Modular Data Center Site Evaluation
Whether the site can keep the equipment running.
What it covers
Modular data centers ship 1-5 MW per module and reach first workload in 8-16 weeks after site prep, against 18-36 months for a traditional build. They trade that speed for thinner redundancy, staffing and security, which is what a lender has to check.
What you should check for
| Power and cooling | Power source and contract | Who supplies the power, at what price and for how long; behind-the-meter generation needs its own fuel and maintenance plan. |
| Density per module | Rated kW per rack inside the module against the GPUs: about 10-15 kW per 8-GPU server, about 120-140 kW per NVL72 rack. | |
| Redundancy | Backup generation, UPS and cooling redundancy against the uptime in the offtake contract. | |
| Physical | Placement | Modules bolted to a slab, about 25 ft apart with no fire bridges, no open sides, locks and extinguishers, the site fenced: the questions property insurers ask of containerized sites. |
| Security and staffing | Cameras, guards, alarm monitoring, and who is on site to swap a failed GPU. | |
| Network | Fiber routes and redundancy to the customer; inference tolerates distance, training clusters do not. | |
| Recovery | Who owns the module and the land | Whether the modules are part of the collateral, and the lender's access to the site in a default. |
| Movability | Whether modules can be relocated to another powered site, and at what cost. |
Sources: NVIDIA, GIGABYTE, American Compute.
Why modular sites need their own checks
Modular sites go where power is: stranded gas, surplus hydro, retired industrial sites with grid connections. A site with $0.03/kWh power from stranded gas can beat $0.08/kWh grid power near a metro.
The trade-off is everything around the power. A remote modular site won't match the network connectivity, staffing depth or physical security of a Tier III colocation facility. Many modular builds lack the redundancy that customer contracts assume.
Containers have failed before. Sun, Microsoft and Google all tried containerized data centers between 2005 and 2010 and moved away from the form factor: limited density, hard maintenance access, cooling in a sealed box. GPU density at 40-80 kW per rack makes cooling the first check.
Red flags
| Single power source, no backup | One generator or one feed carries the whole site and every SLA. |
| Air-cooled container for 40+ kW racks | Past roughly 40 kW per rack, direct-to-chip liquid cooling becomes mandatory. |
| Nobody on site | A remote site with no hands means every failed GPU waits for a truck. |
Discuss a transaction
Send us the equipment list and the loan or lease you have in mind. We will come back with an indication and the questions we would need answered.
Prefer email? hello@amcompute.com