Armada will manufacture standardized, factory-integrated superintelligence infrastructure based on the NVIDIA DSX POD reference design for the Vera Rubin platform. Together with Johnson Controls, the company plans to build 500 MW of distributed, sovereign capacity in 2027, according to its announcement.
In the race for AI capacity, the bottleneck is increasingly the building rather than the chip. At NVIDIA GTC, Armada, a provider of sovereign superintelligence (SI) infrastructure, announced its support for the NVIDIA DSX POD reference design. The approach moves cabling, wiring, piping and functional testing from the construction site into the factory.
With DSX POD, partners can develop standardized, factory-integrated container solutions to deploy SI compute faster and at scale. Armada and Johnson Controls plan to develop, manufacture, integrate and deliver prefabricated SI factories based on the NVIDIA Vera Rubin platform. All cabling, electrical wiring and piping work, as well as functional tests, are completed at the factory before the individual units ship.
Armada joined the NVIDIA Inception program in 2024 and began working with NVIDIA’s network simulation technologies in the same year. Today, the company uses the NVIDIA DSX platform across the lifecycle of an SI factory, from simulation to operation. NVIDIA DSX Air simulates the AI factory and validates Armada’s Bridge GPU management software before hardware is delivered. The DSX POD reference design underpins container manufacturing. Once a site is running, NVIDIA DSX MaxLPS maximizes token generation within its power limits. The Armada platform also supports NVIDIA AI Grid, which lets telecom operators, service providers and enterprises deploy and operate geographically distributed AI infrastructure.
The company points to a familiar problem. According to Armada, a conventional 50 MW data center can take years to become ready for IT installation and requires around 600,000 on-site working hours. At the same time, the construction industry is projected to be short of roughly 288,000 workers by 2027. Permitting, grid connection and local approvals extend timelines further. Even organizations that already have chips, land, power and capital still have to wait.
“The race for superintelligence will not be won by whoever pours concrete fastest,” said Dan Wright, co-founder and CEO of Armada. The company builds the SI factory at one of its Galleon Forge manufacturing sites, ships and connects it, so customers can generate tokens within months on infrastructure they own and control, he added.
In practice, the reference design shifts standardizable work into the factory. Compute and networking, rack-level power, liquid-cooling piping and supporting infrastructure are integrated and tested before delivery. On site, Armada connects the containers to prepared power, water, cooling and network connections. As demand grows, customers can add further standardized containers. NVIDIA DSX MaxLPS is meant to help them use available power more efficiently across sites through policy-based power control, dynamic workload orchestration and high-precision telemetry.
“Demand for compute is growing faster than the world can build the facilities for it,” said Charlie Boyle, vice president of DGX Systems at NVIDIA. With the reference design, Armada can supply factory-built Vera Rubin containers wherever customers’ energy and data resources are located.
Armada cites experience with data centers in unusual locations: in Alaska, aboard a US Navy littoral combat ship, at desert sites in the Middle East and, in future, on a North Sea oil platform. Johnson Controls CEO Joakim Weidemanis said the partnership combines Armada’s edge-compute expertise with his company’s thermal management technology, engineering and global manufacturing capacity.
Armada describes its approach as a distributed network of modular SI factories rather than a single large facility. The pitch is aimed at governments and enterprises that want to avoid renting capacity from providers outside their jurisdiction. Sites are located near energy sources, including previously unused energy, renewables and off-peak surplus power. Each site runs on an open architecture without cloud dependencies, so customers keep control over their models, data and the infrastructure underneath.
Responsibilities are split. NVIDIA develops, tests and validates the DSX POD reference design for Vera Rubin. Armada handles engineering, manufacturing, integration, customization, maintenance, support and delivery.
Recent company announcements add context. On 11 November, Armada and Johnson Controls reported the start of production at Galleon Forge Two in Dublin and the completion of a first Leviathan for Fossefall in Norway. Fossefall has placed an initial order for five Leviathan units totaling more than 9 MW, with delivery scheduled by the first quarter of 2027. The companies are targeting at least 500 MW of capacity in 2027. Earlier, on 1 October, Palantir named Armada its inaugural certified modular data center partner. How quickly the factory approach converts into operating capacity remains to be seen: the timelines, capacity figures and labor estimates cited here come from the companies themselves and have not been independently verified.
Dr. Jakob Jung is Editor-in-Chief of Security Storage and Channel Germany. He has been working in IT journalism for more than 20 years. His career includes Computer Reseller News, Heise Resale, Informationweek, Techtarget (storage and data center) and ChannelBiz. He also freelances for numerous IT publications, including Computerwoche, Channelpartner, IT-Business, Storage-Insider and ZDnet. His main topics are channel, storage, security, data center, ERP and CRM.
Contact via Mail: jakob.jung@security-storage-und-channel-germany.de