The math behind why public Bitcoin miners keep signing AI hosting deals instead of buying more mining rigs just got a lot more explicit. CoinShares estimates that AI compute now generates roughly $1.5 million in annualized profit per megawatt for operators who redirect their power capacity toward it, compared with about $500,000 per megawatt from traditional Bitcoin mining, a three-to-one gap that helps explain why the industry's capital has been flowing so decisively in one direction over the past year.

The gap isn't just about revenue, it's about how the two businesses are actually built. CoinShares' own mining report pegs the build cost for AI and high-performance computing infrastructure at roughly $8 million to $15 million per megawatt, versus $700,000 to $1 million per megawatt for conventional mining hardware, an 8-to-21x gap in upfront capital intensity. AI infrastructure contracts, though, typically come with multi-year terms and margins CoinShares puts above 85%, giving operators revenue visibility that mining, priced entirely off a volatile hash rate and an even more volatile bitcoin price, simply cannot offer.

Mining economics have deteriorated enough to make that trade-off an easy call for public operators. Hash price, the standard measure of daily mining revenue per unit of computing power, fell to roughly $28 to $30 per petahash per second per day by early this year, a new post-halving low, while the weighted average cash cost for listed miners to produce one bitcoin climbed to around $80,000. With margins that thin, some operators are cutting mining capacity outright rather than merely diversifying alongside it. Public miners have already cut 23% of network hashrate to chase AI cloud revenue instead, and CoinShares projects listed miners could derive as much as 70% of total revenue from AI by the end of this year, up from roughly 30% today.

The deals underpinning that shift are already sized at a scale that dwarfs anything mining alone could generate. Miners have signed over $70 billion in cumulative AI and HPC contracts across the sector over the past year or so, with individual operators now running hundreds of megawatts of hosting capacity backed by long-term customer commitments rather than spot hash price. That kind of contracted, multi-year revenue is precisely what a business exposed to a halving-driven hash price collapse, like the one that pushed some operators below cash breakeven as production costs climbed past $75,500 earlier this year, has never been able to offer its shareholders.

Related: Public Bitcoin Miners Cut 23% of Hashrate to Chase AI Cloud Revenue

The individual deals show how far this has already gone. Core Scientific has energized 350 megawatts for HPC hosting even as its mining capacity has stagnated, Hut 8 signed a $7 billion, 15-year lease with Fluidstack covering 245 megawatts, and TeraWulf has contracted 522 megawatts of capacity backed by $12.8 billion in long-term customer commitments while running just 39 megawatts of its own critical IT load. Those are the kinds of balance-sheet-altering, multi-year contracts that no amount of hash rate could generate on its own, and they explain why miners with access to cheap power and existing grid connections have become some of the most sought-after landlords in the AI buildout rather than simply commodity block producers.

None of this means mining itself is going away. CoinShares is careful to frame the shift as a reallocation of capital toward the more profitable use of the same power infrastructure, not an abandonment of the network, and the report notes that if mining profitability recovers meaningfully, some operators may reassess how they split capacity between the two businesses again. But for now, with AI contracts offering both higher per-megawatt returns and multi-year certainty that spot hash price can't match, the direction of travel for public miners' capital looks set to keep tilting toward compute rather than blocks.