Bitcoin miners’ AI pivot deepens as hashrate slides

By April, the economics were already breaking against operators, and CoinDesk Podcast Network’s interview with Luxor showed the damage was persisting rather than self-correcting. By April 2026, Luxor reported the network hash-rate had fallen below a major threshold for an extended period, consistent with a large realized-hash-rate contraction. Luxor said “we are now four difficulty epochs in to being below the one zeta hash threshold for network hash rate… we’ve now been below that level for coming up to 50 days,” while also describing hash rate as “if not plateauing, actually falling off the map,” evidence that weaker conditions were translating into sustained capacity withdrawal, not a temporary pause in deployment.

What turned that weakness into a June-to-September contraction was that marginal mining no longer paid: in the same CoinDesk Podcast Network discussion, the possibility of miners moving equipment was raised directly — “Do you think some of this might be miners moving equipment to lower cost jurisd…” — and Luxor’s answer was blunt, “Marginal hash rate came offline at $30. Hash price has not returned even with hash price now near $39.” Luxor’s April 2026 analysis thus linked deteriorating mining economics directly to hash rate weakness, including hash rate “falling off the map,” consistent with miners reducing capacity rather than scaling up. Even as “some of the publicly traded mining companies caught this bid” and Hash Rate Index’s mining stock index rose 27% in April, Luxor said “we continue to see a decline in average network” difficulty, showing market relief was not enough to bring rigs back.

Corey Trinetti said gigawatt-scale power is a negotiation with utilities built around three key contractual terms, including minimum bills and exit fees or collateral, with commitments reaching hundreds of millions of dollars annually. In one Oracle example, the minimum bill was about 336 million a year, and the exit fee amortizes until about 2047, which makes AI and HPC the preferred destination for scarce electricity and capital.

Once those contracts are signed, the power is no longer readily available to swing back into Bitcoin mining, so the hashrate weakness looks structural rather than cyclical. The August 22 analysis said miners are locked in and their power is locked into big delivered AI HPC builds, making reversal nearly impossible; it also noted resistance near 950 exahashes, a secular network hash rate decline since November 9, 2025, losses of 50 exahash a month, and a smooth average at 915 exahash. The August 27 economics work called inference the clear winner on net revenue per megawatt-hour and projected that by 2030 data output could exceed 50% of data center demand, while Iren’s Microsoft deal showed the shift in practice with roughly 200 MW of IT load supplied by Nvidia GB300 NVL72 GPU systems.

By June 2026, the behavior of listed miners showed the pivot was already operational, not rhetorical. The Wolf Den argued that “the biggest publicly traded Bitcoin miners are not pivoting to survive the squeeze. They are pivoting out of the business,” and IREN was the clearest case: formerly Iris Energy, it had already signed a five-year, $9.7 billion AI cloud contract with Microsoft and, as The Wolf Den put it, “IREN currently holds zero bitcoin,” underscoring that its identity had shifted from miner to AI infrastructure operator.

Other public miners were making similarly concrete moves inside their balance sheets and operating plans. Core Scientific sold $175 million of Bitcoin in March, cutting holdings from 2,537 coins to roughly 600 to fund its transition; Hut 8 told investors on its fourth-quarter earnings call that “bitcoin is no longer a long-term strategic focus”; MARA Holdings softened its stance and now sells newly mined bitcoin opportunistically, with roughly twenty eight percent of its holdings either loaned or pledged; CleanSpark said on its Q1 2026 call that mining investment no longer made sense versus AI infrastructure; and across Core Scientific, Bitdeer, Riot Platforms, and Bitfarms alone, public miners have sold roughly fifteen thousand bitcoin from their treasury holdings since the start of the year, with the proceeds going almost entirely toward financing the pivot.

The contraction was visible first in the public miners’ own output: WEEX reported realized hashrate for Q2 2026 at 319.0 EH/s, down from 344.4 EH/s in Q1 and 368.3 EH/s in Q4 2025, a sharp sequential and multi-quarter retreat rather than normal noise. Excluding Bitdeer, the decline was 21.2%, and WEEX described the drop as “a weakening of mining competitiveness,” adding that some companies had turned off equipment or reallocated power to AI and high-performance computing infrastructure.

The weakness also lasted beyond quarter-end, which is what made it matter at the network level: mid-summer operating changes showed the pullback was still underway, with Core Scientific saying that as of mid-July its power capacity for leasing customers was about 1.1 GW. By early September, CryptoNews.net was blunt enough to run the headline “Difficulty Rises, Hashprice Rips 22% as Bitcoin Hashrate Stalls,” and by mid-September TipRanks, citing a KuCoin LinkedIn post, said Bitcoin’s network hashrate had remained below its late-2025 peak for more than 300 days.