Anthropic and TeraWulf sign $19 billion infrastructure and compute deal, with TeraWulf securing power-constrained data center sites to supply Anthropic's training and inference workloads.

On July 6, Bitcoin mining company TeraWulf signed a 20-year data center lease agreement worth approximately $19 billion. Two weeks later, another mining company, Hut 8, announced a long-term lease worth $9.8 billion to build an AI data center in Texas, leasing 352 MW of server space and power capacity to an undisclosed major client for 15 years. This is Hut 8's second contract of equal scale at the Beacon Point campus, bringing the total contract value of both phases to $19.6 billion.

Both companies started out mining Bitcoin. Until recently, TeraWulf and Hut 8 primarily made their living by mining cryptocurrency—adding equipment when prices rose and shutting down capacity when prices fell or electricity costs became prohibitive, causing revenue to fluctuate sharply with Bitcoin's price. Now, mining rigs are being phased out and original mining facilities are being renovated into AI data centers. The companies are shifting from mining cryptocurrencies themselves to leasing space and power to AI companies. The land, electricity, and grid connection rights once prepared for mining may now be more valuable than the mining equipment itself.

This pivot traces back to the financial collapse of Core Scientific. From the second half of 2020 through the end of 2021, Bitcoin experienced a major bull market, with its price rising from approximately $10,000 to nearly $69,000. Core Scientific aggressively expanded during this period, purchasing mining equipment and constructing large-scale operations in Texas, North Carolina, Georgia, and elsewhere, becoming one of North America's largest publicly traded mining companies. The logic was straightforward: as long as Bitcoin prices remained high, buying more miners and building larger facilities offered a faster path to recouping investments. Bitcoin mining companies invest computing power to compete in the network; the higher their share of total hash rate, the more Bitcoin they typically earn.

However, this expansion relied on an uncontrollable condition: Bitcoin's price had to remain high enough to cover equipment and electricity costs. In 2022, this premise collapsed. Bitcoin prices dropped sharply from their peaks while energy costs such as natural gas continued rising. Mining companies had to keep paying electricity bills and equipment costs, yet the Bitcoin mined could no longer support previous expansion efforts. By year end, Core Scientific's cash reserves were depleted, and the former North American "mining king" filed for bankruptcy protection.

The pressure was industry-wide. Riot, another publicly traded mining company, mined 5,554 bitcoins in 2022—a 46% increase from the prior year—but its mining revenue decreased from $184 million to $157 million. The company reported a net loss of $509.6 million for the year, primarily due to impairments of mining equipment, bitcoins, and acquired assets.

Mining operates as an endless equipment race. As more mining machines join the network, the Bitcoin network automatically increases mining difficulty, reducing the amount of Bitcoin each individual machine can mine. Older machines are slower and consume more power, meaning the same electricity cost yields progressively lower returns. To maintain production and reduce electricity costs, mining companies must continuously purchase new miners with higher hash rates and lower energy consumption. This means paying not only electricity bills but repeatedly investing capital to upgrade equipment. When coin prices drop, both newly purchased miners and held bitcoins lose value while electricity costs remain unchanged.

The situation worsened after April 2024, when Bitcoin underwent its fourth halving, reducing the block reward from 6.25 BTC to 3.125 BTC. The term "halving" refers to the Bitcoin network reducing the block reward by 50% approximately every four years. This means that even with identical computing power, mining companies receive significantly fewer new bitcoins. Mining equipment, electricity, and labor costs do not decrease accordingly, while output is halved, further compressing profit margins. Meanwhile, the total network hash rate continues to grow, and the equipment race has not slowed down due to the halving.

Core Scientific was affected particularly severely. In 2024, the company's Bitcoin mining output decreased by 52% year-over-year. By 2025, annual production fell further from 6,595 to 2,276 bitcoins, and self-mining revenue dropped from $409 million to $229 million, nearly halving. The solution became clear: mining facilities could be repurposed. Even if no longer used for Bitcoin mining, the land and power infrastructure previously purchased by Core Scientific could support other power-intensive computing equipment—precisely what AI companies lack most.

This is where CoreWeave entered the picture. CoreWeave is now one of the most prominent AI cloud computing companies, currently using only GPUs from NVIDIA, which serves as its supplier, technology partner, and major shareholder with an additional $2 billion investment planned for 2026. CoreWeave leverages large-scale GPU clusters to provide the computing power that large model companies need for training and running models.

But before becoming a major buyer of NVIDIA GPUs, CoreWeave was also a mining company that mined Ethereum. In 2016, CoreWeave's founding team bought their first GPU and placed it on a pool table in their New York office, mining their first Ethereum block. One graphics card grew into hundreds, then into tens of thousands. After the crypto market entered a winter from 2018 to 2019, CoreWeave took advantage of numerous mining companies exiting the market to purchase GPUs and data center resources at low prices. Crucially, CoreWeave used GPUs to mine Ethereum. Unlike Bitcoin ASIC miners, which are designed for specific computations, GPUs have broader applications. After the crypto market cooled, CoreWeave began renting out these devices to visual effects companies and machine learning clients, gradually shifting from cryptocurrency mining to cloud computing services. This transition allowed CoreWeave to ride the wave of the generative AI boom, propelling it into the heart of the AI cloud computing market.

As more GPUs are purchased, however, a critical problem emerges: where should these machines be placed? High-end GPUs can only generate real computing power when connected to sufficient power and network resources. CoreWeave can buy more chips, but finding powered data centers in a short time is difficult. Core Scientific, meanwhile, has existing mining facilities, reliable electricity access, and grid connection eligibility—but needs new customers who can make long-term rental payments. Two companies with complementary needs had found each other.