Bitcoin Mining Was Supposed to Be Dead, Then AI Compute Changed Everything

A row of Antminer S19s sits in the dark somewhere in west Texas inside a low metal structure that appears from the highway to be a livestock barn. Fans are off. Turn off the lights. There is no longer the constant hum that used to shake the walls. A team of electricians is replacing it with new, thicker cabling that is rated for a different type of load. The building will stop mining Bitcoin by the summer. It will use Nvidia GPUs to train a frontier model.

This is not an isolated incident. It is occurring throughout North America, including decommissioned smelters in Washington State, the wind corridors of the Dakotas, and Quebec’s inexpensive hydro belt. The publicly traded Bitcoin miners, the same businesses that told investors in 2021 that they were creating the foundation of a decentralized future, are subtly evolving into something else. landowners. brokers of power. AI hosts. Whatever you want to name it. The term “Bitcoin” is gradually disappearing from their pitch decks.

There is little room for disagreement because the math is brutal. In the first quarter of 2026, the average publicly traded miner spent about $79,995 to create one Bitcoin. The price of Bitcoin has fluctuated between $67,000 and $70,000. For each coin extracted from the hash, that represents a loss of $10,000 to $20,000. Morgan Stanley did the repurposing math. Their analysis shows that switching one megawatt of power from mining to AI hosting results in a valuation premium of more than ten times. Ten times. Not thirty percent. Not twice. a magnitude order.

The first and most decisive action was taken by Core Scientific. It sold roughly 1,900 Bitcoins for $175 million back in January and declared that it would clear all of its holdings by Q1. Its mining facilities in Texas are being transformed into 1.3 gigawatts of AI hosting capacity. They signed a $10.2 billion contract with CoreWeave. In March, MARA, a company that built its reputation over the years by never selling a coin, discreetly approved the liquidation of all 53,822 BTC listed on its balance sheet. Wu Jihan, the founder of Bitdeer, stated that the company needs liquidity to seize land and electricity while the AI window is still open. Despite this, Bitdeer cleared its treasury and increased its hash rate. The CEO of Bitfarms spoke softly into a live microphone, saying, “We are no longer a Bitcoin company.”

Hardware is the intriguing wrinkle. ASICs, the specialized devices that hash Bitcoin using SHA-256, are ineffective for artificial intelligence. It is not enough to simply “flip a switch.” They are being torn out by the miners, who are then replacing them with GPU racks and selling them for scrap or secondary markets. The boring stuff is what really transfers and what no one considered to be worth properly five years ago. Cooling systems and substations. contracts for ten years of electricity that were signed when it was inexpensive. permits for the environment. capacity of the transformer. As it happens, the miner was never the valuable asset. It was the building’s power line.

Beneath all of this is a more subdued tale about Bitcoin. The market absorbs those coins somewhere if the institutional holders, who are supposedly the most dedicated, start acting as net sellers to finance AI buildouts. The network continues to function. Adjusting is difficult. A new floor is found by the hash rate. A publicly traded company pursuing hyperscaler contracts has entirely different incentives than a guy with a laptop in 2009, as one Reddit commenter put it. In good faith, intelligent people cannot agree on whether that is bullish or bearish for Bitcoin’s security model.

The historical rhyme is difficult to ignore. Telecom companies overbuilt fiber networks that failed during the dot-com bust in the 1990s. Ten years later, the modern internet was constructed on top of that same abandoned fiber by Google and Netflix. It’s possible that the Bitcoin miners unintentionally overbuilt the power infrastructure for an entirely different revolution. They simply believed that they were constructing it for themselves.

Whether each of these pivots is successful is still up in the air. Some people will. Digital and Core ApplicationsWeave has already demonstrated that the model can produce actual income. Others will find that managing an AI data center differs greatly from mining in terms of customer relations and skill requirements. As this develops, there’s an odd feeling that an entire industry is subtly changing its name in real time in the hopes that no one will notice that the ASICs have disappeared. The machines are subject to change. The power remains. It turns out that was the true issue all along.