Vertiv Releases 7MW End-to-End Critical Power and Cooling Reference Architecture for NVIDIA Blackwell AI Data Centers

On October 15, 2024, Vertiv announced the release of a complete reference architecture for the NVIDIA GB200 NVL72 platform, co-developed with NVIDIA. The design targets deployments of up to 7MW and supports per-rack power densities of up to 132kW. Covering critical power and cooling infrastructure from end to end, the architecture helps customers retrofit existing facilities for large-scale AI workloads and provides a standardized deployment blueprint for greenfield data center builds. By aligning AI cluster capacity with data center power modules, the design reduces stranded power and lowers capital overhead. It forms part of the Vertiv 360AI Reference Design portfolio, which provides integrated infrastructure pathways for high-performance and AI-accelerated computing applications.

Power and Cooling Infrastructure

The power layer is built around the Vertiv Trinergy Uninterruptible Power Supply system, paired with Vertiv EnergyCore lithium-ion battery cabinets. Compared to conventional lead-acid UPS configurations, this combination reduces physical footprint by approximately 40% while maintaining high-availability power protection. Integrated load-leveling via lithium-ion batteries is engineered specifically for the dynamic, burst-intensive power draw of GPU-accelerated workloads. During large-scale AI training or inference tasks, instantaneous power fluctuations across GPU clusters are significantly higher than those of conventional IT loads — a demand profile that static UPS architectures cannot manage efficiently. The integrated design actively smooths these fluctuations, stabilizing power delivery and extending battery service life. The architecture also includes engineering guidance for OCP-inspired DC power shelf integration, offering an optional path for operators requiring open-standard infrastructure alignment.

The cooling layer uses a hybrid liquid and air cooling topology, integrating Vertiv AFC chillers, Vertiv Liebert CW chilled-water room cooling units, and Vertiv XDU coolant distribution units. Liquid cooling handles the primary thermal load of high-density racks, while the air-side systems use low global warming potential (GWP) refrigerants to manage residual heat. Operating in coordination, the two cooling modes reduce annual cooling operating costs by up to 20% compared to fixed-speed compressor solutions. At a per-rack density of 132kW, achieving a balance between cost efficiency and operational reliability through a single cooling method is not practical under current engineering constraints; the hybrid topology represents a validated approach to this challenge.

Modular Deployment and Service Support

The Vertiv MegaMod CoolChip prefabricated module system enables factory-integrated, pre-configured delivery, achieving deployment speeds up to 50% faster than conventional on-site construction. Prefabricated delivery moves the majority of commissioning work into a controlled factory environment, reducing on-site variables and compressing the timeline from contract to operation. The modular approach applies to both legacy facility retrofits and new data center builds, and supports standardized infrastructure replication across multiple sites. Vertiv maintains approximately 4,000 field service engineers across more than 130 countries, providing installation, commissioning, and lifecycle operational support for complex system-level deployments.

Industry Background

The NVIDIA GB200 NVL72 platform, based on the Blackwell architecture, reaches per-rack power densities of up to 132kW — beyond the design parameters of conventional data center power and cooling infrastructure. Enterprise AI infrastructure is shifting from general-purpose server environments toward purpose-built AI facilities centered on accelerated compute density, driving sustained demand for high-density, liquid-cooling-ready critical infrastructure. Next-generation AI data centers differ fundamentally from conventional facilities in architectural complexity, power density, and dynamic workload characteristics. Decision-making models that rely on accumulated case-by-case engineering experience are no longer sufficient to support the pace and scale of current AI deployment requirements. A jointly validated reference architecture standardizes engineering decisions, shortens design review cycles, and systematically reduces implementation risk across both retrofit and greenfield projects.

About Songli and Inquiry

Songli is an authorized Vertiv distributor with a long-standing focus on critical power and cooling infrastructure for data centers. Our product coverage includes UPS systems, precision cooling, telecommunications power, and associated infrastructure components. We serve data center contractors, system integrators, and enterprise procurement teams, providing technical support across product selection, solution assessment, and on-site delivery.

If you are planning a new AI data center build, evaluating a retrofit for an existing facility, or require technical consultation and pricing for Vertiv product lines, we welcome your inquiry. Our technical team will provide product recommendations and solution support based on your specific project requirements.