AMD to invest up to $5bn in Anthropic

AMD said on 22 July 2026 it will invest up to $5bn in Anthropic. The equity is committed upon achieving deployment milestones. Terms including the valuation were not disclosed.

Anthropic will deploy up to two gigawatts of AMD Instinct MI450-series accelerators. The systems ship as AMD Helios rack-scale platforms with EPYC processors, Pensando networking and ROCm software. The first gigawatt begins in the first half of 2027.

The arrangement inverts AMD's earlier AI deals. Its OpenAI agreement covered up to six gigawatts with warrants over AMD stock; a February deal with Meta covered up to six gigawatts with performance-based warrants. Here AMD invests and takes a stake.

Anthropic plans to use Claude to optimise workloads for Instinct GPUs and accelerate ROCm development. AMD will deploy Claude across its engineering and product-development teams. Anthropic already uses Instinct MI355X GPUs; this is a substantially larger commitment.

The deal follows Anthropic's April statement that Claude demand caused 'inevitable strain' on infrastructure. Its compute stack spans Google TPUs, Amazon Trainium and Nvidia GPUs, and the lab was reported in July to be in early talks with Samsung about a custom accelerator.

AMD's shares rose on the announcement. The company has signed three gigawatt-scale AI agreements in this cycle: OpenAI at up to six gigawatts, Meta at up to six gigawatts, and now Anthropic at up to two. Only the Anthropic deal has AMD investing rather than issuing warrants over its own stock.

Independent adoption remains the open question. AMD's Instinct silicon has been competitive on memory bandwidth for two generations, while ROCm has trailed Nvidia's CUDA on tooling maturity. No timetable beyond the deployment milestones has been published.