The $60M-per-MW playbook: can miners replicate Hyperscale's mining-to-AI

On September 1, Hyperscale Data switched off its Michigan Bitcoin miners to begin converting the site to AI hosting under a master services agreement its own press release says could be worth $1.2 billion for 20 megawatts of compute. Divide the headline two ways and the whole stock thesis appears in miniature: $1.2 billion over 20 megawatts is $60 million per MW of lifetime contract value, and against the roughly $110 million the company says would sit in stockholders' equity at mid-year, a single 20 MW customer looks like the asset that justifies the company. The instinct to price the other ~320 MW of the campus at the same $60 million-per-MW rate — a ~$19 billion lifetime figure — is exactly what the deal's own structure warns against.

The $60M number, divided properly

The yardstick only seems clean. Hyperscale's $1.2 billion figure is not annual revenue and it is not committed money — it is the company's estimate of what the 10-year agreement with two five-year customer extension options would produce if the customer stayed for the full maximum term of 20 years. Normalized, $1.2 billion ÷ 20 MW ÷ 20 years is about $3 million per MW per year; count only the initial 10-year term and it is twice that. The anchor contract also only covers 20 MW, with an option for an additional 32 MW, while the Michigan campus is planned for roughly 340 MW of eventual power. So even in the best case, the deal visibly prices a small slice: Hyperscale estimates the full exercise of the option would use no more than about 20% of capacity.

That is the gap the reader has to keep separate. The headline is one 20 MW customer. The site is a 340 MW project with no tenant for the other ~320 MW.

What the market actually pays per MW

The peer set that has already converted mining power to AI hosting is large enough to establish a market price — and it does not support $60 million per MW. Core Scientific's CoreWeave agreements, $8.7 billion over 12 years for 500 MW of critical load, work out to about $1.45 million per MW per year. TeraWulf's 200 MW Fluidstack deal anchors $3.7 billion over 10 years, near $1.85 million per MW per year. Applied Digital's CoreWeave contract — the 250 MW, $7 billion over 15 years that Hyperscale itself cites as a comparable — is roughly $1.9 million per MW per year. Hut 8's River Bend lease works out similarly, and Cipher's AWS lease below that. The peer cluster sits near $1.2–1.9 million per MW per year, versus Hyperscale's implied ~$3 million.

This is the narrative-versus-numbers tension the deal sets up, and it cuts in a specific direction. The company's own benchmark is roughly 1.6–2.5 times what comparable, contracted, hyperscaler-backed peers earn per MW. That is not proof of fraud — the anchor could be an above-market tenant — but it is the first falsification test, and the peer set fails the "consistent with the benchmark" check by a wide margin.

The other 320 MW, and the upgrade gap

Extrapolate the $60 million-per-MW figure to the uncontracted ~320 MW and you get a ~$19 billion lifetime number that has no counterparty, no financing, and no completion date attached. The more defensible read: if that capacity ever finds a tenant at peer-typical pricing (near $1.5 million per MW per year), it is roughly $500 million a year of potential revenue — large, but an order of magnitude below the extrapolated figure and entirely contingent on sales that have not happened.

That distinction is the real "revenue upgrade gap." The contracted 20 MW at the anchor's premium is worth maybe $60 million a year if fully ramped. The uncontracted ~320 MW is worth $0 in contracted AI revenue today, still earning mining-level dollars on the ~30 MW currently drawing power — and every additional watt needs both a customer and capital. The gap between what the anchor is priced at and what the rest of the site would command on market terms is the difference between the headline thesis and the deliverable economics.

Who funds the buildout without dilution

The conversion is capital-hungry at a company with thin equity. Hyperscale estimates the initial 20 MW retrofit of 60,000 square feet at $100–120 million — about $5–6 million per MW, roughly its entire mid-year stockholders' equity of ~$110 million. Its levers so far: roughly $62 million of Bitcoin (about 961 BTC), a ~$30 million Bitcoin-backed DeFi loan at under 5% designed to "reduce reliance on more dilutive forms of equity financing", another ~$9.6 million raised by selling 150 BTC, and a $125 million shelf that, as of July 2025, carried no immediate raise.

The contrast with peers is the structural point. Cipher, Core Scientific, TeraWulf, and IREN funded their conversion buildouts with project-level debt made cheap because a hyperscaler — Google, Microsoft, AWS — backstops the tenant's lease, converting a volatile miner's credit into Big Tech's credit, at loan-to-cost ratios up to 85%. Hyperscale's anchor tenant is an unnamed California-based neocloud with no disclosed hyperscaler backstop. Absent that credit enhancement, the non-dilutive debt path that let peers convert without heavy share issuance is not established here; the alternatives — more BTC loans, selling coin, or the shelf — each trade equity value or asset for the buildout. With a market capitalization near $22 million against ~$110 million of book equity and heavily shorted stock, the market is already pricing in that execution risk rather than rewarding the contract.

The forward observables that would resolve this are specific. The first 10 MW is slated to be operational before the end of 2026, the next 10 in Q1 2027; a confirmed first-customer go-live, a named tenant, or a comparable peer deal booked above ~$3 million per MW per year would each support the benchmark. A restatement of the $1.2 billion estimate, or a customer decision not to exercise extensions, collapses the multiple back toward the peer range. Until one of those lands, the $60-million-per-MW playbook is a price the market has not paid — it is a claim on a premium that only a single, unnamed tenant is so far willing to anchor.