Bitcoin miners face shifting economics as power value to AI data-center operators exceeds cryptocurrency mining returns; Core Scientific and others reposition capacity.
Bitcoin ended June at $58,400 against average ex-tax cash production costs of roughly $75,500 per BTC across the listed sector. Yet rather than simply installing more efficient mining machines, the industry is increasingly reallocating power. AI developers require enormous quantities of electricity, and miners already control sites connected to grids where new data centers can wait years for access.
Bitcoin mining economics deteriorated sharply in the quarter. June hashprice reached a record low of $27.70 per PH/s/day, while transaction fees contributed less than 1% of block rewards. The response from some operators goes beyond slowing expansion.
Core Scientific paid $41.9 million to cancel approximately 15 EH/s of next-generation mining equipment. Its remaining self-mining operation posted a negative 56% gross margin, with some machines continuing to operate partly because power obligations remained during the conversion of sites toward AI infrastructure. Keel stopped mining entirely on June 29. Cipher does not plan additional mining capex and expects mining to become immaterial by 2030. CoinShares estimates at least 35 EH/s will leave the listed-miner cohort as announced transitions proceed.
IREN provides perhaps the clearest indication of where the economics are heading. Its quarterly AI cloud revenue reached $70.5 million, exceeding $66.7 million from mining, while 71% of TeraWulf's Q2 revenue came from HPC leases. These companies are no longer merely diversifying away from Bitcoin. They are deciding whether an available megawatt produces a better return connected to ASICs or AI servers.
CoinShares estimates that current AI economics can generate approximately $1.5 million of annualized profit per MW, compared with about $500,000 from Bitcoin mining. That comparison reshapes how mining infrastructure can be valued. Instead of asking only how efficiently a company converts electricity into BTC, investors increasingly must ask what alternative return that electricity could generate.
The headline advantage for AI comes with a substantial qualification. Building or converting AI-grade infrastructure can require $8 million to $15 million per MW, compared with roughly $700,000 to $1 million for mining infrastructure. AI facilities require different cooling, networking, redundancy and reliability standards. A mining campus cannot simply replace ASICs with GPUs and capture three times the profit.
The strongest part of the AI thesis is not computing hardware. It is electricity. Core Scientific's colocation revenue rose sharply as capacity entered service, but Q2 capital expenditure simultaneously approached $798 million. For investors, a headline backlog figure should therefore carry less weight than the speed at which contracted megawatts become billing megawatts.
AI currently offers superior economics per MW, but mining retains something that long-term hosting contracts can sacrifice: direct operating leverage to Bitcoin. Hashprice recovered from its June low toward $38 as BTC rebounded toward $77,000, bringing most operators back above cash breakeven according to CoinShares. A further Bitcoin rally could materially improve mining profitability without requiring years of construction or billions of dollars in conversion spending. Miners locked into long-duration AI contracts cannot simply reclaim that electricity when BTC economics improve. That makes flexibility valuable.
A company that converts every available megawatt to AI may secure predictable contracted cash flows but surrenders upside during another highly profitable mining cycle. A pure miner retains that upside but remains exposed to Bitcoin prices, network difficulty and increasingly thin transaction-fee revenue. The strongest infrastructure portfolios may ultimately be those capable of allocating capital selectively rather than making an all-or-nothing bet on either industry.
Hashrate, joules per terahash, power costs and BTC production still matter. But they no longer tell investors enough about companies turning themselves into digital-infrastructure operators. The additional metrics are energized MW, contracted MW and billing MW. Energized capacity measures the scarce infrastructure already secured. Contracted capacity indicates potential demand. Billing capacity shows how much of that opportunity has actually become revenue.
The gap between contracted and billing megawatts may be the most revealing number over the next several quarters. It separates AI ambition from AI execution at a time when equity valuations already assume substantial success. CoinShares expects AI/HPC revenue to accelerate through the second half of 2026 as additional capacity starts billing. Financing will matter just as much: miners are committing billions of dollars to infrastructure capable of supporting those contracts.
Bitcoin mining is not being simply replaced by AI. The industry's scarce resource is being repriced. Electricity once mattered because it could produce Bitcoin cheaply. In an increasingly power-constrained data-center market, having that electricity available in the right location, with a viable route to AI-grade infrastructure and a customer willing to pay for it, may be considerably more valuable.