Anthropic invests $9.1 billion in Riot Platforms infrastructure

The artificial intelligence company Anthropic, developer of the chatbot Claude, has signed a massive $9.1 billion deal with Bitcoin mining company Riot Platforms. As part of the agreement, Riot, which recently began selling data center capacity for artificial intelligence, will provide the company with large-scale computing power.

The move highlights Anthropic's growing efforts to secure enough computing power to meet the increasing demand of its customers for AI tools. Earlier, Riot reported that it had secured a 20-year contract to supply 191 megawatts of computing power — the amount of electricity required to power about 143,000 homes at any given moment — from the company's facility in Rockdale, Texas, for a "leading AI company."

Sources involved in the matter confirmed to Bloomberg that it is indeed Anthropic. Following the reports, Riot's stock jumped by 25% to a price of $24.40 in after-hours trading.

Riot, which previously operated under the name Bioptix and was engaged in the production of diagnostic devices for the biotech industry before switching to Bitcoin mining, is one of several crypto companies now shifting their resources to cloud computing to profit from the artificial intelligence boom.

The company expects that the contract, which runs until June 2048, will generate $9.1 billion in revenue, with the agreement also including an option for two five-year extensions, which could increase total sales to $16.1 billion. Additionally, Riot announced a separate deal with Advanced Micro Devices (AMD) to build computing infrastructure.

The current deal joins a series of agreements signed by Anthropic in recent months. Among other things, the company recently signed a $10 billion deal with infrastructure startup Volta Infra Holdings, and as early as last May, it agreed to purchase computing power worth about $45 billion from Elon Musk's xAI company.

In parallel with the cloud deals, Anthropic recently updated the biological safety mechanisms in its Claude Fable 5 model. The update is intended to reduce by about 85% the erroneous blocks for legitimate queries on health topics, laboratory test interpretation, symptoms, and academic research, while maintaining strict limitations on sensitive topics such as virology and toxicology.

Additionally, the company is working on designing dedicated chips for Claude models alongside continuing its existing collaborations with technology giants Nvidia, Google, and AWS.