Alpine Fox founder Mike Alfred stated at the Energy Investors Forum in Dallas that he does not consider the boom in artificial intelligence infrastructure to be a bubble, reports TheEnergyMag.
At the same time, the investor warned: for bitcoin miners, diversification into AI comes down to choosing a model—selling sites and power or additionally taking on the risk of purchasing GPUs.
Alfred expressed confidence that the demand for AI infrastructure will persist over a 20–30 year horizon, even if the market experiences corrections and economic downturns.
He described the key fork in the road for miners as a choice of operating model. The first option is to own GPUs at their own sites. This offers the chance of higher revenue and greater control but sharply raises the stakes: the operator needs to finance equipment purchases, account for the risk of rapid chip obsolescence, and ensure constant utilization of capacity.
"If you decide to own them, you must have a very good reason," emphasized the expert.
The second option is colocation. The client places servers and chips themselves, and the site owner sells space, electricity, water, and cooling for the "hardware." Alfred called this scheme more conservative and likened it to real estate. That is precisely why it is much easier to attract financing for it, he noted.
Alfred serves on the board of directors of IREN as a non-executive director. According to him, about 80% of Alpine Fox's investments are concentrated in IREN, as well as in Cipher Digital.
Both companies started as bitcoin miners and later moved into the AI sphere. Among recent deals in this direction are IREN's five-year agreement with Microsoft for approximately $9.7 billion to supply cloud infrastructure based on Nvidia GB300 chips, as well as Cipher's 10-year contract with Fluidstack for equipment placement and allocation of 168 MW of capacity at the Barber Lake site in Texas.
Most Bitcoin Mining Sites Are Not Ready for AI
A significant portion of active cryptocurrency mining facilities are difficult and expensive to convert into AI data centers. This conclusion was reached by participants in a thematic panel at the Energy Investors Forum.
The forum discussed the concept of a model where mining acts as a flexible load:
- temporarily monetizes unused electricity during the construction phase of an AI campus or absorbs excess generation;
- shuts off when power is needed for more expensive computations.
SATOKIE founder and CEO Jay Zapata noted that when competing for a suitable substation, cryptocurrency mining will always lose to AI in business attractiveness.
According to TheEnergyMag's observations, investors are increasingly evaluating miners through the lens of available power capacity and project portfolio, not just by bitcoin mining volume. As examples, the publication cited TeraWulf's July agreement with Anthropic for 20 years on a campus in Hawesville (Kentucky) with approximately 401 MW of critical IT capacity and expected revenue of about $19 billion, as well as Galaxy Digital's 15-year deal with CoreWeave for the Helios facility in West Texas at 133 MW and about $4.5 billion in revenue. However, both cases are closer to fully pivoting sites towards AI than to a sustainable coexistence of the two models.
Zapata estimated the cost of building a mining site at $200,000-300,000 per 1 MW versus over $10 million per MW for AI infrastructure excluding chips. According to him, an intermediate scheme only works with cheap electricity: at a price of $0.06-0.08 per kWh, the economics of mining become challenging, and it is more profitable for a developer to go straight into AI or simply hold onto power access without mining revenue.
Panelists emphasized that electricity for an AI project is just the starting point. Typical mining facilities are often located in remote areas, built to simplified standards, and have limited redundancy. AI tenants need redundant fiber-optic lines, backup generation, complex cooling, water, large plots for development, and service availability at about 99.999%.
Compass Mining's Director of Mining, Shannon Squires, stated that after evaluating many new sites for Tier 3 data centers, it became clear that the vast majority of them are unrealistic for this purpose. According to him, the intersection of the two models mostly ends at land, substations, and step-down transformers.
A more sustainable version of the hybrid scheme was named by the speakers as sites with their own generation. Upstream Data CEO Steve Barbour noted that gas power plants are often built with excess capacity, which can be taken by mining while the AI site uses the guaranteed 24/7 load. Then digital asset mining remains a secondary consumer and can be quickly shut down without violating service obligations.
At the same time, panel participants believed that expensive GPUs are less suitable for such flexibility, and long-term AI contracts carry concentration risk. According to the assessment of all four speakers, before the next Bitcoin halving in 2028, at least one major tenant, lender, or lessor in the AI segment could default or break a contract.
Recall that public bitcoin miners are increasingly converting power capacities and data centers into infrastructure for artificial intelligence, according to data from Stanford University.
end-content
