How to build a 100 MW hyperscale AI data center: STULZ, Siemens and NVIDIA blueprint

AI workloads are pushing rack power densities, cooling demand and electrical dynamics far beyond traditional data center design limits. Operators planning new capacity need a repeatable way to combine compute, power, cooling and automation without lengthy custom engineering for every site. A new reference blueprint from Siemens, STULZ and NVIDIA addresses exactly that challenge for IEC markets. Softprom is the official distributor of STULZ.

What was announced

Siemens, STULZ and NVIDIA have jointly developed a reference blueprint for a 100 MW hyperscale AI data center designed for IEC markets. The architecture is built around NVIDIA GB300 NVL72 systems and also accommodates GB200 NVL72 deployments. It integrates Siemens electrical infrastructure and automation, STULZ mission-critical cooling and NVIDIA AI compute into a single modular concept intended to accelerate time-to-compute and simplify repeatable rollouts.

STULZ contributes the thermal backbone: CyberCool SCS 15102 WU Cooling Distribution Units deliver liquid cooling directly to the compute racks, while ABR 2500 CW Computer Room Air Handlers cover air cooling for support racks and residual heat loads.

Why this matters

For CIOs, data center directors and procurement leaders, AI capacity planning has shifted from a server question to an infrastructure question. Rack densities above 100 kW, dynamic GPU power profiles and hybrid liquid-air cooling demand a coordinated design across vendors. A validated reference architecture reduces integration risk, shortens design cycles and provides a defensible basis for capital planning on 100 MW-scale builds.

Technical details

• Scale: 100 MW hyperscale reference design for IEC markets
• Compute target: NVIDIA GB300 NVL72, compatible with GB200 NVL72
• Liquid cooling: STULZ CyberCool SCS 15102 WU Cooling Distribution Units at the rack
• Air cooling: STULZ ABR 2500 CW Computer Room Air Handlers for support loads
• Power and automation: Siemens electrical infrastructure and automation stack
• Architecture: modular, repeatable, scalable to project requirements

FAQ

What is the 100 MW AI data center blueprint?

It is a jointly developed reference architecture from Siemens, STULZ and NVIDIA that defines how power, cooling, automation and AI compute integrate at 100 MW hyperscale for IEC markets.

Which NVIDIA platforms does the STULZ blueprint support?

The design is centered on NVIDIA GB300 NVL72 systems and also supports GB200 NVL72 deployments within the same modular infrastructure concept.

How is high-density cooling handled?

Liquid cooling is delivered to compute racks via CDUs, while CRAH units handle air cooling for support racks and residual heat loads in a combined thermal concept.

Softprom and STULZ

Softprom is the official distributor of STULZ. Our team supports data center operators and integrators in planning, specifying and deploying mission-critical cooling for high-density AI environments based on the Siemens-STULZ-NVIDIA reference architecture.

Prepared by the Softprom editorial team based on vendor materials (Softprom DistriFlow). Source: original article.