Two Launches, One Direction
On 21 October, at GTC Berlin, NVIDIA unveiled DSX POD, a modular AI factory reference design built around its Vera Rubin platform. The same day, Schneider Electric published the RapidScale Modular Data Center Reference Design 115 for the same platform. Both point to one conclusion: AI infrastructure is shifting from item-by-item site installation to factory-integrated, factory-tested, whole-module delivery.
1. What DSX POD Changes
DSX POD is neither a server nor a CDU. It is a reference design defining how computing racks, networking, rack-level power and liquid cooling piping are assembled and functionally verified in a factory before shipping to site as dedicated modules.
NVIDIA states that, where site power, cooling and network capacity are ready, deployment can be compressed from roughly six months to as little as 14 days, with partner solutions planned from early 2027. The structural change is where the work happens: cabling, piping and verification move upstream, module manufacturing runs in parallel with civil works, and duplicated on-site integration disappears.
Figure 1 — Site-built sequence versus factory-integrated module delivery.
2. Prefabrication Is Not a New Market
Prefabrication spans several scopes, from factory welding, cleaning and pressure testing of coolant piping, through pre-assembled CDUs, cold sources and racks, to integrated modules combining power, cooling and IT equipment. The scope covered defines both product position and delivery responsibility.
Vertiv targets high-density modular infrastructure delivered as a whole, with 2026 configurations of its MegaMod HDX range supporting up to 10 MW. Schneider Electric draws on power distribution, liquid cooling and prefabricated modules to build platform-specific reference designs, while Delta combines power and liquid cooling into prefabricated AI data center solutions. Chinese suppliers address the market with Kehua's WiseMDC liquid-cooled POD, Shenling's SKY-ACMECOL modular cold source, CDU and piping system, and Envicool's XSpace range.
| Vendor | Primary position | Integration scope |
|---|---|---|
| Vertiv | High-density modular infrastructure | MegaMod HDX configurations up to 10 MW |
| Schneider Electric | Power, liquid cooling, prefabricated modules | RapidScale Reference Design 115 for Vera Rubin |
| Delta | Power and liquid cooling portfolio | Prefabricated AI data center solutions |
| Kehua Data | Liquid-cooled POD | WiseMDC combined with power distribution |
| Shenling Environment | Liquid cooling integration | SKY-ACMECOL cold source, CDU and piping |
| Envicool | Equipment and modular applications | XSpace modular data center and CDU |
None of these vendors starts from zero; the question is whether existing products can meet the unified requirements of a high-density AI platform.
3. From Single Equipment to Complete System
The innovation lies less in prefabrication than in extending platform requirements into infrastructure design, integration and validation. The scope covers rack layout, power interfaces, coolant piping, network connections, and transport and maintenance clearances; partners must validate a complete module, not individual units.
This mirrors a trend already visible in liquid cooling. Large AI projects increasingly evaluate cold source, CDU, piping and server-side loops together to reduce interface and responsibility gaps between vendors. Capacity and efficiency remain essential, but piping cleanliness, interface compatibility, control coordination, transport reliability, factory acceptance testing and field service now decide admission to a full-system solution.
Figure 2 — From single-unit performance to system-level delivery capability.
A unified architecture lowers repeated design cost but raises adaptation and validation requirements. NVIDIA defines the architecture only; engineering, manufacturing, deployment and service stay with partners, so the platform owner is more likely to reshape collaboration than replace established vendors.
4. Schneider Electric's Parallel Move
Schneider Electric's RapidScale Reference Design 115 integrates power, liquid cooling, IT space and services around Vera Rubin and names DSX POD support as the next step — a redirection of existing prefabrication capability rather than a reactive product.
A joint case study with Compass Datacenters found that, in a high-density liquid-cooled project using prefabricated IT modules, on-site installation time per module fell from 75.5 hours to 15 hours, about 80 per cent. The figure is not a DSX POD measurement but quantifies the engineering value of prefabricated delivery.
NVIDIA's ecosystem lists computing partners such as Dell, HPE, Lenovo and Wiwynn alongside infrastructure providers including Vertiv, Schneider Electric, Delta and Eaton. Whole-Pod delivery requires joint work across computing, power and liquid cooling, and ecosystem listing does not equal confirmed orders.
5. Implications for the Supply Chain
- Integrated vendors can connect existing prefabricated products to next-generation AI factory projects, but must deliver deeper interface adaptation and system validation.
- CDU, cold source, piping and connector suppliers gain new supporting demand, while qualification shifts from product performance to interface compatibility and batch delivery.
- Owners and EPC contractors see acceptance move toward the factory gate, shortening schedules but concentrating technical risk in fewer, larger supply relationships.
DSX POD does not create the prefabricated data center market, but it brings platform owners into the rules governing its design and delivery. What matters next is how many specialists can convert liquid cooling, power and module manufacturing capability into validated, batch-deliverable AI infrastructure.
