AMD to invest up to $5bn in Anthropic under AI chip partnership

Advanced Micro Devices said Wednesday it will invest up to $5bn in Anthropic as part of a strategic partnership that will put AMD chips into the AI developer’s computing infrastructure. Under the agreement, Anthropic will deploy as much as 2 gigawatts of AMD Instinct MI450 Series graphics processing units in AMD Helios rack-scale systems, according to AMD’s announcement.

The first gigawatt of capacity is scheduled for deployment in the first half of next year, AMD said. Gigawatts, a power measure, have become a common shorthand for the scale of AI data centres because large model training and inference require dense clusters of chips, high electricity supply and advanced cooling systems.

AMD chief executive Lisa Su said in a statement that the company was “thrilled to deepen our partnership with Anthropic and deploy AMD Helios at gigawatt scale.”

The deal strengthens AMD’s position in the market for AI accelerators, where its graphics processing units are used to train models and process large AI workloads. Nvidia remains the dominant supplier in the sector, while AMD has been seeking to expand its share through agreements with large AI companies, including a previously announced arrangement with OpenAI, an Anthropic competitor.

For Anthropic, the AMD pact adds another supplier and another pool of computing capacity as demand for its Claude AI models puts pressure on infrastructure. In April, Anthropic said demand for Claude had placed “inevitable strain” on its systems, affecting “reliability and performance” for users, particularly during peak usage periods.

Anthropic adds to infrastructure commitments

The AMD agreement follows several infrastructure deals announced by Anthropic this year. In May, the company reached an agreement with Elon Musk’s SpaceX to use all available computing capacity at SpaceX’s Colossus 1 data centre in Memphis, Tennessee. Anthropic will pay SpaceX $1.25bn a month through May 2029 under that arrangement, according to SpaceX’s prospectus.

Anthropic also announced a multibillion-dollar agreement with Amazon in April, along with a multi-gigawatt compute agreement involving Google and Broadcom. CNBC has previously reported that Anthropic is in preliminary discussions to lease computing power from Meta.

The sequence of transactions reflects a broader capital race among AI model developers, cloud providers and chipmakers. Large AI systems require specialised processors, networked server racks and substantial electricity commitments. Companies seeking to run frontier models often sign long-term capacity agreements because building or securing data-centre resources can take years.

AMD’s Helios systems are designed as rack-scale platforms, meaning chips, networking and supporting hardware are organised as integrated units rather than sold only as individual processors. That structure can matter for AI workloads because performance depends not only on the GPU itself, but also on how efficiently thousands of processors exchange data across a cluster.

AMD did not disclose the exact timing or terms under which its investment in Anthropic would be made beyond saying it could reach up to $5bn. The company framed the arrangement as a strategic partnership tied to the deployment of its Instinct MI450 Series GPUs.

This story draws on original reporting from CNBC.