Why Public Bitcoin Miners Are Pivoting to AI—and Why Many May Not Return

Public Bitcoin miners are increasingly treating themselves as owners of scarce power-and-data-centre infrastructure rather than purely Bitcoin producers. CoinShares’ Q2 2026 mining report argues that AI and high-performance computing (HPC) can offer higher, more contracted returns per megawatt—and that long-duration customer commitments make much of the shift difficult to reverse.34

The core economic case: more profit per megawatt

CoinShares estimates annual profit of roughly $1.5 million per MW for AI/HPC infrastructure, compared with about $500,000 per MW for Bitcoin mining. That approximate threefold gap gives operators a strong incentive to direct available electricity and site capacity toward compute tenants rather than ASICs.6

The comparison became especially stark during Q2 2026. Bitcoin finished the quarter at $58,400, while the weighted average ex-tax cash cost of producing one Bitcoin among listed miners was about $75,500. On that aggregate measure, the sector was below cash breakeven.3

Mining revenue conditions were also weak: June’s average hash price—a measure of expected mining revenue per unit of hash rate—hit a record low of $27.70 per PH/s/day. A later recovery to around $38 as Bitcoin approached $77,000 can improve the outlook for operators that still have uncommitted mining capacity. But it does not automatically outweigh contracted AI/HPC economics for capacity already assigned to customers.34

Why the transition is described as structurally irreversible

The strategic asset is no longer just an ASIC fleet. It is permitted, grid-connected power capacity and the ability to operate a data centre at that location. CoinShares points to the scarcity and difficulty of reproducing energised sites as a central reason those assets have become so valuable.3

For a miner that has retrofitted a facility and signed a long-term AI or HPC lease, reversing course is not simply a matter of switching machines back on. It can involve giving up recurring contracted revenue, undoing physical conversion work and potentially breaking or buying out commitments. Long-duration arrangements are already visible across the sector: CoinShares has highlighted 12-year hosting agreements at Core Scientific and a 15-year lease at Hut 8, among other compute deals.8

Core Scientific illustrates the cost of reallocating capacity. According to CoinShares’ report, it paid $41.9 million to cancel deliveries for roughly 15 EH/s of next-generation mining hardware. The decision suggests that the company viewed AI capacity as more valuable than expanding an efficient mining fleet at those sites.34

That is the meaning of “structurally irreversible” in this context. Bitcoin price appreciation may raise mining margins, but it cannot easily make contracted power capacity available again. CoinShares therefore concludes that a BTC recovery is unlikely, by itself, to reverse the broader AI transition.4

The AI pivot has substantial upfront risk

Higher potential profit per MW does not mean the conversion is immediately lucrative. Building AI-ready data-centre capacity requires far more capital than conventional mining infrastructure.

Nine comparable miners spent $5.11 billion on capital assets in the first half of 2026 while reporting $341.2 million in directly reported AI/HPC revenue—a roughly 15-to-1 capex-to-revenue gap. Their combined AI/HPC revenue reached $205.8 million in Q2, up 52% quarter over quarter, but the figures show that the buildout remains capital-intensive and execution-dependent.1

In other words, the pivot rests on the prospect of long-lived contracted cash flows, not on a quick conversion of spending into revenue. Operators must fund construction, deliver facilities on schedule and sustain the performance required by compute tenants.

Which miners are exiting, and what it could mean for Bitcoin’s network

CoinShares identifies a growing separation between miners converting power capacity to compute and companies retaining mining-first flexibility.

• Keel ceased mining on June 29 and was expected to have no mining revenue in the third quarter.3
• IREN was identified as planning to complete its mining exit in 2026, while Cipher Digital was expected to exit in 2027.3
• TeraWulf is reducing mining capacity by 145 MW as it reallocates infrastructure toward higher-value compute uses.
• CoinShares estimates that at least 35 EH/s could leave publicly listed miners—about 4.7% of a 750 EH/s Bitcoin network.36

If hash rate leaves the network, competition for block rewards can decline, improving the relative position of miners that remain online, all else equal. But that network effect does not make a megawatt already leased to an AI customer easy—or necessarily rational—to reclaim for mining.

The miners that could still expand Bitcoin operations

CoinShares identifies Riot Platforms, MARA Holdings, HIVE Digital Technologies and Bitdeer Technologies as operators that may still expand mining if Bitcoin prices rise. Their advantage is flexibility: they retain greater mining exposure and, in some cases, capacity that is not yet tied up in long-term AI/HPC commitments.4

That distinction is more important than a single Bitcoin price move. A company with uncontracted power can compare current mining returns with available alternatives and react quickly. A company that has signed a multi-year compute lease has made a much more durable allocation decision.

The new dividing line for public miners

CoinShares’ report points to a bifurcated industry. Bitcoin strength can reward mining-focused companies with available capacity, while converted operators increasingly resemble data-centre businesses whose economics depend on power access, buildout execution and long-term tenant contracts.

The key question is therefore not only whether Bitcoin rises. It is whether a miner still controls economically reversible, uncontracted capacity. For much of the sector, long-term AI/HPC agreements, retrofit costs and scarce grid access make compute infrastructure a strategic commitment rather than a temporary hedge against weak mining margins.34