Anthropic has finalized a 20-year cloud agreement worth $9.1 billion with the bitcoin mining company Riot Platforms.

      Anthropic has completed another significant deal to satisfy its demand for computing power, this time partnering with a company that recently profited from bitcoin mining. The twenty-year contract, valued at $9.1 billion, provides the developer of Claude with ample capacity at a Texas facility that Riot Platforms is currently transforming from cryptocurrency mining operations into rented computing resources for AI. This agreement is part of a series of substantial infrastructure contracts that the company has secured to obtain the necessary hardware for its models.

      The agreement will last until June 2048 and includes two five-year extension options, which, if utilized, could increase the total value to $16.1 billion. This arrangement resembles more of a long-term partnership between an innovative AI lab and a past bitcoin miner rather than a simple lease.

      In exchange for this investment, Anthropic will have access to 191 megawatts of IT capacity at Riot’s Rockdale facility, with the capacity being deployed in phases rather than all at once. Riot anticipates that 96MW will be operational by December 2027, with the full 191MW available by June 2028, a schedule that assumes the ongoing rapid pace of data center construction in Texas continues.

      To support this development, Riot is turning to financial assistance from Wall Street. Morgan Stanley is providing $573 million in temporary financing to cover the costs associated with building the site, a common move as the demand for AI's processing power exceeds the available funds of companies striving to meet it.

      Investors responded positively to the news, with Riot's shares rising approximately 25% in after-hours trading on August 10, reaching around $24.30, reflecting the market's enthusiasm for anything related to AI as much as the specifics of the contract.

      A reliable stream of payments over twenty years from a well-funded client is significantly more stable than the fluctuations in bitcoin prices. The leasing company is an unexpected choice for this role; Riot Platforms, a historically successful bitcoin miner that produced 1,587 bitcoins in the second quarter of 2026, is now repositioning itself as a provider of computing facilities.

      CEO Jason Les noted that the company has "secured leases totaling 241 megawatts of capacity, translating to approximately $9.8 billion of long-term, contracted revenue" with prominent AI companies within a six-month timeframe. This shift is not exclusive to Riot, as several former bitcoin miners have realized that their vast, power-intensive facilities initially built for mining are ideally suited for training and running AI models, prompting them to lease this capacity to willing clients.

      Anthropic has adapted to a model where it relies on others to build the data centers, which it then occupies with its workloads. The reasoning for Anthropic is simple, albeit somewhat overwhelming. The company has been rapidly entering into computing agreements that would have seemed preposterous a few years ago, establishing multi-billion-dollar contracts with cloud providers, chip producers, and now cryptocurrency miners, all to secure computing resources ahead of its competitors.

      The demand for its models is continually increasing, and the only way to meet this demand is by persistently acquiring, leasing, and reserving processing power wherever available, even in repurposed facilities from the crypto industry.

      There is an interesting irony in this scenario. The energy that used to be expended on generating digital currencies is now being redirected towards creating digital intelligence, transforming the same facilities filled with operational machines to serve new purposes. However, whether the economics of these two-decade commitments will still seem sound in 2048, when the current excitement may be a distant memory, remains a question that those signing the contracts seem hesitant to explore.