Hyperscale Data (NYSE American: GPUS) says it is in advanced talks to convert part of its Michigan Bitcoin mining site into a 52 megawatt AI colocation hub for a single unnamed customer, a 20 year deal the company projects could top $2.5 billion in revenue.
A small-cap Bitcoin mining operator called Hyperscale Data is trying to do something that, two years ago, would have been a fringe bet: turn a chunk of its Michigan power plant into an AI data center.
In a press release issued Monday, the Las Vegas-based company said it is in "advanced negotiations" with a prospective customer over a master services agreement (MSA) for AI colocation services at its Michigan data center campus. Colocation, in plain terms, is the business of renting floor space, cooling, and power to AI customers so they can run their own GPU servers on someone else's site.
The first phase would deliver about 20 megawatts of "critical power capacity," the rated IT load a data center can deliver to servers, with another roughly 32 megawatts slated for 2028. Together, the company projects more than $1 billion from the initial tranche and roughly $1.5 billion from the later one, for more than $2.5 billion in total revenue over 20-year terms, including renewal options. The customer is not named.
The deal, if it closes, would redirect about 17% of the 300 megawatts of total power capacity Hyperscale Data says the Michigan campus can eventually support, a figure the company itself calls "potential" and which is contingent on regulatory approvals, financing, utility agreements, and engineering studies the company has not detailed.
Hyperscale Data trades on NYSE American under the ticker GPUS. The company describes itself as "an AI data center company anchored by Bitcoin," and the Michigan campus has historically been used for Bitcoin mining. The press release says the company will "likely cease all Bitcoin mining operations" at the site over several months once the AI services come online, effectively repurposing mining-anchored power for AI and high-performance computing workloads.
The framing matters. AI colocation, the business of leasing power and space to AI customers who bring their own chips, has been one of the tightest markets in 2026, with operators such as Equinix, Digital Realty, and a handful of private hyperscale developers competing for new gigawatts. Legacy Bitcoin mining sites, with their pre-existing grid interconnects, substation agreements, and on-site cooling, are an increasingly common starting point. Hyperscale Data's pitch is essentially: we already have the power, and we can flip it from crypto to AI faster than a greenfield build.
That is also the bet, and it is the part of the story the press release does not fully resolve. Hyperscale Data's announcement describes the deal as "advanced negotiations," not a signed agreement. The company offered no executed signature date, no financial guidance for revenue or EBITDA impact, and no third-party corroboration. The prospective customer's name, vertical, and end-use workload are also undisclosed, leaving readers to weigh a 20-year, $2.5 billion revenue projection against a single, unnamed counterparty.
The scale of the proposed pivot is also worth anchoring. 52 megawatts of critical power capacity is enough to run a medium-sized AI training cluster continuously, with peak draws comparable to the load of roughly 16,000 U.S. households at average residential consumption. It is small relative to the multi-hundred-megawatt campuses being built by the actual hyperscalers, but it is also far larger than what most microcap data-center operators have been able to monetize in the AI era.
The release also offered no timeline for when Bitcoin mining at the site would actually stop, only that it would happen "over several months" as AI capacity comes online. For now, both businesses run in parallel at Michigan, and the company has not disclosed how it plans to manage the transition, the cost, or the potential impact on short-term cash flow from the existing mining fleet.
What to watch: whether the master services agreement is signed in the coming weeks on the company's current timeline, whether the unnamed customer is ever publicly identified, and whether Hyperscale Data discloses the contractual penalties, performance guarantees, or take-or-pay terms that would convert the $2.5 billion projection into something closer to booked revenue. Investors and partners will also be watching for any independent confirmation of the 300-megawatt site ceiling, which is currently a company-stated target rather than a commissioned figure.
For now, Hyperscale Data has given the market a pattern to think about: a small, Bitcoin-anchored operator is trying to recycle a piece of its power into a 20-year AI bet, on the strength of an unsigned deal, with an unnamed customer, and a price tag that lives or dies on whether the framework contract is ever signed.