From Proof-of-Work to Proof-of-Power: Why Bitcoin Miners Are Becoming AI's Landlords

The consensus in early 2025 was tidy: Bitcoin miners were getting crushed. Halvings compressed margins; AI stole the narrative. Companies like Core Scientific, Iris Energy, and TeraWulf looked like candidates for a second wave of bankruptcies.

That reading missed the only thing that matters in compute infrastructure right now — not the machines, but the meter.

Connecting a new data center to the U.S. electrical grid takes three to five years. Not because construction is slow — because the interconnect queue (the waiting list to plug into the high-voltage transmission network) is saturated through 2027-2028 in every major grid region: ERCOT in Texas, PJM along the Eastern corridor, MISO in the Midwest. A hyperscaler (a cloud-at-scale operator like AWS, Microsoft, or Google) can announce a campus today, close financing, break ground, and still not power a single GPU rack before late 2028.

Bitcoin miners built their facilities in the 2020-2022 cycle, when grid access was cheap and open. That interconnect is already paid for, already energized, and it does not expire.

Most miners operate under PPAs — power purchase agreements, long-term contracts that lock in an electricity price and capacity for ten to twenty years. A PPA is not a spot purchase. It guarantees a fixed cost of power regardless of market volatility. In a world where AI training runs consume gigawatts around the clock, that predictability is worth almost as much as the power itself.

The conversion arithmetic is direct. A facility running Bitcoin ASICs at $0.04/kWh can often generate higher gross margins hosting NVIDIA H100 clusters for an AI lab — without permitting a single new substation or waiting in line for grid interconnection.

Core Scientific signed a twelve-year HPC (high-performance compute) hosting contract with CoreWeave for approximately 200 MW of capacity — roughly the electricity draw of 160,000 U.S. homes, locked in under a single long-term agreement. Iris Energy decommissioned mining rigs across multiple sites and brought GPU clusters online within one to two quarters of the decision. TeraWulf announced a dedicated HPC wing at its Lake Mariner campus in upstate New York.

These are not pivots of desperation. They are arbitrage: the same megawatt that generated X dollars in Bitcoin revenue now generates Y dollars in AI compute revenue — with Y > X and Y secured under a multi-year contract.

What hyperscalers are paying for isn't square footage or cooling efficiency. It's the right to consume power now, not in 2028. Miners operating in deregulated markets like ERCOT carry an additional layer: demand response programs that let them sell capacity back to the grid when spot prices spike. That flexibility commands a direct premium from operators who need reliable baseload compute.

The absorption of ex-mining infrastructure into AI compute supply compresses lead times structurally. It doesn't eliminate greenfield construction — hyperscalers will still build campuses. But it creates a parallel supply channel with a twelve to twenty-four month structural advantage over any new site that hasn't already secured interconnection.

For investors, the pricing lens shifts. An ex-mining company is no longer valued on Bitcoin hash rate economics. It's valued on MW under contract, lease duration, and counterparty quality — the same metrics applied to data center REITs.

The assets didn't change. The demand curve moved.