthropic’s Claude across its software, engineering, and product development.
The hardware runs on Helios, AMD’s rack-scale system built to compete at the data center level, not just chip for chip. That’s the part worth watching. Selling one GPU is a sale. Selling a full rack architecture that a frontier lab trains and serves models on is a platform win.
Why it matters
For years, Nvidia has owned AI training. Its chips and its CUDA software have been the default, and rivals kept losing the argument on performance and ecosystem. AMD landing a customer like Anthropic changes that story. What stands out here is that a top-tier lab is willing to run production Claude workloads on AMD silicon, not just kick the tires in a test cluster.
The engineering collaboration cuts both ways. AMD gets Anthropic’s real training and inference workloads to optimize against, which is exactly the feedback loop it has lacked. Anthropic gets hardware tuned to how Claude actually runs. As Tom Brown, Anthropic cofounder and chief compute officer, put it in the press release quoted by The Verge AI: “By partnering with AMD across the stack, we are securing the capacity we need and optimizing it for training and serving Claude.”
The bigger pattern
This isn’t a one-off for Anthropic. It’s the latest move in an aggressive compute land grab. The AMD partnership builds on recent data center deals with SpaceX and TeraWulf, and it sits alongside infrastructure agreements with Google, Broadcom, and Amazon. The Verge AI notes rumors that a deal with Meta could follow.
Read those deals together and the strategy is clear:
- Diversify suppliers. Anthropic is spreading bets across Nvidia alternatives, custom silicon, and multiple cloud partners rather than depending on any single vendor.
- Lock in capacity early. Compute is the binding constraint on frontier AI. Signing 2027 capacity now is about guaranteeing there’s room to grow.
- Push chip competition. More buyers backing AMD means more pricing and roadmap pressure on Nvidia, which benefits everyone building on these platforms.
What to expect next
The first gigawatt won’t land until early 2027, so don’t expect an overnight shift in who trains what on which chips. But the direction is set. If AMD can prove Helios and the MI450 hold up under Anthropic’s workloads, expect other labs and cloud providers to take a harder look at breaking their Nvidia dependence.
For practitioners, the near-term signal is about optionality. A serious second source for AI compute means more competition on price, availability, and tooling over the next 18 months. Watch three things: whether the MI450 delivers on real Claude training runs, whether AMD’s software stack closes the gap with CUDA, and whether that rumored Meta deal turns into another name on Anthropic’s supplier list.
Anthropic is buying insurance against the one thing that could slow it down: not enough chips. AMD, for its part, just got the reference customer it needed to be taken seriously in frontier AI. More details are available in the original report from The Verge AI.