Major players in the Bitcoin mining industry are increasingly transforming their energy capacities and infrastructure into data centers for artificial intelligence (AI) and high-performance computing (HPC). This is no longer just a trend, but a strategic pivot dictated by the harsh realities of the market.

By the end of 2025, the global capacity of AI data centers reached 29.6 GW. The key shortage for this industry is not computing hardware, but stable energy supply: substations, cooling systems, and permitting documentation. These are precisely the assets miners have been building up for years for their ASIC farms.

To grasp the scale: in 2022, AI infrastructure capacity was less than 1 GW. In three years, it has grown roughly 200 times. The total energy consumption of the AI sector in 2024 was 9.4 GW, and individual resource-intensive training clusters could require over 100 MW. Meanwhile, the cost of GPU computing has fallen by more than 99% since 2006.

Total capacity of AI data centers
Total capacity of AI data centers.

There are 5,427 data centers in the United States alone.

Electricity Has Become More Valuable Than ASICs

Specialized mining equipment (ASICs) is unsuitable for either model training or inference. AI sector clients need ready-made energy sites with long-term electricity supply contracts, grid connections, and physical buildings for server racks. Such facilities are typically located in US regions with cheap electricity, such as Texas or the Gulf Coast.

The deals already signed by leading miners are impressive:

  • In November 2025, IREN signed a five-year agreement with Microsoft for cloud GPU services worth approximately $9.7 billion (a 750 MW campus in Texas);
  • In December 2025, Hut 8 entered into a 15-year contract with Fluidstack for $7 billion (the River Bend facility in Louisiana, 245 MW). Payments are guaranteed by Google;
  • TeraWulf reported contracted HPC revenue of $12.8 billion and is already earning more from leasing than from mining;
  • Core Scientific expanded its agreement with CoreWeave to $10.2 billion over 12 years.

The AI Pivot Is Changing Company Valuations

According to CoinShares estimates, public miners have announced AI and HPC contracts worth over $70 billion. Companies with such agreements trade at a multiple of 12.3 times annual revenue, while miners focused solely on Bitcoin have a multiple of only 5.9.

Analysts predict that the share of AI in public miners' revenue will grow from approximately 30% in the first quarter of 2026 to 70% by the end of that year.

The economics of Bitcoin mining have sharply deteriorated. The average cost to mine one coin is $78,000, while the current price is around $60,000. At this price level, more than 20% of miners are operating at a loss.

However, transitioning to AI is expensive. While typical mining infrastructure costs $700,000 to $1 million per 1 MW, AI facilities with liquid cooling require $8 to $15 million per 1 MW.

There are also risks: rising debt loads, delays in facility commissioning, and dependence on a narrow circle of hyperscalers. For example, Hut 8 plans to launch River Bend only in the second quarter of 2027. By the end of March 2026, IREN had disclosed about $3.75 billion in convertible debt, and in May it raised another $3 billion. In the same month, the largest public miners sold a record 32,000 BTC—more than in all of 2025.

Expert commentary from Cryptalist: The current situation is a classic example of "creative destruction" in the crypto industry. Miners who fail to adapt in time risk being left with outdated equipment and negative margins. The shift to AI is not diversification—it is a matter of survival in the new economic realities. The question is whether miners can handle the enormous capital expenditures without drowning in debt.