Armada Launches 10MW Orion Modular Data Center for AI Workloads

Key Takeaways

• The newly launched Orion delivers 10MW of capacity and supports up to 2,880 GPUs across 40 racks.
• The prefabricated system is designed for deployment within existing structures and alongside distributed power.
• Manufacturing is scaling as Fossefall prepares to deploy five Leviathan modules across the Nordic region.

Armada has launched Orion, a 10MW modular data center designed to compress the time required to deploy high-density AI and cloud infrastructure. The configuration is the largest system in the company's Galleon family and can support up to 2,880 GPUs across 40 racks.

Orion represents an almost 10x increase compared with the Leviathan launch last year. Leviathan provides 1.77MW of capacity and was itself roughly 10x larger than previous offerings. That pace of expansion illustrates how quickly modular infrastructure suppliers are adapting their designs for larger AI clusters.

According to DCD, Orion targets fast deployment of high-density compute for AI and cloud workloads. Each prefabricated six-module core integrates compute, networking, power and cooling rather than treating those systems as separate construction projects. The company reported that units can be delivered within months.

That timeline is central to the pitch. Conventional data center developments can involve lengthy permitting, utility interconnection and on-site construction processes. A factory-built configuration does not remove every site constraint, but it can shift more assembly and testing into a controlled manufacturing environment.

Orion is designed for installation within existing structures and can be scaled across distributed sites to hundreds of megawatts. The system can also be colocated with distributed energy resources, including stranded or behind-the-meter power.

Access to GPUs is only part of the AI infrastructure bottleneck. Operators also need available electricity, cooling capacity, network connectivity and a physical site capable of accommodating dense racks. When those components cannot be assembled quickly enough at a conventional campus, standardized modules located near usable power can provide another route to deployment.

The hardware architecture is compatible with Nvidia Blackwell systems and is described as Vera Rubin-ready. That gives prospective operators a migration path as accelerator platforms evolve, although customers will still need to evaluate rack density, cooling design, network topology and power availability for their selected hardware.

“AI infrastructure should be fast to deploy, easy to scale, and sovereign wherever it needs to live,” said the co-founder and CEO of Armada. The chief executive added that customers can own and operate Orion on their land and power, positioning the system as an alternative to consuming equivalent capacity solely through public cloud services.

Ownership transfers operational responsibility to the customer and supports data residency, infrastructure sovereignty or predictable access to compute for governments and regional cloud providers. Buyers will need to examine availability targets, maintenance arrangements and resilience against frameworks such as the Uptime Institute Tier Standard: Topology and ISO/IEC 22237.

The launch comes as demand for prefabricated capacity accelerates. Stratview Research projects that the global modular data center market will grow from approximately $36.75 billion in 2025 to $111.62 billion by 2032, representing a 17.2% compound annual growth rate. Other estimates place 2025 revenue between $28 billion and $33 billion, with forecasts generally pointing to 17% to 19% annual growth through 2030 to 2033.

Longer-range projections are larger still. NextMSC forecasts that the broader modular data centre market could reach $233.62 billion by 2035, based on an 18.13% compound annual growth rate from 2026 through 2035. Forecast methodologies differ, but the direction is consistent: prefabrication is moving deeper into hyperscale, enterprise and edge deployments.

Established competition exists from Schneider Electric and its EcoStruxure-modular portfolio, as well as Vertiv and Huawei. Orion’s differentiators will therefore need to extend beyond raw capacity. Delivery performance, supply-chain resilience, cooling efficiency, accelerator compatibility and support across multiple jurisdictions could all influence procurement decisions.

A regional test is already underway. Earlier this week, a contract was announced with Nordic-based data center operator Fossefall for five Leviathan modules. The systems will provide 8.85MW in total and are scheduled for delivery by the first quarter of 2027. The agreement, valued at several hundred million NOK, will support projects in Norway, Sweden and Finland, with Fossefall owning and operating the GPU infrastructure.

Following a $230 million funding round, manufacturing space is expanding with the development of Galleon Forge One alongside Johnson Controls. The site is set to span up to 400,000 square feet, or 37,161 sqm, create 500 jobs and enable continuous module production. Commercial impact will now depend on whether that factory-based model can translate ambitious capacity claims into repeatable deployments measured in months rather than years.