Corvex, Inc. (Nasdaq: MOVE) announced on November 16, 2026, the execution of a definitive three-year agreement valued at $103 million to provide NVIDIA Blackwell computing capacity to a scaled AI inference platform. The contract is structured under take-or-pay terms, obligating the customer to pay for the reserved capacity regardless of actual utilization.
Under the agreement, Corvex will deliver more than 1,000 NVIDIA HGX B200 GPUs connected with NVIDIA Quantum-2 InfiniBand networking. The deployment is expected to utilize 1.6 megawatts of IT capacity and cost approximately $78 million, including GPU servers, storage, networking, and spares.
Financing for the deployment will be provided through a combination of a $26 million customer prepayment and $38 million in committed debt financing from USD.AI. These funds are expected to cover $64 million, or more than 80% of the deployment cost. The debt commitment is for a three-year asset-backed loan that is non-recourse to Corvex, Inc.
The company expects to begin delivering the capacity in the first quarter of 2027, with full service expected by the end of that quarter. Upon completion of this deployment, Corvex anticipates its contracted annual recurring revenue at its Mid-Atlantic 1 data center will increase from $22 million to approximately $56 million.
The filing notes that Corvex has entered into a binding term sheet for the debt financing, but the company cautions that the availability of funds is subject to technical due diligence and other customary closing conditions. Failure to satisfy these requirements could result in delayed, reduced, or unavailable funding.