FAQ
Frequently asked questions
Common questions about products, delivery times, cooling and global compliance.
What is SwiftMDC?
SwiftMDC is SINOWISE's container data center. Racks, power distribution, liquid cooling, monitoring and fire protection are integrated and tested in the factory, shipped to site as complete containers and put into service once power and cooling are connected. Compute, networking, liquid cooling, power and monitoring are delivered as one deployable system, either as infrastructure only or as an integrated version delivered with servers.
How much power can SwiftMDC cover?
Three configurations cover 80 kW to 20 MW: AI-Edge, a 20 ft container with 80–200 kW and 8–16 GPUs; AI-Core, a 40 ft high-cube container with 300–600 kW and 32–64 GPUs; and AI-Flex, 4 to 20+ interconnected 40 ft modules with 2–20 MW and 256 to 2,048+ GPUs. Rack density goes up to 120 kW.
How long does delivery take?
The target is 8–12 weeks for a single container and 3–5 months for megawatt-scale projects, counted from when the site and power are ready. Final timing follows the project contract.
How large is one POD?
Taking 8-GPU B300 servers as the example, one POD holds 1,024 GPUs in 128 servers with an IT load of about 2 MW and cold-plate liquid cooling. The module measures 21.6 × 7.2 m, a footprint of about 155.5 m².
Which cooling methods are supported?
Air cooling, cold-plate liquid cooling and immersion cooling. Liquid-cooled racks come in cold-plate and immersion versions; a site without liquid cooling can start with the air-cooled rack and convert the same rack to cold-plate cooling later, without changing the layout, power or cabling.
What specification are the liquid-cooled racks built to?
They are built to the NVIDIA MGX v1.1 rack geometry, in three models: liquid-cooled, liquid-cooled with a side-mounted CDU, and air-cooled. Racks are cabled, fitted with manifolds, leak-tested and burned in at the factory; on site they only need to be placed and connected to power and coolant.
How large can modular data halls and campuses be?
From 1 MW to 1 GW, in three tiers: micro (under 7.5 MW), city (7.5–20 MW) and hub (100 MW–1 GW), for new builds, retrofits and ongoing expansion. The 8 MW reference design has a design-estimated annual average PUE of 1.204.
Can projects be deployed in other countries?
Yes. Key components are selected from the latest NVIDIA approved vendor list (AVL), and projects are prepared for the grid, certification and installation codes of ten regions: North America, Europe, Japan, South Korea, Southeast Asia, South Asia, the Middle East, Africa, Latin America and Oceania. Other markets are assessed project by project. Certification and testing paths are planned per project, for example NRTL listing in North America (with reference to UL 2755), CE and UKCA in Europe, and PSE and VCCI in Japan; the certification scope and status of specific products follow the project documents.
What does the site need to provide?
The site provides the footprint and service space, transport and lifting access, the utility power connection, the network connection, and local permits and external utilities. SINOWISE delivers the containers, liquid-cooled racks and modular compute units; power distribution, cooling sources and liquid-cooling piping; factory integration, on-site assembly and system testing; central monitoring and the digital twin; and maintenance and expansion support.
What services do you provide?
The full lifecycle of design, delivery and operations: assessment and design (including digital-twin simulation), factory integration and testing, on-site delivery and acceptance, and operations and expansion (maintenance, repairs and spare parts, and performance tuning).
How do I start a project?
Use the configurator on this website to get a first-pass configuration and delivery estimate from power, cooling and region, or request a site assessment: tell us about the site, available power and target scale, and we'll recommend racks, containers or a campus and prepare a proposal. Email: contact@gpuaidc.com