Hyperscale AI data center buildouts have begun directly competing with Bitcoin miners for the same power and transmission infrastructure across several U.S. and Canadian jurisdictions. Texas's ERCOT queue is the most visible example, with more than 200 gigawatts of AI-related interconnection requests against roughly 5 to 10 gigawatts from mining operators. The dynamic has fundamentally reshaped power-contract negotiations in markets that miners had quietly considered their own, and the implications are reverberating through every public-miner equity story.
Several large miners have privately lost out on follow-on power deliveries that they had assumed were locked in, as utilities and independent power producers reallocated capacity to AI tenants offering longer-duration commitments and higher-margin contracts. One mid-size operator described being told by its utility, mid-renewal-discussion, that the previously preliminary terms were no longer available because a hyperscaler had presented a 15-year take-or-pay structure on the same megawatt block. The miner had been negotiating a 7-year deal at lower per-megawatt rates and had no way to match the duration commitment without taking on balance-sheet risk that public-market disclosures could not easily absorb.
The willingness-to-pay gap is structural. Inflexible AI training workloads cannot easily curtail without setbacks measured in compute-weeks, because each interruption requires re-loading multi-terabyte model checkpoints and resuming from a known-good state on multi-million-dollar GPU clusters. That inflexibility makes AI tenants willing to pay materially higher rates for predictable, firm power than miners typically can. In ERCOT in particular, the spread between what hyperscalers will pay and what mining operations can sustain has widened to roughly 30 to 50% over the past eighteen months, with the gap continuing to widen as AI capex commitments accelerate.
The transmission side of the equation is equally tight. New high-voltage transmission projects are taking five to seven years from initial filing to energization, and the marginal megawatt of available transmission capacity in concrete-busbar terms is now genuinely scarce in much of Texas, the Pacific Northwest, and the Mid-Atlantic. Utilities have started running their own internal triage models, and miners increasingly find themselves competing not on price-per-megawatt-hour alone but on how much transmission upgrade cost they are willing to accept and how flexible their contractual minimums are. The flexibility argument has helped at the margin, but it has not closed the willingness-to-pay gap.
The competition has forced the largest public miners to either accelerate their own AI/HPC pivots — as Riot, Iris Energy, and Cipher have done — or accept tighter geographic constraints on future expansion. The Wyoming, Alberta, and Paraguay grid markets have all seen an uptick in mining-operator interest specifically because they are not yet on the hyperscalers' geographic shortlist. For some private operators, the answer has been to retreat into stranded-resource niches like methane capture and remote hydroelectric pockets where AI hyperscalers cannot economically follow, since AI training workloads require predictable bulk power that stranded resources cannot reliably supply.
For the broader category, the implication is that the era in which mining could quietly accumulate cheap power slots is closing. The most attractive interconnections in the most attractive grids are now contested assets, and the contestants on the other side of the table have deeper pockets and longer time horizons. The miners that thrive will be those that either pivot into selling power to AI tenants at the higher clearing rate, or that build genuinely defensible cost structures in geographies the hyperscalers cannot reach. The next eighteen months will reveal which public miners belong to which group, and the equity-multiple implications of that sorting are likely to be substantial.