Alpha Compute Corp. has signed a binding four-year agreement worth $81.7 million to lease the entire ALPHA-03 GPU cluster — all 592 liquid-cooled Nvidia B300 chips — at its Missouri "superintelligence factory," the company announced on Wednesday. The deal hands the Nasdaq-listed infrastructure firm roughly $20.2 million a year in recurring revenue, payable whether or not the customer ever uses the capacity.
The customer is described as a venture-backed AI compute marketplace valued at more than $1 billion in a recent funding round, serving more than 100 AI customers. It has already paid a $1 million data-center booking deposit and agreed to prepay 25 percent of the contract value — about $20.2 million — with 20 percent due at signing and the remaining 5 percent on delivery. Delivery of the cluster is targeted for early in the first quarter of 2027.
Missouri becomes Alpha Compute's first contracted US site
The ALPHA-03 cluster will run in Missouri under a separate agreement for 1.3 megawatts of liquid-cooled data-center capacity The 592-GPU count covers 576 cards under the customer contract plus 16 hot spares, a standard cushion for maintenance and failover. It is Alpha Compute's first contracted GPU cluster in the United States: ALPHA-01, with 504 Nvidia B200 GPUs, runs in Canada, and ALPHA-02, a 576-GPU B300 deployment in Sweden, secured a $71.9 million contract announced earlier this month.
Liquid cooling matters at this scale. Blackwell-class chips draw far more power per rack than prior generations, and landlords increasingly require direct liquid cooling for dense AI rows. Alpha Compute has positioned that expertise — purpose-built, high-density GPU builds — as its edge against larger cloud providers.
A take-or-pay bet on AI compute demand
The striking term is the take-or-pay structure: fees are owed for the full 48-month term regardless of utilization. The customer is effectively buying an option on scarce GPU capacity, while Alpha Compute converts uncertain future demand into a contracted revenue stream.
That pattern is spreading across the AI infrastructure trade. Morgan Stanley estimated this week that US startups serving AI data-center power and compute constraints have raised $33.5 billion in 2026 alone — nearly double the 2025 total — as power availability becomes the binding limit on expansion. Leasing firms with real, deliverable clusters are signing multi-year deals before the hardware is even racked.
Execution risk remains the real test
Alpha Compute still has to deliver. The announcement pegs ALPHA-03 delivery to early Q1 2027, and any supply-chain slippage on Nvidia B300 hardware or data-center buildout would flow straight into that timeline. The company has run this play before: its Sweden cluster secured bridge financing in the spring with a June go-live target, and the binding $71.9 million contract for that site followed later.
For investors watching the small-cap AI infrastructure names, the next checkpoint is the quarterly report. The ALPHA-02 and ALPHA-03 contracts together imply a meaningful jump in contracted revenue — but only reported revenue will confirm how much of it has actually landed.
— Hamza Faisal
