Fermi America has signed a binding 15-year lease agreement with TensorWave valued at 6.5 billion dollars for a 222-megawatt GPU-ready data center facility, according to the announcement.
The facility is designed to house tens of thousands of AMD Instinct GPUs and includes expansion rights that could grow the installation to more than 650 megawatts. The contracted revenue of 6.5 billion dollars spans the full initial lease term, with power capacity positioned to serve large-scale artificial intelligence and machine learning workloads. TensorWave operates as a provider of GPU compute infrastructure.
Fermi America develops and operates data center capacity across the United States. The company has emerged as one of several operators racing to build power-dense facilities capable of supporting the computational demands of large language model training and inference. Data center operators have faced constraints from both grid capacity and power availability, pushing companies to secure long-term contracted arrangements with equipment providers and compute infrastructure firms.
The 222-megawatt baseline represents a substantial single-facility commitment in the current GPU infrastructure buildout. Industry players including CoreWeave, Lambda Labs, and others have signed similar multi-year arrangements with cloud providers and AI companies seeking dedicated compute access. The binding nature of the agreement locks both parties into the contract terms, reducing execution risk compared to non-binding preliminary arrangements.
Expansion rights embedded in the deal allow the facility to scale beyond its initial 222-megawatt footprint if demand materializes. The potential 650-megawatt expansion would position the site among the larger dedicated AI compute installations in development, though many operators are still in early construction phases.
The announcement carries relevance for power infrastructure planners, since data center buildouts of this scale require coordinated grid connections and often trigger state and local permitting processes. Fermi's willingness to commit to 15-year contracted revenue streams reflects confidence that GPU demand will sustain at current price levels, though TensorWave's ability to fill the facility at projected capacity remains unverified.
A facility at 80 percent average occupancy generates vastly different returns than one operating at 40 percent, and market rates for GPU compute have fallen 60 to 70 percent since late 2023, compressing margins for both operator and tenant. If Fermi discloses quarterly occupancy figures in public filings or investor updates, those numbers will show whether the facility is tracking to the revenue expectations that justified the 6.5 billion dollar contract.
