The Great Compute Pivot: How Bankrupt Crypto Miners Are Becoming AI Infrastructure Kings

You might not know what you’re looking at when you walk into one of these facilities in rural Wyoming or west Texas. Rows of industrial buildings, cooling systems humming steadily against the flat landscape, power lines running in from substations that were expanded specifically for this tenant. A year ago, these sheds were filled with ASIC mining machines that were constantly processing SHA-256 calculations to produce Bitcoin at whatever level of network difficulty was required. There are still some of those rigs in operation. However, GPU racks are now located in the building’s corner where liquid cooling loops were installed and new floor space was recently cleared. Furthermore, the business renting that space isn’t a cryptocurrency fund. An AI model is being trained by this company.

The economic factors that led to this change are striking. The weighted average cash cost to produce one bitcoin among publicly traded miners was roughly $79,995 in Q4 2025, according to CoinShares’ Q1 2026 mining report. The price of Bitcoin was between $68,000 and $70,000. When cooling, electricity, hardware depreciation, and facility overhead are taken into consideration, each coin produced results in a loss of about $19,000. In the words of one industry observer, “the math doesn’t work.” The situation worsened in April 2024 when the block reward was cut overnight from 6.25 BTC to 3.125 BTC, resulting in a 50% decrease in revenue for pure mining operations while operating costs remained essentially unchanged. The industry wasn’t the only one harmed by that combination. It compelled an examination of these businesses’ true nature.

Increasingly, the response is that they are infrastructure firms with exposure to Bitcoin, not Bitcoin firms with control over infrastructure. The public mining industry has now signed contracts worth over $70 billion for high-performance computing and artificial intelligence. The expanded agreement between CoreWeave and Core Scientific is valued at $10.2 billion over a twelve-year period. TeraWulf’s contracted HPC revenue is $12.8 billion. At its River Bend campus, Hut 8 inked a $7 billion, 15-year lease for AI infrastructure. Google-backed Fluidstack and Cipher Digital have a multibillion-dollar deal. These collaborations are not exploratory. These agreements, which span ten years, were made with counterparties who require computing power and are able to pay for it at prices that mining is currently unable to match.

Perhaps the most obvious indication that Wall Street has chosen to see these companies differently was Morgan Stanley’s involvement in the discussion in February 2026. The bank started covering the mining industry as an energy infrastructure play for the AI economy rather than as a cryptocurrency play. Cipher Mining and TeraWulf received Overweight ratings from analyst Stephen Byrd, who also set price targets of $38 and $37, respectively.

Byrd framed the thesis around a potential “REIT endgame” in which mining facilities transform into data center real estate that generates consistent, rent-like income. MARA Holdings was given an Underweight rating and a $8 price target because it has placed significant bets on buying Bitcoin rather than diversifying into AI. The market responded right away; TeraWulf and Cipher both saw daily gains of 13%. The fact that pure-play miners’ revenue per terahash dropped from $70 to $35 made the point very evident.

The storyline—a troubled cryptocurrency asset turns into an AI infrastructure hero—has an alluring simplicity, but in practice, there is a major infrastructure obstacle that the breathless headlines frequently ignore. An Antminer S19 cannot be disconnected and replaced with an NVIDIA H100 and referred to as a data center. Bitcoin mining is built to be resilient and interruptible: the rigs don’t require low-latency fiber connections, can shut down in a matter of seconds during a grid stress event, and rely on simple evaporative cooling. Workloads for AI training are the opposite. They require 99.99% uptime because stopping a training run corrupts the work and wastes a significant amount of compute time.

To control GPU heat density, they need liquid cooling systems. They require fiber connectivity, which was not intended for the majority of rural mining sites. The majority of miners are unable to finance the real costs associated with converting a mining shed to Tier 3 data center standards using their current balance sheets. Either direct partnerships with hyperscalers who co-invest in the upgrade or debt secured against the AI contracts themselves are being used to finance the transition. Due to the capital barrier, the industry is becoming divided into two groups: those who are stuck in mining economics and have nowhere else to go, and operators who are creditworthy enough to make the leap.

What this pivot means for the Bitcoin network itself is a question that is not sufficiently addressed in the coverage of this change. A miner’s power capacity is essentially locked out of the Bitcoin network for five years when they sign a five-year AI hosting contract. When the network needs security, it cannot dynamically switch back to hashing. Additionally, the incentive to prioritize Bitcoin at all gradually wanes if AI revenue continues to outpace mining revenue for the coming years, as current projections indicate is likely. In this industry, energy has always been the most valuable resource.

Bitcoin only offered the initial use case. Currently, that energy is being reallocated to the next high-bidder, which is obviously AI compute in 2026. A Bitcoin price rebound toward $100,000 might alter the calculations once more and encourage some mining capacity. A smaller, more concentrated group of less expensive miners may be able to secure the network if the multi-year contract structure makes capacity simply unavailable when the price changes.

As this develops, it seems as though the mining industry is going through the same forced evolution that every commodity sector eventually experiences: the point at which the infrastructure built around the original product can no longer be justified on its own and the infrastructure finds a second life somewhere nearby. The sheds remain in place. There are still power connections. Running high-density compute operations in low-power locations is still a discipline. The industry’s Bitcoin maximalists will contend that this is merely a stopgap measure until the price recovers. A more nuanced picture is revealed by the balance sheets. The $70 billion in contracts indicates that many of these operators have no intention of returning.