Fred Thiel, CEO of MARA Holdings, one of the world’s largest publicly traded Bitcoin mining companies, has made a stark admission: the same megawatt-hour now generates dramatically higher returns when allocated to artificial intelligence workloads rather than maintaining Bitcoin hash rate. In a July 23 interview with journalist Natalie Brunell, Thiel said, “You get a lot more money per electron if you're doing it for AI than for Bitcoin mining.” The remarks, highlighted in an original report, confirm a reshaping of miner balance sheets that has been accelerating for months.
The profitability gap is punishing. AI computing—particularly GPU-driven inferencing and training—can produce multiples more revenue per kilowatt-hour than Bitcoin’s current block reward plus fees. Even with Bitcoin trading around $64,000, revenue per terahash is being squeezed by rising difficulty and intense competition. In contrast, an Nvidia H100 cluster leased to an AI startup or hyperscaler yields a predictable, high-margin income stream that does not depend on crypto price swings.
MARA is moving toward AI data center operations while continuing to mine Bitcoin only where energy is near-free or wasted, such as curtailed renewable generation or flared gas. This dual strategy leaves pure-play mining exposure looking increasingly untenable for publicly listed firms that must answer to shareholders each quarter. Other major miners are following suit: Core Scientific, once a Bitcoin mining giant, now generates 83% of its revenue from AI-related colocation services ($136.7 million in Q2, up from $10.6 million a year earlier), with a 59% gross margin on that business, while its self-mining operation produced a negative margin. Hut 8 and other firms have also inked AI hosting deals. CoinShares estimates that public miners have announced more than $70 billion in cumulative AI and high-performance computing contracts, and that listed Bitcoin miners could derive up to 70% of their revenue from AI by end-2026.
Yet the pivot is not without risk. André Dragosch, head of research at Bitwise Europe, told CryptoSlate that miners may be making the shift at the wrong point in the cycle. He argues that AI compute demand—including demand from autonomous agents—could take longer to materialize than current investment implies, while Bitcoin is approaching the end of its downturn. Bitcoin is trading roughly 50% below its October peak, and miner daily revenue has fallen nearly 40% year over year to about $28.5 million (30-day average), with hashprice dipping to record-low territory near $30 per petahash per second per day. If Bitcoin’s price recovers and mining profitability improves, miners who committed billions to long-term AI contracts could regret the capital outlay and lost optionality.
The transition is expensive. CoinShares estimates Bitcoin mining infrastructure costs $700,000–$1 million per megawatt, while AI-grade facilities run $8–$15 million per megawatt. That deeper, longer-duration capital commitment is being made during one of the largest technology infrastructure spending cycles on record; the Bank for International Settlements warns that the five largest hyperscalers may spend over $1 trillion on AI capex in 2025–2026, raising the risk of overinvestment. Additionally, AI data centers draw continuous, massive power loads that strain grids differently than interruptible Bitcoin mining, potentially drawing regulatory and community pushback.
Thiel’s candor essentially tells the market that MARA’s growth will rely less on Bitcoin price appreciation and more on executing as an energy and compute infrastructure company. The shift could decelerate Bitcoin’s total hash rate, easing difficulty for remaining miners, but also threatens to strand assets if AI demand cools. Miners are betting the AI wave outlasts immediate halving pressure, but the correlation risk remains underexplored.
