What is a GPU shipping container?
A GPU shipping container is a physical modular data center designed to house GPU servers and their supporting infrastructure. It brings the enclosure, rack layout, power distribution, cooling and network connections into a coordinated deployment project.
The term “GPU container” can also describe software used to run accelerated applications. This page is about the physical infrastructure: a container-format facility for AI training, inference and other GPU workloads.
QIT builds liquid-cooled GPU containers in this category: 1 MW and up to 1,200+ GPUs in a 20 ft container. We begin with your workload and the site, then define the hardware, thermal design and installation scope together. The container is one part of the facility; power delivery, heat rejection and connectivity need to be planned alongside it.
QIT GPU container specifications
Every QIT container is a complete AI data center in a 20 ft shipping container: 1 MW of IT capacity, up to 1,200+ density-packed GPUs and direct liquid cooling on every processor. It is built and tested in the factory, trucked to site and installed in 15 hours.
| Specification | QIT GPU container |
|---|---|
| Format | 20 ft shipping container |
| Dimensions | 5.8 m × 6.8 m × 2.44 m |
| IT capacity | 1 MW per container |
| GPUs | Up to 1,200+ density-packed GPUs |
| GPU platforms | Any architecture, including NVIDIA HGX B300, GB300 NVL72 and RTX PRO 6000 |
| Workloads | AI training and inference |
| Cooling | Direct liquid cooling, with a water block on every GPU and CPU |
| Heat rejection | Dry coolers on a sealed closed loop |
| Water | 1.5 m³ recirculated continuously; no water consumption or wastewater |
| Thermal stability | Within 2 °C under sustained load |
| Factory build | 3 weeks on average |
| On-site installation | 15 hours at a prepared site |
Heat leaves through a sealed loop to dry coolers, so the container draws no water and produces no wastewater. Each GPU and CPU has its own water block, which keeps processors at full speed under sustained load. See the liquid cooling guide and the 15-hour installation plan.
Start with the system, not the enclosure
A useful specification connects every system to the workload it supports. An enclosure alone does not establish GPU capacity, available power or a deployment date.
- Compute: GPU platform, server configuration, rack layout, service clearances and transport requirements.
- Cooling: server-side cooling requirements, coolant distribution, monitoring and the interface to site heat rejection.
- Electrical: rack demand, incoming supply, distribution, protection, grounding and any required backup systems.
- Network: workload fabric, storage connectivity, external bandwidth and management access.
- Operations: commissioning criteria, maintenance access and ownership of alarms and service response.
QIT scopes these interfaces before committing to a configuration. GPU models, rack count, power capacity and redundancy are project-specific, rather than fixed specifications published for every container.
What needs to be ready at your site?
Begin with a location, the power you can actually deliver and the date it becomes available. A container can move factory work offsite, but it still needs a prepared destination.
Site review should cover the foundation, delivery and crane access, local approvals, electrical connection, network routes and cooling interfaces. Ambient conditions and the heat-rejection approach influence the thermal design. Who supplies and commissions each interface should be recorded in the project scope.
Use the 15-hour installation plan to understand how site work and the container build can progress together. The schedule is confirmed against component availability and site readiness.
How much does a GPU container cost?
There is no useful single price without a hardware and site specification. The GPU and networking configuration, cooling equipment, electrical scope, transport and installation all affect the project budget. The container price and a fully commissioned site price can cover very different things.
Ask for a scope that separates compute equipment, container infrastructure, site works, logistics, commissioning and ongoing operations. Confirm whether the quote includes power and cooling outside the container, spare parts and service support.
For a QIT project, share your location, available power, workload and target date. We will use those inputs to discuss a configuration and a delivery plan.
For the broader facility decision, read our containerized data center guide. For the thermal system, explore liquid-cooled GPU containers.
Planning around 1 MW GPU blocks
Each QIT container delivers 1 MW of IT capacity with up to 1,200+ GPUs, which makes it a natural building block for a phased 5–10 MW site. Electrical overhead and resilience requirements still determine how many containers a given supply can carry.
Start with the available facility load, reserve the supporting infrastructure demand, and then size the compute phase. See the capacity example and project budget guide.
A GPU container is one form of modular data center. That guide compares containers with skids, prefabricated halls and other modular formats.
Prepared by QIT · Updated November 2, 2026
Explore all field notes