Crypto miners pivot to AI datacentres

A handful of companies that once ran massive bitcoin farms are quietly becoming the hardware arm of the AI boom. (businessinsider.com) Those firms already had exactly two things that AI builders need most: access to huge, steady power contracts and buildings wired to handle megawatts of electricity. (businessinsider.com) Because training and serving large models burns electricity at scale, the fastest way to host AI racks is to reuse sites where miners once placed thousands of energy‑hungry ASICs. (coindesk.com) Companies such as Applied Digital and Cipher Digital turned those advantages into immediate capacity: together they control multiple gigawatts of energized capacity and ready land that hyperscalers can occupy faster than building new campuses. (businessinsider.com) Investors noticed. The market value of a group of former miners rose sharply from a few billion dollars in 2022 to many times that as the market priced the switch from coin production to selling AI compute. (businessinsider.com) The shift looks like this on the ground: operators rip out custom bitcoin rigs, install GPU clusters from vendors like NVIDIA, and sign contracts to host model training or inference workloads for cloud companies and AI startups. (coindesk.com) That path is cheaper and faster than building fresh sites because the hardest bottlenecks for hyperscale compute are not servers; they are power hookups, local lines, and the physical cooling infrastructure. (businessinsider.com) The cooling problem is also why some investors are looking offshore. Seawater is a massive, free heat sink, so putting data centers on the ocean floor or on barges can cut the portion of energy spent on air conditioning, which often consumes a third to nearly half of a center’s power. (investorplace.com) Underwater experiments have shown the idea can work: Microsoft’s Project Natick put a sealed module of servers beneath the North Sea and reported surprisingly low failure rates and good thermal performance during its test run. (subseacables.net) For engineers and students, the practical lesson is where industry dollars flow next. Power‑delivery planning, rack‑level cooling, low‑latency coastal networking, and ruggedized hardware design are suddenly central problems—more than raw algorithm speed for a moment. (businessinsider.com) That does not erase the need for classic software skills. Companies still hire engineers who can design distributed systems that tolerate hardware faults, optimize I/O for GPUs, and reduce data movement during training. But those systems run where electricity and cooling are cheapest, and former miners already control many of those places. (coindesk.com) One concrete signature of the trend: a tested subsea module once contained 864 servers inside a 40‑foot cylinder during Microsoft’s Natick trials—small compared with continent‑scale campuses, but proof that ocean cooling and sealed, maintenance‑focused designs can work. (subseacables.net)