At the recent Energy Investors Forum in Dallas, leading industry experts agreed: the boom around artificial intelligence infrastructure is not a bubble, but a long-term trend. However, for Bitcoin miners considering diversification into AI, choosing the right strategy is critical. The founder of Alpine Fox noted that demand for AI infrastructure will persist for the next 20–30 years, despite possible market corrections.
The main dilemma for miners is choosing an operating model. The first option is purchasing their own GPUs. This promises higher margins and control, but comes with serious risks: the need to finance expensive equipment, rapid chip obsolescence, and the obligation to maintain constant capacity utilization. As the expert emphasized, the decision to own GPUs must be economically justified.
The second, more conservative path is colocation. In this model, the miner acts as an infrastructure provider: selling space, electricity, cooling, and water, while the client hosts its own hardware. According to analysts, this approach resembles the real estate market and significantly simplifies fundraising. This is the path taken by companies like IREN and Cipher Digital, which started as Bitcoin miners and successfully pivoted to AI services. A prime example is IREN's five-year contract with Microsoft, worth approximately $9.7 billion, to supply cloud infrastructure based on Nvidia GB300 chips.
Are Existing Sites Ready for AI?
The reality is that most current mining facilities are not suitable for high-performance AI data centers. Retrofitting is expensive and complex. Forum participants suggested viewing mining as a flexible load: it can temporarily monetize excess electricity during the construction phase of an AI campus and shut down when power is needed for more expensive computations.
However, as noted by SATOKIE founder Jay Zapata, when competing for a suitable substation, mining always loses to AI in terms of business appeal. Building a mining site costs $200,000–300,000 per 1 MW, while AI infrastructure costs over $10 million per 1 MW, excluding chips. An intermediate scheme is only viable with ultra-cheap electricity (around $0.06–0.08 per kWh). At higher prices, the economics of mining become unfeasible, and it is more profitable for a developer to build an AI center directly.
Panel participants emphasized that energy is just the beginning. AI tenants require redundant communication lines, backup generators, complex cooling systems, and 99.999% availability. Typical mining facilities in remote locations, built to simplified standards, are unrealistic for this. The overlap between the two models ends at the land, substation, and transformers.
Sites with their own generation were described as a more sustainable version of the hybrid scheme. Gas power plants often have excess capacity that mining can absorb, while the AI site uses guaranteed load. However, expensive GPUs are not suitable for such flexibility, and long-term AI contracts carry concentration risk. According to speakers' estimates, before the next halving in 2028, at least one major tenant or lender in the AI segment could default.
My expert analysis: The market is entering a phase of maturity where miners will have to make a strategic choice. Chasing GPUs without a clear monetization plan is a path to losses. Focusing on infrastructure, energy, and flexibility is what will allow not only survival but also profit from the AI boom, without repeating the mistakes of past cycles.