Edgevana has signed more than $700 million in multi-year, take-or-pay agreements to deploy NVIDIA Blackwell-based AI computing infrastructure across North America and Europe. The privately held company also reports $7 billion in contracted customer demand and an active commercial pipeline exceeding $20 billion. Announced November 9, the commitments cover deployments of NVIDIA B300 GPUs and GB300 systems, positioning Edgevana’s distributed infrastructure model as an alternative to waiting for newly constructed AI data center campuses. The financial figures are company-reported and have not been independently verified.

The company’s approach combines pre-energized land, existing data center facilities, power arrangements and high-speed connectivity into what it calls a virtual campus. Rather than concentrating all compute resources in a single greenfield development, Edgevana coordinates infrastructure across geographically distributed locations through a unified management platform. It says this approach can bring tens of megawatts of high-density AI compute online within 90 to 120 days. Its footprint comprises 17 operational data center locations and 10 in development. Edgevana has not disclosed aggregate energized capacity, GPU counts under the new agreements, individual customers or deployment schedules.

Edgevana emphasizes infrastructure for production AI inference, which can benefit from placing compute near user demand where latency and network conditions justify geographic distribution. The company says platforms relying on its infrastructure serve more than 10,000 downstream enterprise customers. Founded in 2019, Edgevana is led by co-founders and co-CEOs Ryan Fay and Subhan Jahromi. The November announcement does not identify the hyperscalers or AI laboratories involved; the $700 million in agreements is not recognized revenue or proof of completed GPU delivery.

Subhan Jahromi, co-CEO of Edgevana, said: “The defining constraint on AI isn’t algorithms or software, it’s physical, heavy-duty infrastructure.”

Analysis

Edgevana’s announcement illustrates an approach to the power and construction constraints affecting AI infrastructure development. Instead of relying exclusively on new gigawatt-scale campuses, its distributed model aims to aggregate existing electrical capacity and facilities across multiple markets. The relevant engineering questions include whether those locations can support Blackwell-generation power density, cooling, network performance and operational requirements without major reconstruction.

Training and inference impose different infrastructure demands. Large training jobs generally require closely coupled accelerators, while distributed inference can place compute closer to applications or end users. Actual latency and cost improvements depend on workload characteristics, networking and utilization. A common virtual-campus management layer does not make geographically separated facilities equivalent to a tightly coupled GPU cluster.

Take-or-pay commitments may provide predictable contractual cash flows, subject to individual contract terms, but they do not establish that all systems are installed, energized or accepted. Edgevana’s $7 billion contracted-demand figure and $20 billion-plus pipeline are distinct measures and should not be added together. Key future milestones include megawatts commissioned, installed GPU systems, customer acceptance, actual 90–120-day deployments and recognized revenue.