Former Bitcoin Miner Signs $1.2B Deal to Convert Michigan Data Center for AI Compute
Hyperscale Data (NYSE American: GPUS) signed a Master Services Agreement on June 24 worth approximately $1.2 billion with an unnamed California-based neocloud provider, converting its Michigan Bitcoin mining campus into AI compute infrastructure.
The initial commitment covers 20 megawatts of AI-ready colocation capacity. The customer holds options to scale to 52 MW — lifting the deal’s total value above $3 billion if exercised. A further 32 MW option sits beyond that. At full build-out, the Michigan campus is designed to hold 300 MW; this deal uses about 17% of that potential.
The Retrofit
Hyperscale Data will invest $100 to $120 million in capital expenditures to reconfigure approximately 60,000 square feet of existing mining space for AI and high-performance computing workloads. About 28 MW of power currently running Bitcoin mining operations will be redirected to the AI tenant’s infrastructure.
Bitcoin mining at the campus continues on the remaining capacity. But the direction is clear: AI compute generates more predictable revenue against more stable margins than mining at current BTC prices and network difficulty. The colocation-plus-services model positions Hyperscale as the infrastructure layer — power, cooling, facility — while the neocloud tenant brings hardware and customers.
The Pattern
Hyperscale Data joins a list of former Bitcoin miners executing the same trade. IREN, now reporting $3.1B in ARR after expanding into Spain, completed its AI pivot earlier. Applied Digital locked a $7.5B, 300 MW deal in Louisiana. Hut 8 signed a $9.8B, 352 MW deal at Beacon Point.
The logic is consistent across all of them: miners hold large existing power contracts and physical data center space at the exact moment AI companies need both urgently and in volume. The arbitrage is structural — redirect megawatts from proof-of-work to inference workloads, capture the higher per-MW revenue that AI companies are willing to pay.
Scale Context
The Michigan deal is smaller than the headline numbers from IREN or Hut 8 — 20 MW versus hundreds of megawatts. But the expansion optionality is the point. A customer who signs at 20 MW with options to 52 MW and rights to additional capacity is signaling a longer-term infrastructure dependency, not a one-time purchase. The neocloud customer stays and grows if AI inference demand continues to compound.
Hyperscale Data’s Michigan campus has 300 MW of eventual potential. This deal fills 17% of it. If the AI infrastructure buildout sustains its current trajectory, the remaining 83% is the business case.