A Blueprint for AI at Scale – A Reference Architecture for IEC Markets
The rapid growth of artificial intelligence is transforming the requirements placed on data center infrastructure. As compute density increases, operators face a fundamental challenge: how can new AI capacity be deployed quickly while ensuring that power, cooling and operational systems work together efficiently and reliably?
Addressing this challenge requires more than optimizing individual technologies. It calls for an integrated approach in which compute, power, cooling and automation are considered as part of one coordinated infrastructure concept. This is where the collaboration between Siemens, STULZ and NVIDIA comes into play. Together, the three companies have developed a reference blueprint for a 100 MW hyperscale AI data center, combining their respective expertise in electrical infrastructure and automation, mission-critical cooling, and AI computing.
An integrated approach to high-density AI
Designed around NVIDIA GB300 NVL72 systems while also accommodating GB200 NVL72 deployments, the blueprint addresses the key challenges of high-density AI environments: increasing rack power densities, rising cooling demand, dynamic electrical loads and the need for scalable, highly available infrastructure. Its modular architecture integrates power, cooling, automation and compute to support efficient and repeatable deployment.
STULZ provides the cooling technologies at the heart of the thermal concept: STULZ CyberCool SCS 15102 WU Cooling Distribution Units deliver liquid cooling to the compute racks, while STULZ ABR 2500 CW Computer Room Air Handlers provide air cooling for support racks and residual heat loads. Combined with Siemens power and automation systems and NVIDIA AI computing technology, these solutions form an integrated infrastructure concept designed to meet the demanding requirements of AI workloads.
From blueprint to deployment
The result is a scalable reference architecture that provides a starting point for planning and designing high-density AI data centers in IEC markets. Its modular approach is designed to help accelerate time-to-compute, support efficient operation and provide the flexibility to adapt the infrastructure to specific project requirements.
Discover the Siemens-owned 100 MW Hyperscale AI Data Center Blueprint, developed in collaboration with STULZ and NVIDIA, and learn how the combined expertise of the three companies can help you plan, scale and optimize high-performance, energy-aware AI data center infrastructure.
