Armada modular
data centers vs QIT.

What the Armada modular data center line includes

Armada, often searched as Armada AI (armada.ai), builds factory-built modular data center systems under the Galleon name and pairs them with software for edge computing and AI workloads. Its site describes the Armada Edge Platform, Atlas (an operational insights platform for connected assets), Bridge (a cloud and GPU-as-a-Service layer) and a Marketplace. The company’s markets include operators in the oil and gas industry, defense, state and local government, mining, telecommunications, neoclouds and sovereign AI infrastructure.

The Galleon range spans from small ruggedized units for remote locations and challenging environments to megawatt-scale systems. Armada’s configurator lists a range from a 300 W Beacon to 50 MW. Edge computing in this sense means processing data locally, near where it is generated, which reduces bandwidth usage at sites far from fiber. Two products matter most for a GPU buyer evaluating Armada alternatives: Leviathan and Orion.

Sources: Armada website and Armada Leviathan product page. Reviewed November 2, 2026.

Galleon modular data centers: Leviathan and Orion

Leviathan

Armada describes Leviathan as its largest Galleon form factor for GPU-dense AI infrastructure. Each unit integrates compute, networking, liquid cooling, UPS-backed electrical distribution and monitoring. The published specifications are:

• IT envelope: 1.77 MW per unit, with 576 NVIDIA GB300 NVL72 GPUs per unit.
• Cooling: liquid-first, closed-loop direct-to-chip, about 80/20 liquid to air, with air-cooled heat rejection through N+1 chillers and 2.2 MW of cooling capacity.
• Footprint: about 119 × 45 ft (0.12 acre) per unit. Armada states that Leviathan is not a shell, enclosure or containerized room.
• Scale: a four-unit pod of 5.35 MW, multi-unit sites of 5–20 MW and a stated path to 50 MW+.
• Schedule: “12 weeks from site-ready.” Armada states that Leviathan deploys on prepared land with available utility power and network connectivity.

Armada contrasts this with 18–24 months for a traditional data center build. The 12-week figure starts once the site is ready, so power, permitting and civil work sit before it.

Orion

Orion, launched on August 26, 2026, is described as a 10 MW Galleon built around a prefabricated six-module core, with up to 2,880 GPUs across 40 racks. The release describes it as deliverable in months from deposit. Its cooling design was not specified in the material we reviewed.

Sources: Armada Leviathan page; Orion launch release (PR Newswire, via Webull), August 26, 2026. Reviewed November 2, 2026. Specifications are Armada’s and may change.

Announced Armada deployments

In August 2026, Nordic developer Fossefall selected Armada for five Leviathan units totaling more than 9 MW, with delivery by Q1 2027. Fossefall stated it will own and operate the GPU infrastructure, which indicates a hardware delivery model rather than cloud capacity.

In Australia, WinDC deploys relocatable modular AI data centers built by Armada at renewable generation sites. Trade press reported an 11 MW partnership in March 2026.

Sources: DCD on Fossefall; WinDC website and W.Media on WinDC–Armada. Reviewed November 2, 2026.

Johnson Controls and modular data center production

In May 2026, Johnson Controls published “Armada announces agreement with Johnson Controls for modular data center production.” The framework agreement outlines plans for Galleon Forge One, a dedicated factory in Arizona of up to 400,000 square feet. Johnson Controls, which describes itself as a global leader in thermal management and mission-critical building systems, expects the site to create 500 jobs, with additional roles in the domestic supply chain.

Continuous production is planned to start with Leviathan, for high-density AI training and inference workloads, sovereign neo-cloud and multi-tenant compute environments.

“The AI race will not be won by one-off projects,” said Dan Wright, co-founder and chief executive officer of Armada. “It will be won by the companies and countries that can manufacture, deploy, and continuously improve AI infrastructure, with speed, scale and sovereignty.”

For buyers, a dedicated factory matters mainly for delivery slots. Ask Armada how Johnson Controls production capacity maps to your delivery window, and ask any vendor where your order sits in the queue.

Source: Johnson Controls, May 22, 2026. Reviewed November 2, 2026.

How Armada and QIT modular data centers compare

Armada and QIT both build modular data centers for AI infrastructure that trains and serves AI models, but the units are defined differently. The table separates published Armada specifications from QIT’s project approach.

Compare schedules only when they start from the same event. A “site-ready” clock and a project clock that includes site preparation measure different things.

Edge computing or GPU containers: when each option fits

Armada may fit when you need ruggedized edge computing in remote locations and other demanding environments, a software layer such as the Armada Edge Platform for fleet operations, sovereign AI infrastructure, or a unit sized around a published 1.77 MW or 10 MW block with a fixed reference GPU configuration that can deploy quickly once the site is ready. Its line also covers small Galleon units that QIT does not offer.

QIT may fit when you have an existing powered property in Texas, want liquid-cooled GPU containers planned in roughly 1 MW blocks, and prefer to phase a 5–10 MW AI factory as power and demand arrive. QIT scopes the GPU platform, enclosure and site interfaces together for each project.

Other modular data center companies also build at this scale. See our modular data center companies guide for a wider list, including Flexnode and Crusoe Spark.

Questions to ask before you accelerate deployment

• What exactly does “site-ready” mean, and who delivers each item on that list?
• Is the quoted capacity IT load or total facility load, and what cooling overhead applies at your design ambient temperature?
• Which GPU platform and rack configuration are fixed, and which can change?
• Will you own the hardware, lease it or buy capacity as a service?
• Who operates the units and responds to faults after handover? What real-time monitoring and predictive maintenance does the operations software provide?
• How many deployed units are running today, and in what kinds of mission critical environments?
• What event starts the delivery clock, and what moves it?
• How is full-system acceptance defined and documented?

Related guides and next steps

Before comparing quotes, establish the power you can use and the boundary of each scope. Useful starting points:

• Modular data centers for AI
• Liquid-cooled GPU containers
• Modular data center cost
• Powered land vs powered shell
• QIT’s proposed deployment stages

QIT is independent of Armada. This page does not claim equivalent capacity or measured performance. Discuss your site with QIT.