Not Every Bitcoin Mine Can Become an AI Training Campus

At the Energy Investors Forum in Dallas on July 23, generators, utilities, data-center developers, miners and investors outlined the physical and nontechnical barriers that can prevent bitcoin mining sites from converting into large-scale AI training campuses.

Panelists identified key infrastructure limits: constrained transmission, long interconnection queues, substation capacity, high-density cooling needs, fiber connectivity, equipment lead times and permitting timelines. They said owning a grid connection is only an initial step; sites also require redundant power, advanced cooling and networking to meet AI tenant demands.

Alexander Neumüller of the Cambridge Centre for Alternative Finance presented preliminary survey results covering more than half of global bitcoin mining activity. The findings estimated annual mining electricity consumption rose to about 190 terawatt-hours between June 2024 and December 2025, up from roughly 138 TWh. Estimated emissions increased to about 48 million metric tons CO2e from about 40 million. The low-carbon share of mining’s electricity mix was reported at 59.4%, compared with 52.4% previously. The presentation also found roughly 10% of respondents had already allocated some power to AI or accelerated computing, more than 40% were exploring AI or high-performance-computing diversification, and nearly nine in ten expected AI/HPC diversification to become an industry theme.

Speakers described different commercial approaches. Mike Alfred, founder of Alpine Fox, said miners that secured land and electricity when those assets were undervalued can pursue contracted data-center revenue. He identified two primary business models: owning GPUs, which may yield higher returns but carries greater operational and market risk, or providing colocation, where tenants supply servers and the operator supplies power, cooling and connectivity. Alfred said the colocation model resembles a real-estate business and can be easier to finance. He also stated, “Texas is the most important data-center market in the world.”

John Belizaire, chief executive of Soluna, outlined a strategy that pairs compute facilities with renewable generation whose output is frequently curtailed. He said modifying an existing interconnection to add data-center load can be faster than securing a new grid connection. Several participants noted that clusters of 10–20 megawatts could sometimes connect faster, require less upfront capital and provide geographic redundancy compared with single, gigawatt-scale campuses.

Speakers discussed the so-called mullet approach: using bitcoin mining to monetize power while preparing a site for AI. They noted limits. Mining can tolerate interruptions and lower connectivity; AI training typically needs firm power, strict service-level agreements, greater redundancy, heavy cooling and plentiful fiber. Only a subset of mining sites, the panelists said, will meet those higher technical standards.

Community acceptance emerged as a nontechnical constraint. Curtis Harris of Compass Mining urged developers to engage residents early, describing one outreach effort that stressed curtailment on request, no water use for mining rigs and hiring local contractors. Texas state Representative Jared Patterson told attendees local consent cannot be replaced by state-level lobbying and urged clear explanations of tax revenue, school funding impacts, grid relationships and water plans.

Panelists said some power-and-land assets may convert to hyperscale AI campuses, others to distributed compute or flexible mining, and some to sale or lease. They recommended that operators classify which megawatts should remain in mining, which can support flexible compute, and which meet the higher standards required for AI training campuses.