Get your free copy of Editor’s Digest
FT editor Roula Khalaf picks her favourite stories in this weekly newsletter.
Bitcoin miners are rushing to sign deals with artificial intelligence developers in a bid to revive sagging revenue by finding new customers for their massive data centers.
Cryptocurrency miners run powerful computing facilities, often spanning acres of land, where they solve complex mathematical puzzles to verify transactions and generate digital coins, but many are struggling to turn a profit due to high energy and computing costs and a recent halving of mining rewards.
Now they hope to profit from a surge in demand for powerful but rare chips used for both cryptocurrency mining and AI processing, known as graphics processing units, or GPUs.Tech companies are competing for access to chipmaker Nvidia’s GPUs in an effort to build more powerful AI systems, and are increasingly signing deals to allow miners to use their chips or install their own in their data centers.
Core Scientific, one of the world’s largest bitcoin miners, is “aggressively pursuing” AI-related deals, Chief Executive Adam Sullivan told the Financial Times. “This is a critical part of our business,” he added.
The Nasdaq-listed miner, which has data centers in Texas, North Carolina and Georgia, signed a deal with AI cloud provider CoreWeave last month, which the two companies estimate will bring in $4.7 billion in revenue over 12 years. Nvidia-backed CoreWeave, a former cryptocurrency miner that pivoted to AI several years ago and saw its valuation jump to $19 billion in May, will use Core Scientific’s data centers to host its AI chips.
AI companies require large amounts of energy and computing infrastructure, two things that bitcoin miners typically have access to, and AI Group believes that tapping into the miners’ high-performance computing (HPC) data centers will be faster and cheaper than building its own.
Big tech companies including Microsoft, Google and Amazon have said they plan to spend tens of billions of dollars developing data center infrastructure to support their AI ambitions. Demand for AI capabilities is also spurring investor interest in new cloud startups such as CoreWeave and Lambda Labs, which focus on renting access to GPUs.
“that [normally] “Building an HPC-grade data center from the ground up takes three to five years,” JPMorgan analysts wrote in a recent note, adding that this time frame is getting longer due to increased demand for AI projects.
“This race for power will give an advantage to companies that currently have access to cheap electricity,” they added.
Other major bitcoin miners are also using some of their data and processing power for AI.
Coatue Management, a US hedge fund founded by Tiger Cub fund manager Philippe Lafont, recently invested $150 million in bitcoin miner Hut8 to help it upgrade its infrastructure to meet the needs of AI companies. The mining company also recently launched a new AI division.
Hut8 chief executive Asher Genault said the company, named after the building at Bletchley Park where mathematician Alan Turing worked during World War II, was focused on “the enormous demand and growth in the data center sector, driven primarily by rising demand for AI.”
Bitcoin miners hope that by shifting their strategies to AI, they will see higher and more stable profits.
In 2022, following the collapse of cryptocurrency exchange FTX and Bitcoin’s price plummeting below $16,000, many miners, including Core Scientific, went bankrupt.
Cryptocurrency prices have soared since then, with Bitcoin hitting an all-time high of more than $73,800 in March and currently trading at around $63,800, but the financial rewards for mining new blocks of Bitcoin have diminished due to the cryptocurrency’s quadrennial halving in April. High energy and technology costs are also hitting profitability.
Canadian miner Hive is also focused on “growing revenue from the suite of Nvidia GPU chips that power data services for the AI revolution,” it said, while New York-based Bit Digital inked a $275 million, three-year deal in January to lease data center space to companies building large-scale language models.
“We understood the halving was coming, but with profit margins shrinking by 50% overnight, we felt it didn’t make sense and was poor business practice to rely on the expectation that Bitcoin would rise,” BitDigital CEO Sam Tabar said.
“We’re just lending out computing power to people who are building AI models, we’re just handling the hardware part of it,” he added.
But the race to build new data centers is straining power grids in some parts of the world, given HPC’s massive electricity demands. Bitcoin mining is also very energy intensive, and both sectors have been criticized for using up huge amounts of electricity.
According to data from the University of Cambridge, Google’s greenhouse gas emissions have soared 48% over the past five years due to the expansion of its data centers for AI processing, and mining bitcoin uses more energy each year than Pakistan or Ukraine.
