Armada said it plans to build 500 megawatts of distributed AI computing capacity with Johnson Controls in 2027, giving their factory-built data center partnership a proposed scale.
The figure is Armada's goal, not capacity reported as operating or contracted. Its Oct. 21 announcement did not name deployment sites or customers, set out project financing for the proposed capacity, or say how the 500 megawatts would be divided among locations.
Armada plans to use NVIDIA's DSX POD reference design for the Vera Rubin platform. Partners would assemble and test computing equipment in a factory along with the hardware needed to connect, power and cool it. NVIDIA says factory work can proceed while sites are being prepared. It identifies the sequence of work normally done at a site as a delay the design aims to reduce, while acknowledging that prefabricated data centers already exist.
NVIDIA estimates deployment at a prepared site could take as little as 14 days, compared with six months in its comparison. That is a conditional estimate, not a measured result from Armada's proposed capacity.
The estimate does not include all the work needed to ready a site. NVIDIA's technical description says crews still need to prepare foundations, power and network connections, water and heat-rejection systems, and complete commissioning. NVIDIA says an earlier start could let customers generate tokens and revenue sooner, but it has not quantified that benefit.
Johnson Controls announced a framework agreement with Armada for modular data center production in May and disclosed an investment in the company. Armada's new capacity goal adds a year and a proposed scale to that relationship, but its Oct. 21 announcement gave no timetable for individual deployment sites.
NVIDIA named Armada as a DSX POD partner and said it expects partner deliveries worldwide to begin in early 2027.
