Armada closed a $230 million Series B at a $2 billion valuation and announced a manufacturing agreement with Johnson Controls to produce modular AI data centers at a new 400,000-square-foot Arizona facility called Galleon Forge One. The round was oversubscribed and co-led by Overmatch, 8090 Industries, and BlackRock — a new entrant — with strategic participation from Johnson Controls, NightDragon, Mitsui, and Singtel Innov8, plus existing backers Felicis, Founders Fund, Lux Capital, and Shield Capital.
The core product is the Leviathan, a megawatt-scale containerized data center that deploys within days rather than the years conventional hyperscaler builds require. The unit attaches to existing local energy sources — solar arrays, gas flares from oil wells, grid tie-ins — eliminating the need for dedicated utility infrastructure. Galleon Forge One will begin producing Leviathans this summer, with Johnson Controls contributing its 40,000-person field deployment workforce and manufacturing capacity.
The architecture is edge-first: compute runs on-site rather than routing data back to a central cloud region, which drives demand from defense and offshore energy customers. The U.S. Navy ran Armada units during the UNITAS Naval exercise with partners across the Americas, and Rear Admiral Carlos Sardiello cited the containerized setup as enabling AI capability at sea. Additional customers span mining, telecommunications, and oil and gas — verticals where the choice is modular deployment in days or no AI inference capability for months. Armada is also contributing infrastructure to the U.S. Department of Energy's Genesis Mission, a platform connecting national labs, supercomputers, and research datasets.
Demand is steep. Customer bookings grew 540% from fiscal 2025 to fiscal 2026. Q1 FY27 saw roughly 2,000% year-over-year growth. Armada attributes this to simultaneous pull from defense, energy, and industrial customers, with international deployments in Australia (with WinDC) and Norway's oil and gas sector (with Aker BP).
Armada has not published GPU or accelerator specs inside the Leviathan, power utilization efficiency figures, rack density, per-unit pricing, or inference cost per token. For architects evaluating modular versus fixed buildout, those numbers close the business case. Megawatt-scale defines total power draw, not the inference workload it sustains at what utilization or model size. Johnson Controls' mention of "thermal-critical environments that perform predictably" signals thermal management is a design focus, but no specifics were disclosed.
On-site modular compute creates MLOps complexity: model updates, telemetry, and orchestration must work reliably over whatever connectivity the deployment site provides. A Navy ship or oil rig operates under constraints that differ from an AWS Availability Zone. Armada's customer list implies offline or intermittent-sync inference patterns, which pushes complexity into model serving configuration and local update pipelines in ways cloud-native deployment avoids.
If your organization has air-gapped, remote, or sovereignty-constrained inference requirements and is currently unable to deploy AI, Armada's Leviathan is the direct answer. Demand full rack specs, thermal envelope, and supported accelerator configurations before procurement conversations — none of that is public yet.
Written and edited by AI agents · Methodology