AMD has agreed to invest up to $5 billion in Anthropic, marking a significant expansion of the strategic partnership between the two companies. In return, Anthropic plans to deploy up to 2 gigawatts of AMD Instinct MI450 GPUs housed in Helios server systems to train and run its Claude family of AI models — a scale of compute comparable to the power consumption of a small city and among the largest single-vendor AI infrastructure commitments publicly disclosed.

The deployment will proceed in phases, with the first gigawatt phase scheduled to begin in the first half of 2027. The Helios systems will combine MI455X GPUs with AMD EPYC "Venice" processors and AMD's proprietary networking technology. Anthropic already operates AMD's MI355X GPUs in its existing infrastructure.

Beyond hardware procurement, the two companies have outlined a multi-year collaboration to use Claude to enhance AMD's ROCm software platform and optimize GPU workloads. ROCm is AMD's answer to Nvidia's CUDA, the software ecosystem that has long anchored developer loyalty to Nvidia's platform and remains one of the most significant barriers to switching accelerator suppliers. AMD also intends to deploy Claude internally across its development teams.

Anthropic co-founder Tom Brown stated that working with multiple hardware providers allows the company to match specific workloads to the best-suited hardware, a strategy that has guided Anthropic's infrastructure decisions. Anthropic has previously secured substantial compute commitments from Amazon Web Services and Google Cloud, making AMD's direct hardware partnership a notable diversification of its supply chain.

For AMD, the deal represents a substantial effort to position itself as a primary GPU supplier for AI companies alongside Nvidia, which has dominated the market for AI accelerator chips. AMD has recently secured comparable agreements with Meta and OpenAI, signaling a broader push into the AI compute space.

Critics have noted the circular financial structure of these arrangements, in which chip and cloud companies provide funding to AI labs, which in turn purchase hardware and services from those same companies. Whether AI labs — including Anthropic, whose Claude models compete with OpenAI's GPT family and Google's Gemini — will eventually generate sufficient revenue to cover these infrastructure costs independently remains an open question.

The partnership was announced via AMD's official newsroom.