AI's power crunch turns Bitcoin miners' grid into an asset

AI's power crunch turns Bitcoin miners' grid access into a strategic asset because hyperscalers need permitted, energized power sites faster than they can build them. Miners spent a decade securing substations, interconnection rights and contracts in Texas and the Gulf Coast, and are now signing multibillion-dollar AI leases—though converting campuses into liquid-cooled data center revenue is far from plug-and-play.

Key Takeaways

• AI data center power capacity reached about 29.6 GW by end of 2025, per Stanford University.
• Listed miners have announced more than $70 billion in AI and HPC contracts, CoinShares reports.
• AI-grade infrastructure costs $8M–$15M per MW versus $700K–$1M for mining sites.
• Miners with HPC contracts trade at 12.3x 12-month revenue vs 5.9x for pure-play operators.
• Many contracted sites, including Hut 8's River Bend, will not commission until 2027.

Why is AI's electricity demand still climbing despite cheaper GPUs?

By the end of 2025, artificial intelligence data centers worldwide tied up about 29.6 gigawatts of power—enough to run all of New York state at peak demand, according to Stanford University's annual AI industry report released in April. GPU computation costs have fallen more than 99% since 2006, yet efficiency gains feed larger models instead of cutting total draw.

Stanford estimates the most demanding training runs can pull upward of 100 megawatts, while AI-dedicated capacity has risen roughly 200-fold in three years from under a gigawatt in 2022. Chips can be delivered in months, but energizing a site—with substation work, interconnection approval and cooling—takes years. The United States hosts 5,427 data centers, more than 10 times any other country.

What does Bitcoin mining infrastructure offer AI developers?

Mining ASICs cannot train or run inference for AI labs. What transfers is everything around the chips: energized sites, power contracts, grid hookups and cooling shells. A miner with an existing grid connection can rent capacity that would take hyperscalers years to permit from scratch.

Miners cluster in cheap-power US states such as Texas and the Gulf Coast—exactly where AI wants to build. Through 2024, AI's cumulative all-in power demand reached an estimated 9.4 gigawatts, close to Switzerland's national electricity use and about half Bitcoin mining's estimated draw, per Stanford.

Which deals show miners' grid access is already priced?

Mining economics tightened the case for leasing. JPMorgan estimated Bitcoin's all-in production cost at about $78,000 per coin, well above market prices near $53,400 in late June, with Cointelegraph reporting hashprice below breakeven for roughly 20% of the industry.

In November 2025, Iren signed a five-year, $9.7 billion GPU cloud deal with Microsoft from its 750-megawatt Childress, Texas campus. Hut 8 signed a 15-year, $7 billion lease with Fluidstack for 245 megawatts at River Bend, Louisiana, backstopped by Google. TeraWulf reported $12.8 billion in contracted HPC revenue, and Core Scientific expanded its CoreWeave agreement to $10.2 billion over 12 years. CoinShares counts more than $70 billion in announced AI and HPC contracts across listed miners.

Why isn't the pivot as simple as swapping out mining rigs?

CoinShares estimates mining infrastructure costs $700,000 to $1 million per megawatt, while AI-grade liquid-cooled builds run $8 million to $15 million. Hyperscalers demand power density, redundancy and uptime guarantees many mining sites never needed. Iren disclosed about $3.75 billion in convertible note debt in March and raised another $3 billion in May.

Investors have rewarded the shift anyway. Hut 8 stock jumped about 20% in premarket trading when its lease was announced, Reuters reported. CoinShares projects listed miners could derive as much as 70% of revenue from AI by end of 2026, up from roughly 30% in Q1. For more on how crypto infrastructure is reshaping finance, see our Fintech & Crypto Alerts coverage.

If AI demand cools or projects slip, miners that tore out ASICs may have fewer options. Signing multibillion-dollar contracts is one thing; delivering the earnings investors expect is another.