Hyperscale Data (NYSE: GPUS) has officially shut down all of its Bitcoin mining operations at its Michigan data center.

‍

As of Sept. 1, the company had turned off every Bitcoin miner at the facility. The company is now preparing the site to provide computing infrastructure for artificial intelligence. I think that's a smart move.

‍

To be sure, Hyperscale Data isn't abandoning Bitcoin entirely. It still operates roughly 10 megawatts of mining capacity in Montana. But in Michigan, the company already has something potentially more valuable to do with its electricity and data center infrastructure.

‍

The company has signed a master services agreement with an unnamed California-based "neocloud" provider for 20 megawatts of AI computing capacity. The initial agreement runs for 10 years, with two optional five-year extensions. If it exercises those extensions, Hyperscale Data estimates the contract could generate more than $1.2 billion in revenue.

‍

The customer also has the option to add another 32 megawatts. If that happens and the relationship runs for the maximum term, Hyperscale Data says total contract revenue could exceed $3 billion.

‍

Why AI makes more sense than Bitcoin mining

‍

Bitcoin mining and AI data centers have something important in common: they both need enormous amounts of electricity and computing infrastructure.

That means Hyperscale Data isn't starting from scratch.

‍

It already owns a 34.5-acre Michigan campus with a 617,000-square-foot facility and existing power infrastructure. The company ultimately believes the property could support roughly 340 megawatts of power capacity. The initial AI agreement, including the customer's expansion option, would consume only about 20% of that potential capacity.

‍

The economics are also different.

‍

Bitcoin mining revenue depends heavily on the price of Bitcoin, mining difficulty, electricity costs, and the amount of computing power competing for the same rewards. It's an inherently volatile business.

‍

AI infrastructure can potentially provide something much more predictable: long-term contracts with customers paying for computing capacity.

‍

Hyperscale's existing agreement is a good example. Rather than mining Bitcoin and hoping the economics remain attractive, the company could have a customer paying it over many years for access to power and infrastructure.

‍

Hyperscale Data is even selling some of its Bitcoin to finance the transition. According to The Block, the company has sold 830 Bitcoin for roughly $53 million over the past five weeks, with proceeds helping fund the Michigan development.

‍

In other words, it's converting a volatile asset into physical infrastructure that could potentially generate recurring revenue.

‍

The stock tells a different story

‍

To be sure, investors aren't exactly celebrating.

‍

Hyperscale Data shares recently traded around $0.22, an all-time low, and have fallen more than 76% this year.

‍

And there are legitimate reasons for the skepticism.

‍

At the end of June, Hyperscale had just $36.8 million in cash, while current liabilities totaled approximately $201.7 million. The company expects the initial 20-megawatt Michigan AI deployment alone to require more than $100 million in investment over time, and Hyperscale has warned that its existing liquidity won't be sufficient to meet its needs over the next 12 months.

‍

So the AI pivot doesn't magically fix the company.

‍

Hyperscale still has to finance the buildout, deliver the promised capacity, keep its customer, and prove that the Michigan facility can generate the kind of revenue management is projecting.

‍

But strategically, moving the Michigan site away from Bitcoin mining makes sense.

‍

Hyperscale already has one of the assets AI companies desperately need: access to large amounts of power. Instead of using that power to compete for Bitcoin rewards, management is trying to sell it under long-term contracts to companies that need enormous amounts of computing capacity. That's a much more compelling use of the Michigan facility. Now Hyperscale Data has to prove it can actually pull it off.

‍