Key takeaways
- A pop-up data center is a set of prefabricated modules. Each holds racks, power distribution and liquid cooling. They are placed on a pad next to a power source, not inside a building.
- The modules are fast. The power usually is not. Vendor timelines start when the site already has power and fiber. Getting that power is often the longest step.
- It fits a powered site that would otherwise sit idle, a GPU requirement that cannot wait for a new building, or capacity you may need to move later.
- It fits poorly when you need certified redundancy on day one, hundreds of megawatts in one place, or a landlord who carries the operating risk.
What is a pop-up data center?
A pop-up data center is a data center delivered as finished modules instead of a building. Each module is an enclosure, often container-sized or larger. Racks, power distribution, cooling, fire suppression and monitoring are installed and tested in a factory. The modules ship by truck, sit on a concrete pad or skids, and connect to power and fiber on site.
The industry uses several names for close variants: modular, prefabricated, containerized, skid-mounted and rapid-deploy. "Pop-up" describes the outcome. A site with power gets usable compute without a construction project. Some units are fully self-contained. Others are prefabricated power and cooling blocks that serve racks in a simple structure. Always confirm what the module includes.
It sits beside the other ways to get wholesale capacity:
| Model | What you get | Who builds it |
|---|---|---|
| Turnkey wholesale | A finished hall in an operator's building | The operator. You install IT equipment |
| Powered shell | A building with utility power. You fit it out | The landlord builds the shell. You build the inside |
| Build-to-suit | A facility designed and built to your specification | A developer, under a long lease |
| Pop-up or modular | Factory-built modules placed on a powered site | A module vendor or provider. The site owner supplies power and pad |
How a pop-up deployment works, step by step
- Find the power. Everything starts with a site that has spare electrical capacity. Common examples are industrial plants with unused utility service, substations with headroom, solar and storage sites, and sites with on-site gas generation. A site that already has power and fiber is the "site-ready" point vendors measure from.
- Agree the site deal. The site owner provides land, power and access. The provider or tenant brings the modules. The deal sets the power price, the term, who pays for the pad and interconnection, and what happens to the modules at the end.
- Permit the site. Expect building, electrical and fire permits, and air permits for any generators. Zoning still applies. Modules are equipment on a pad, but expect the local jurisdiction to review them.
- Prepare the pad. Foundations, grading, drainage, the electrical connection, fiber, and space for heat rejection such as dry coolers.
- Build in the factory while the site is prepared. This parallel work is where the time saving comes from. In a conventional build the building has to exist before the inside can be installed.
- Deliver, set and connect. Modules are craned into place, connected to power, cooling loops and fiber, then commissioned.
- Operate. The units described below include remote monitoring. Hands-on work still needs a field team. Someone must own maintenance, spares, security and response times. Put it in the contract.
How big are pop-up data centers?
Single modules range from a few hundred kilowatts to around 10 megawatts. Sites grow by adding modules. AI racks in these units are liquid-cooled. Nvidia's GB200 NVL72 rack platform runs up to about 132 kW per rack in Vertiv's reference design (October 2024). Newer platforms run higher. These are published examples. They are vendor specifications and announcements, not a claim that capacity is available in any of them.
| Provider and product | What the vendor states | Status |
|---|---|---|
| Armada Leviathan | 1.77 MW of IT load and 2.2 MW of cooling per unit. 8 compute racks at 200 kW plus 4 networking racks. About 576 GPUs per unit. Up to 8 units in a cluster. Footprint about 0.12 acres. Armada, accessed October 2026. | Product specification. No public count of energized units found |
| Vertiv MegaMod HDX | Prefabricated power and liquid cooling. A compact module holds up to 13 racks and 1.25 MW. A combined module holds up to 144 racks and 10 MW. Racks from 50 kW to more than 100 kW. Vertiv, January 14, 2026. | Product maximums |
| Crusoe Spark | Units integrate power, cooling, monitoring, fire suppression and high-density racks. Deployments scale from hundreds of kilowatts to tens or hundreds of megawatts. Built in Brighton, Colorado. Crusoe, March 12, 2026. | Operating in Nevada since June 2025 on a solar and battery microgrid, per Crusoe. Expansion from 4 to 24 units announced March 2026. A site of up to 25 MW in Snyder, Texas announced February 2026 |
| Schneider Electric EcoStruxure Pod | Factory-built pods with power and cooling for high-density racks. Supports liquid cooling through coolant distribution units. Schneider Electric, accessed October 2026. | Product line. Deployed inside or beside buildings |
How fast is a pop-up data center, really?
Vendors quote short timelines, and they are real for the module. Armada says a Leviathan is operational 12 weeks from site-ready, against 18 to 24 months for a traditional build. Crusoe says delivery can take as little as three months.
Read the starting line. "Site-ready" means the land already has utility power and network connectivity. The clock does not include finding power, signing the utility agreement, permitting or pad work. Those steps decide your real date.
Power is the slow part. JLL puts the average wait for a grid connection in primary data center markets at more than four years (2026 outlook, January 2026). Enverus says connection routinely takes five years or more in constrained markets (May 2026). That is why pop-up deployments go where power already exists, or bring their own generation.
A realistic way to plan the date
- Site has energized spare capacity and fiber: the vendor timeline plus permitting and pad work.
- Site needs a utility upgrade: the utility's committed date sets the schedule. The modules will wait for it.
- Site relies on new on-site generation: generator lead times and air permits set the schedule.
What does a pop-up data center cost?
No provider publishes an all-in price per megawatt for liquid-cooled AI modules. Pricing is quoted per project. These public figures frame it.
- The conventional benchmark. JLL puts average global construction cost at $10.7 million per MW in 2025 and forecasts $11.3 million in 2026. That covers shell and core only. JLL says AI fit-out can add up to $25 million per MW. Our construction cost guide breaks this down.
- Where modular saves money. In a vendor model, Schneider Electric found about 27% lower capital cost for a prefabricated design. More than half of that came from not building capacity before it was needed. The model predates AI-era designs, so treat it as direction, not a quote.
- Where it does not. Per megawatt, the module itself is not cheap. The saving is time, phasing and avoided building cost. If a powered building already exists, a conventional fit-out may cost less.
Compare the full cost over the term: module cost or lease payments, the power price, site rent or revenue share, pad and interconnection, operations, and the value of earlier revenue from GPUs that go live sooner.
How pop-up deals are structured
- Buy the modules
- You own the equipment, place it on a site you control or lease, and operate it or hire an operator. Most control, most capital.
- Lease or capacity service
- A provider owns and runs the modules on a site, and you pay for capacity or compute. Less capital, less control over design and location.
- Site partnership
- A landowner or power owner supplies the pad and power, and shares the economics with the provider. Energy Vault and Crusoe announced this kind of structure for a Texas solar and storage site in February 2026. If you control a powered site, our landowner and developer path is the place to start.
Pop-up vs powered shell vs turnkey
| Pop-up or modular | Powered shell | Turnkey wholesale | |
|---|---|---|---|
| Speed once power exists | Fastest: months | Slower: fit-out on your critical path | Fast if a hall is built and available |
| Depends on | A site with spare power | A shell with committed utility power | An operator with space at your density |
| Scale in one place | Grows in module steps. Limited by the site's power | Large, set by the building | Set by the operator's campus |
| Redundancy | Varies by design and site. Confirm it in writing | What you build | The operator's standard, usually concurrently maintainable |
| Can it move? | Yes, at a cost | No | No |
| Operating burden | Yours or the provider's, by contract | Yours | The operator's |
Risks to price in
- Redundancy. A single utility feed and a single module have single points of failure. Uptime Institute's TIER-Ready program reviews modular designs in advance. Each site still needs its own certification.
- Power quality and backup. Behind-the-meter generation, solar and batteries need a plan for outages and maintenance. Ask what happens to your racks when the source goes down.
- Site and weather. Modules sit outdoors. Ask about wind, flood, heat and snow ratings, perimeter security and access for repairs.
- Scale. The site's power caps growth. A second phase may mean a second site.
- Financing. In AI infrastructure deals, equity holders rather than lenders tend to carry technology, ramp-up and customer concentration risk, according to J.P. Morgan (August 2026). Relocatable equipment on someone else's land needs clear title and access terms.
- End of term. Who removes the modules, who restores the site, and whether the equipment has resale value.
When a pop-up data center fits, and when it does not
| Good fit | Poor fit |
|---|---|
| You control or can reach a site with spare energized power | No site has spare power. The utility timeline will dominate anyway |
| A GPU deployment of a few megawatts that cannot wait for a new building | A single campus of hundreds of megawatts |
| Training or batch workloads that tolerate lower redundancy | Workloads that need certified concurrent maintainability from day one |
| Capacity you may need to move, or a bridge until a permanent site is ready | A team that wants an operator to carry all facility risk |
Questions to ask a pop-up provider
- Is power energized at the site today? How many megawatts are spare, and what is the source?
- When does your timeline start, and what does "site-ready" include?
- How many of these units are energized and running today, and where?
- What rack density and cooling does the unit support, and for which GPU platform?
- What is the redundancy design for power, cooling and network? Is it Tier certified?
- Who operates and maintains the units, and what are the response times?
- What happens at the end of the term: removal, restoration, purchase option?
Frequently asked questions
Is a pop-up data center the same as a modular data center?
Close. Modular describes how it is built: in factory modules. Pop-up describes how it is used: placed on a powered site fast, without a building. Most pop-up deployments are modular. Not every modular data center is a pop-up, because modules are also installed inside conventional buildings.
How long does a pop-up data center take?
Vendors quote around three months from a site that already has power and fiber. The full timeline from contract to energized capacity depends on power, permits and pad work, and is usually longer.
Can a pop-up data center run Nvidia GB200 or GB300 racks?
Units designed for AI use direct-to-chip liquid cooling and support rack densities above 100 kW. Confirm the specific platform, cooling design and power per rack with the provider before you commit.
Can I put a pop-up data center on my land?
If the land has spare power, fiber nearby and a workable permitting path, it may suit a pop-up deployment. Tell us about the site and we will say what it would support.
Do you charge tenants to evaluate pop-up options?
No. Our advice is free to tenants. We are paid by the provider you choose, and we disclose that on every shortlist. How we are paid.
Sources
- Armada, Leviathan product page, accessed October 5, 2026. Vendor specifications and delivery target.
- Vertiv, MegaMod HDX launch release, January 14, 2026. Product maximums.
- Schneider Electric, EcoStruxure Pod Data Center, accessed October 5, 2026, and the launch release, June 19, 2025.
- Crusoe, new manufacturing facility for modular AI factories, March 12, 2026.
- Crusoe, Crusoe and Redwood Materials expand partnership, March 24, 2026.
- Energy Vault and Crusoe, strategic framework agreement for Crusoe Spark units, February 11, 2026, as filed with the SEC.
- JLL, 2026 Global Data Center Outlook, January 5, 2026.
- Enverus, Why data centers are looking to natural gas for behind-the-meter power, May 28, 2026.
- Schneider Electric, White Paper 164, TCO analysis of a traditional data center vs a scalable, prefabricated data center, revision 2, December 19, 2023. Vendor model.
- Uptime Institute, TIER-Ready program, accessed October 5, 2026.
- J.P. Morgan, Financing AI infrastructure and data centers, August 10, 2026.
- DCD, Vertiv and Nvidia reference architecture for GB200 NVL72, October 2024.