Armada Packs Nearly 3,000 GPUs Into Single Modular Data Center as AI Infrastructure Race Accelerates
Armada announced Wednesday a 10-megawatt modular data center that houses 2,880 graphics processing units across 40 racks and can be deployed in a matter of months. The prefabricated unit, called Orion, marks a sixfold increase in capacity from the startup's 1.77 MW Leviathan product that emerged about 13 months earlier.
The new offering targets governments and large enterprises racing to build sovereign AI infrastructure on sites with stranded power capacity or behind-the-meter locations where conventional hyperscale construction timelines stretch across years. It comes as global demand for turnkey AI computing solutions intensifies, with the modular data center market expected to triple to $111.62 billion by 2032 from $36.75 billion in 2025, according to Stratview Research projections cited by TMC Insight.
Orion ships as a six-module core that integrates compute, networking, power distribution, and cooling systems in a factory-assembled configuration designed to fit inside existing structures, Armada said. The system accommodates NVIDIA's Blackwell GPUs and includes what the company describes as "greater-than-N+1 redundancy" for its forthcoming Vera Rubin chips. Customers can link multiple Orion units across distributed sites to scale deployments into the hundreds of megawatts while retaining operational control under an own-and-operate model.
Each installation includes Armada Bridge software for orchestrating GPU workloads across racks and geographic locations, bundled with a model marketplace and integration support for NVIDIA's Token Factory framework.
Early Traction in Northern Europe
Norwegian AI infrastructure provider Fossefall has ordered five Leviathan units delivering more than 9 MW of combined capacity, with the first systems slated for delivery in the first quarter of 2027 under a contract worth several hundred million Norwegian kroner, Armada disclosed last week. "We selected Armada for its best-in-class technology and ability to significantly accelerate time to market," Fossefall CEO Øyvind Vesterdal said.
Those Leviathan modules, manufactured at Armada's Dublin facility, employ closed-loop cooling technology that eliminates daily water consumption.
Startup Trajectory and Manufacturing Push
Dan Wright and Pradeep Nair founded Armada in late 2022. The company raised a $230 million Series B round in May at a $2 billion pre-money valuation, with co-leads Overmatch, BlackRock, and 8090 Industries bringing total capital raised to nearly half a billion dollars. A separate $131 million strategic round closed in July 2025, coinciding with the Leviathan launch.
In May, Armada partnered with Johnson Controls to construct Galleon Forge One, a 400,000-square-foot factory in Arizona expected to generate 500 jobs and begin continuous production starting with the Leviathan line. "Johnson Controls is working with Armada to rapidly deliver secure modular data centers at scale," Johnson Controls CEO Joakim Weidemanis said at the time.
Government and Cloud Integrations
The company has woven its technology into several high-profile partnerships. Armada integrated its Galleon product line with Microsoft Azure Local to enable sovereign private-cloud deployments, with both companies "actively engaging customer deployments" as of March, according to a joint statement. "Sovereign cloud capabilities are essential, not optional," Microsoft Specialized Clouds President Douglas Phillips remarked.
Palantir has deployed Armada infrastructure for field-ready data centers equipped with NVIDIA B300 accelerators, TechRadar reported in July. Armada also opened a demonstration facility in Reston, Virginia, in April, showcasing units for government buyers through its partnership with Carahsoft.
Crowded Competitive Landscape
Armada enters a market already populated by established infrastructure giants. Schneider Electric, Vertiv, Delta Electronics, Dell Technologies, and Hewlett Packard Enterprise all offer modular or prefabricated AI-capable data center solutions, Fortune Business Insights noted three weeks ago. Schneider introduced cooling distribution units scalable beyond 10 MW in January. Crusoe launched its Spark modular AI data centers powering Crusoe Edge Zones in March, adding pressure to the segment.
"AI infrastructure should be fast to deploy, easy to scale, and sovereign wherever it needs to live," Wright said in Monday's announcement. Armada claims Orion units can ship within months of deposit, though actual deployment timelines will likely depend on site-specific factors and supply chain conditions that have bedeviled the broader data center industry.
The question facing Armada now is whether demand for sovereign, rapidly deployable GPU clusters will justify the capital intensity of scaling modular production before larger competitors mobilize their manufacturing networks.
