AMD plans up to $5 billion Anthropic investment in AI chip deal

AMD said Wednesday it has struck a strategic partnership with Anthropic that includes a planned investment of up to $5 billion in the artificial intelligence company. For everyday investors watching the AI trade, the deal is another sign that demand for computing power remains a central driver for chipmakers and AI labs.

Under the agreement, AMD said Anthropic will deploy 2 gigawatts of AMD Instinct MI450 Series graphics processing units in AMD Helios rack-scale systems. A graphics processing unit, or GPU, is a chip built to process many calculations at once, which makes it useful for training and running large AI models.

The companies are using gigawatts to describe the size of the deployment. In this context, a gigawatt is a measure of power tied to AI data center capacity, which has become a common way to describe how large these computing projects are. AMD said the first gigawatt is expected to be deployed in the first half of next year.

AMD CEO Lisa Su said in the company’s announcement that AMD is “thrilled to deepen” its partnership with Anthropic and deploy AMD Helios at gigawatt scale.

What AMD gets from the deal

AMD has become one of the main public-market ways investors track AI infrastructure spending because it sells the GPUs used for large AI workloads. CNBC reported that AMD competes directly with Nvidia, which dominates the GPU market, while AMD has signed agreements with major AI companies, including OpenAI.

The Anthropic agreement gives AMD another high-profile customer for its AI hardware. Rack-scale systems, such as AMD Helios, package chips, networking and other hardware together so customers can deploy large clusters more efficiently than buying individual components.

For AMD, the investment also ties capital to a customer relationship. The company said the investment could reach $5 billion, though AMD did not disclose the full structure of the investment in the announcement cited by CNBC.

Anthropic keeps adding compute

Anthropic, the company behind Claude, has been lining up major infrastructure deals this year as it works to expand computing capacity. Compute capacity means the amount of processing power available to train models and serve users who send requests to AI systems.

In April, Anthropic said demand for Claude had placed “inevitable strain” on its infrastructure and affected reliability and performance, especially during periods of heavy use. More infrastructure can help an AI company handle more users, run larger models and reduce slowdowns, though it also requires significant spending.

CNBC reported that Anthropic reached a deal in May with Elon Musk’s SpaceX to use all of the compute capacity at SpaceX’s Colossus 1 data center in Memphis, Tennessee. According to SpaceX’s prospectus cited by CNBC, Anthropic will pay SpaceX $1.25 billion per month through May 2029 under that arrangement.

Anthropic also signed a multibillion-dollar agreement with Amazon in April, according to CNBC. The company separately announced a multi-gigawatt compute agreement with Google and Broadcom that same month. CNBC has also reported that Anthropic is in early talks to lease computing power from Meta.

Taken together, the agreements show how quickly leading AI companies are trying to secure the hardware and power needed to keep their products running. For investors, AMD’s Anthropic deal adds another data point in the competition among chipmakers, cloud providers and AI labs for control of the infrastructure behind generative AI.

This story draws on original reporting from CNBC.