U.S. semiconductor corporations AMD will invest up to $5 billion (about 7.4 trillion won) in generative artificial intelligence (AI) developer Anthropic. In return, Anthropic decided to adopt AMD's next-generation AI accelerators at scale.
According to the Wall Street Journal (WSJ) on the 22nd (local time), Anthropic plans to secure AMD's latest graphics processing unit (GPU), the MI450, with a total capacity of 2 gigawatts (GW) starting in the first half of 2027. Some of the units will be purchased directly by Anthropic and installed in its own data centers, while the rest will be rented through major cloud providers and NeoCloud, a specialist in GPU infrastructure.
AMD is said to inject funds into Anthropic in stages in line with chip supply and infrastructure build-out conditions. This is AMD's first direct investment in Anthropic, and the actual amount will be determined by whether certain business milestones, such as chip deployments and data center construction, are achieved. The two companies are jointly searching for candidate sites for the necessary data centers, and AMD is also reportedly reviewing ways to support future facility lease agreements.
The two companies will go beyond a simple investment and purchasing relationship and embark on joint engineering. They are also pushing plans to use Anthropic's AI model Claude in AMD's chip design and performance improvement process.
The deal is seen as AMD's strategy to broaden its supplier and customer base in the AI Semiconductor market led by Nvidia. AI corporations, including Anthropic, are also expanding their computing infrastructure and diversifying chip supply chains to reduce their reliance on Nvidia products.
Google earlier agreed to provide guarantees for certain data center contracts so that Anthropic can use its own tensor processing units (TPUs). Lisa Su, AMD chief executive officer (CEO), emphasized that AMD will play a central role in Anthropic's AI infrastructure.