Summary
- TensorWave has signed for a first 222MW facility at Fermi's Project Matador in Texas.
- The 15-year first-phase lease represents about $6.5bn of contracted revenue, subject to project conditions.
- Expansion rights could take the relationship beyond 650MW, demonstrating the size of anchor contracts being used to support AI campus financing.
Fermi America has signed its first binding customer lease at Project Matador, agreeing to develop a 222MW data centre for AI cloud provider TensorWave with expansion rights capable of taking the relationship beyond 650MW.
The first phase is structured as a turnkey data centre lease. Fermi will develop, construct, and deliver the facility, while TensorWave will take occupancy in stages beginning in the second half of 2027.
Fermi puts contracted revenue from the initial 15-year term at approximately $6.5 billion, excluding renewal periods. TensorWave has rights covering two additional data centres that, if exercised, would take total facility power across the partnership above 650MW.
The first building is intended to support tens of thousands of AMD Instinct GPUs for large-scale training and inference. TensorWave operates an AMD-focused AI cloud and is expanding physical capacity to meet demand for accelerator infrastructure.
A binding tenant changes the campus finance equation
Project Matador has been promoted as a 17GW private-power and AI campus in the Texas Panhandle. Fermi says about 6GW of that planned capacity has already been permitted and more than $1.5 billion has been invested in the wider development.
Those figures describe an exceptionally large masterplan, but the TensorWave lease provides a different kind of evidence: a named customer with a contractual commitment to occupy a defined first phase.
For data centre financing, the difference between planned megawatts and leased megawatts is fundamental. Land, permits, and power infrastructure can establish development potential, but lenders and infrastructure investors generally place greater value on contracted cash flow from a tenant capable of supporting the project economics.
The lease contains conditions including project guarantees and financing, so the $6.5 billion figure should not be treated as cash already secured. It represents expected contracted revenue across the initial term if the project is delivered and the agreement performs as intended.
Fermi says phased delivery will begin in the second half of 2027. That timetable requires construction of the data centre itself alongside the private electrical infrastructure intended to support the campus.
Private power is being matched to long leases
Project Matador is designed around a behind-the-meter energy system rather than relying solely on a conventional utility connection. Fermi has described plans combining natural-gas generation, nuclear capacity, utility supply, solar, and battery storage in a private grid.
The approach reflects the same constraint visible across newer AI campuses: developers want to align power construction directly with customer demand rather than wait for grid infrastructure whose delivery may be outside their control.
A 222MW anchor tenant gives that energy buildout a clearer commercial base. If TensorWave exercises its expansion rights, the relationship could exceed 650MW across three phases, creating a load large enough to justify substantial dedicated generation and network investment.
Scale also increases concentration risk. A campus depending on a small number of very large customers becomes exposed to their financing, technology choices, deployment schedules, and future demand. The long lease term is intended to reduce that uncertainty, but it does not eliminate execution risk during construction.
The project is a useful comparator for Europe, where large AI sites increasingly need customer commitments before developers can finance the electrical and mechanical infrastructure required to secure hundreds of megawatts. The contractual structures will differ, but the underlying sequence is similar: power rights and permits create potential, a tenant makes the capacity bankable, and financing turns the design into physical infrastructure.
TensorWave’s reliance on AMD hardware adds another variable. Accelerator generations are changing quickly, and facilities due to open from 2027 need electrical distribution and cooling systems capable of accommodating equipment whose final rack densities may continue to rise before deployment.
Fermi describes the first phase as a complete turnkey solution, which places the responsibility for delivering that environment with the developer. The campus therefore has to bring land, generation, grid systems, buildings, cooling, and customer-ready data halls together on the same schedule.
The TensorWave agreement does not validate all 17GW envisioned for Project Matador, but it does convert the first 222MW from proposed capacity into a binding commercial programme. The next milestones are financing, construction, first power, and phased delivery — the points at which a very large AI infrastructure plan becomes an operating data centre.
