Anthropic plans to deploy up to 2 gigawatts of AMD Instinct MI450 GPUs for training and running Claude, while AMD is investing up to $5 billion in the AI lab. The Decoder reports that the first gigawatt phase is expected to start in the first half of 2027.
The systems will use AMD’s Helios server design, pairing MI455X GPUs with EPYC “Venice” processors and AMD networking technology. Anthropic already uses AMD’s MI355X GPUs, and the two companies also plan a multi-year effort to use Claude to improve AMD’s ROCm software platform and GPU workloads.
For Anthropic, the deal diversifies the hardware base behind Claude at a time when frontier AI companies are trying to secure enormous amounts of compute. For AMD, it is another attempt to prove it can compete with Nvidia as a supplier for top AI labs, following other large AI infrastructure agreements.
The deal also raises a familiar industry question. Chip and cloud companies are funding AI labs that then buy their infrastructure. Whether AI revenue can eventually support these commitments without circular financing remains a key uncertainty.