Armada closed a $230 million oversubscribed Series B at a $2 billion pre-money valuation, lifting total capital raised to nearly $500 million and pairing the round with a manufacturing agreement that moves modular data centers from bespoke, one-off builds toward factory-scale output. The San Francisco company, which builds an edge-computing and satellite-enabled compute platform, announced the raise alongside a Johnson Controls framework deal on May 19.
The cap table is a who's who of infra and defense capital
The round was co-led by Overmatch, BlackRock and 8090 Industries. New strategic backers include Johnson Controls, NightDragon, Mitsui and Singtel Innov8, alongside existing investors Felicis, Marlinspike, Shield Capital, Lux Capital, Founders Fund, Silent Ventures, Veriten and Gladebrook. The mix — a top-tier asset manager, a building-systems incumbent, a Japanese trading house and a Southeast Asian telco arm — signals that demand is coming from defense, energy and sovereign buyers rather than the hyperscaler core.
Manufacturing is the actual product
The headline isn't just the cash; it's the supply chain. The "Galleon Forge One" agreement with Johnson Controls covers an Arizona facility of up to 400,000 square feet, expected to create more than 500 jobs, with production beginning summer 2026. The line will mass-produce Armada's Leviathan megawatt-scale modular data centers, purpose-built for high-density AI training, inference, sovereign neo-cloud and multi-tenant compute.
That's the strategic bet: while the industry waits years on gigawatt campuses tethered to grid interconnect queues, Armada is treating compute as a manufactured, shippable unit. Its portable systems — the Galleon and defense-deployed Triton lines — are designed to drop into environments with no specialized power infrastructure, drawing on solar, natural-gas flares and localized energy. Stated target environments include military operations, oil fields, mining sites and remote industrial facilities.
Demand curve is steep
Armada reports bookings up 540% between fiscal 2025 and 2026, and a 2,000% year-over-year jump in Q1 FY27. Co-founder and CEO Dan Wright framed the thesis bluntly: "The AI race will not be won by one-off projects. It will be won by the companies and countries that can manufacture, deploy, and continuously improve AI infrastructure, with speed, scale and sovereignty."
What changes for builders
The interconnect bottleneck and power-availability crunch are now the binding constraints on AI capacity, not GPU supply alone. Armada's pitch — standardized, transportable megawatt blocks that ship without waiting on substations — is a direct play on that gap, and the Johnson Controls deal gives it the production capacity to test whether modular edge compute can scale past pilots. For enterprise and government buyers shut out of the hyperscaler queue, or operating in disconnected and sovereign environments, a factory-produced data center is a materially different procurement model than leasing rack space. The open question is unit economics versus centralized campuses once Arizona is at volume.
