What happened
Bernstein published a research note on Tuesday arguing that listed Bitcoin miners have become strategic suppliers to the AI infrastructure cycle, not bystanders. The bank put the combined planned power capacity across the listed miner cohort at 27 gigawatts and tallied roughly $90 billion in AI-related deals signed or in late-stage negotiation, per CoinTelegraph's writeup of the report.
The argument rests on a specific bottleneck. Hyperscalers can buy GPUs. They can't conjure substations, transmission upgrades, or interconnect queue slots that took five to seven years to file. Miners spent the 2020-2024 cycle filing those queue positions to chase cheap power for hashrate. Those same filings now look like the most valuable real estate in the AI supply chain.
Names cited in the note include Core Scientific, which inked a $3.5 billion expansion of its CoreWeave deal earlier this year, Iris Energy with its Childress, Texas campus, TeraWulf at the Lake Mariner site in upstate New York, and Cipher Mining. Bernstein's framing: these are no longer mining stocks with an AI option. They're power infrastructure stocks with a mining option.
Why it matters
The binding constraint on AI infrastructure has shifted. Through 2024 it was GPU supply. Through 2025 it became cooling and rack density. In 2026 it's the grid. The Department of Energy flagged in March that interconnect queue wait times in MISO and ERCOT now run six to eight years for new large loads. A miner with a 300 MW substation already energized in West Texas can hand a hyperscaler 18 months. Nobody else can.
That asymmetry is what Bernstein is pricing. The 27 GW figure isn't theoretical. It's the sum of signed power purchase agreements, owned substations, and active interconnect filings across the listed cohort. Even at a 40% conversion rate to actual delivered MW by 2028, that's more than 10 GW of supply moving to AI workloads, equivalent to the entire installed base of the largest US colocation operator three years ago.