Core Scientific and TeraWulf flipped their revenue mix in Q2 2026. The wider public cohort cut mining capacity 13% over six months as data center tenants absorbed the same power and buildings.
Two US-listed Bitcoin miners now earn more from renting out their data centers to AI and high-performance computing (HPC) tenants than they do from mining Bitcoin. The wider public cohort is cutting its mining capacity faster than the Bitcoin network itself.
Core Scientific, the largest of the group, booked $136.7 million in colocation revenue (renting space, power, and cooling to outside tenants) in Q2 2026, against $27.5 million from Bitcoin mining, according to Cointelegraph's report on BlocksBridge Consulting's Miner Weekly. TeraWulf, a smaller peer, reported $31.9 million in HPC lease revenue versus $12.8 million from mining in the same period. Both companies are still legally structured as Bitcoin miners. Their P&Ls no longer are.
That is the cleanest evidence yet that the reallocation is no longer a forward-looking thesis. It is in the quarterlies. The same power purchase agreements, substation builds, fiber drops, and operations benches that the industry built to secure Bitcoin are now being rented to AI tenants. Miners discovered, in the middle of a brutal margin cycle, that the bottleneck-relieving supply for the AI buildout is the infrastructure they already own.
BlocksBridge's data: the realized hashrate of public miners (the actual computing power pointed at the network, not the nameplate capacity of their machines) fell 13.4% over the trailing six months, from 368.3 exahashes per second in Q4 2025 to 319 EH/s in Q2 2026. Strip out Bitdeer, the contrarian, and the remaining cohort cut 21.2% over the same period. The wider Bitcoin network's average hashrate fell 10.6%. Public miners are shrinking their mining footprint faster than the network itself.
That gap is the mechanism. When a cohort cuts faster than its own network, the machines are not being shut down for lack of power. They are being switched to a higher-paying use of that power.
The transition is not uniform. Riot Platforms and Bitdeer still earn the vast majority of their revenue from Bitcoin mining in Q2 2026, even as both have signaled capacity for HPC tenants. Hut 8 and CleanSpark sit in between, with material AI or HPC contracts ramping but not yet dominant. The cohort is splitting into two trades: sell the megawatt, or mine the coin.
Bitdeer is the natural test of which trade wins. Its realized hashrate rose 44% to 63 EH/s over the same six-month window. Every other major public miner cut. The bet is that mining margins recover before the AI colocation pipeline fills. If Bitdeer's cost per coin keeps falling through the next halving, its contrarian bet will look prescient. If the AI pipeline keeps absorbing capacity at the pace the Q2 numbers suggest, the cohort has decided the question for it.
BlocksBridge frames the pullback as the unwinding of the post-2021 North American expansion cycle, one halving later. The 2021 China mining ban pushed a generation of capital into Texas, Georgia, and the Pacific Northwest. Five years on, the cheaper power contracts are spoken for, the transmission upgrades are built, and the operating teams are trained. The marginal AI tenant is now bidding against the marginal Bitcoin miner for the same buildings. Marathon Digital booked a $611 million loss in Q2 2026 as mining economics compressed; the AI colocation line, where present, is the offset.
The Q3 earnings will show which trade won. Bitdeer's hashrate growth and all-in mining cost are the markers. If it keeps expanding, the rotation is not over. If it pauses, the rest of the cohort has decided the question.