Edgevana books $700M of Blackwell deals, says it can power up in 90-120 days

Edgevana, a San Francisco data center and compute orchestration company, said it has signed more than $700 million in multi-year take-or-pay agreements to deploy Nvidia Blackwell systems in North America and Europe. The company says it can bring capacity online in 90-120 days.

The announcement, issued through PR Newswire on Nov. 9, 2026, is a company statement. None of the figures below have been independently verified.

What Edgevana says it signed

According to the company, the agreements are fully committed and cover Nvidia B300 GPUs and GB300 systems. Edgevana says the deals sit within $7 billion of contracted customer demand and an active pipeline above $20 billion. It did not name any customers, the number of GPUs, or the sites involved.

The company describes its product as a “virtual campus”: pre-energized land, data center facilities and high-speed networks combined into one platform. Edgevana says this lets customers bring tens of megawatts of dense AI compute online in 90-120 days, and keep a single control point over a distributed footprint. It contrasts that with the years a new greenfield build can take.

On scale, Edgevana says it runs 17 operational data center locations with 10 more in development. It also says platforms running on its infrastructure serve more than 10,000 downstream enterprise customers. The company was founded in 2019 and says it has supported two of the world’s largest hyperscalers with edge production, and has worked on low-latency systems for high-frequency trading.

Betting that inference moves compute to the edge

Co-CEO Ryan Fay framed the timing around a shift in AI workloads. Training favors giant centralized campuses, he said, but “inference is live production traffic.” Co-CEO Subhan Jahromi put the constraint in physical terms: “The defining constraint on AI isn’t algorithms or software, it’s physical, heavy-duty infrastructure.”

The commercial structure matters as much as the hardware. Edgevana says it combines long-term energy supply, its own GPU allocation and committed customer demand, and that signing contracts before deployment lowers financial risk. In a take-or-pay deal, the customer pays for reserved capacity whether or not it uses it. That moves demand risk from the operator to the buyer, which is what lets a provider finance expensive hardware before it is switched on. The company also claims a lower cost basis than traditional providers, but gave no pricing or comparison.

Why power, not chips, sets the schedule

The 90-120 day claim rests on one thing: power that is already connected. Blackwell-generation racks are far denser than what most data halls were built for. Barrack AI, a vendor blog that cites figures from HPE, Lenovo, TrendForce and Schneider Electric, puts a fully loaded GB300 NVL72 rack at roughly 132 to 140 kW, against an average rack density of about 8 kW. Other outlets report higher numbers, and these are not Nvidia specifications, so treat them as estimates.

Grid strain from this kind of load is also a design problem for Nvidia. In its developer blog and in coverage by Utility Dive, Nvidia says GB300 power shelves with built-in energy storage smooth the sharp swings between idle and full load that synchronized GPU work causes, and can cut peak grid demand by up to 30%. That figure is Nvidia’s own estimate.

Together, those numbers explain Edgevana’s pitch. Tens of megawatts of racks drawing over 100 kW each cannot be dropped into an ordinary facility, and waiting for new grid connections is the slow part. A site that already has power is worth more than the chips that go in it, if the claim holds.

What the release leaves unproven

Several things are not established. The company did not say who the customers are, how the $700 million splits between North America and Europe, when the first systems go live, or which sites are involved. It did not explain how the $700 million relates to the $7 billion in contracted demand, or whether the larger figure is binding in the same way. The 90-120 day timeline is a target the company states, not a delivered result.

For buyers, the practical question is whether pre-energized capacity can be reserved at a competitive price compared with the large cloud providers and other GPU hosts. Edgevana’s own materials give no pricing. Until it names customers or publishes delivery dates, the announcement is best read as a signal of where demand for powered, ready-to-use capacity is heading, not as proof the capacity exists.