Vertiv has introduced SmartIT, a pre-engineered infrastructure solution designed to simplify the deployment of high-density AI systems in existing facilities and new data centers. The architecture integrates rack infrastructure, power conversion and distribution, direct-to-chip liquid cooling, cooling distribution and monitoring into a factory-validated design. Previously deployed with NVIDIA systems, SmartIT targets enterprises, research institutions and colocation providers that need to accommodate increasingly demanding AI workloads without coordinating every infrastructure component separately on-site.
Vertiv Packages Power and Liquid Cooling for AI Data Centers
The rapid growth of AI computing is putting pressure on data center infrastructure, particularly as high-density accelerator systems demand coordinated power delivery and more advanced thermal management. Vertiv’s new SmartIT solution aims to address part of that challenge by combining critical infrastructure components into a pre-engineered architecture that can be deployed as a repeatable building block.
The company says SmartIT is available globally and supports deployments ranging from a single rack to multi-rack environments. It has previously been deployed with NVIDIA infrastructure, including the NVIDIA DGX GB300, and is designed to support systems based on NVIDIA GB300 NVL72, Vera Rubin NVL72, DGX Rubin NVL8 and HGX Rubin NVL8 platforms. The architecture also supports NVIDIA MGX and Open Compute Project ORV3 designs.
The announcement reflects a broader infrastructure shift in AI: computing hardware is increasingly being designed and deployed alongside power and cooling systems, rather than treated as a separate layer to integrate after servers arrive.
Integrating power, cooling and monitoring
Vertiv SmartIT combines rack infrastructure, DC power conversion and distribution, direct-to-chip liquid cooling, in-rack cooling distribution and monitoring. Factory validation is intended to reduce the amount of integration required at the deployment site, leaving customers to connect the system to facility power and cooling infrastructure and complete commissioning.
This approach could help organisations that lack the time or space to design a bespoke infrastructure configuration for each AI installation. It also offers a standardised architecture that can be expanded as demand increases, rather than requiring a complete redesign each time a deployment grows.
Vertiv says SmartIT can support existing facilities, including brownfield enterprise sites and Class A commercial buildings, as well as newly built data centers. Whether a particular building can accommodate high-density AI equipment will still depend on available electrical capacity, heat rejection, floor loading, redundancy requirements and the facility’s existing cooling design. A pre-engineered rack solution cannot eliminate those site-level constraints.
The integrated architecture also aims to reduce coordination among multiple vendors. Instead of sourcing and validating rack, power, cooling and monitoring components independently, customers can procure a combined solution through one infrastructure partner. That may simplify responsibility during deployment, although enterprises will still need to evaluate compatibility with their own data center management, networking and operational requirements.
Liquid cooling becomes central to AI infrastructure
High-performance AI systems can generate concentrated heat loads that are difficult to manage with conventional air cooling alone. Direct-to-chip liquid cooling transfers heat from processors and accelerators to a liquid-based cooling system, reducing reliance on airflow as the sole means of removing heat from high-density racks.
SmartIT incorporates direct-to-chip cooling and in-rack cooling distribution as part of its standard architecture. The package also includes leak detection, monitoring and automated response capabilities, according to Vertiv. These functions are important because cooling incidents can threaten expensive computing equipment and interrupt workloads.
The company says the solution supports monitoring through Redfish, SNMP and Modbus, protocols used in infrastructure management and system integration. These interfaces can help operators connect equipment telemetry with existing monitoring tools, though the depth of integration and available controls will depend on the deployment.
Vertiv is also positioning SmartIT within a broader infrastructure portfolio. SmartRun extends integration across a data center row, rack power distribution units provide power distribution and monitoring, and Vertiv Unify coordinates monitoring, analytics and controls across power and thermal infrastructure. Together, these products are intended to support deployments that grow beyond individual racks.
Designed for evolving NVIDIA platforms
SmartIT’s compatibility with multiple NVIDIA architectures is central to its positioning. The listed platforms include the GB300 NVL72 generation and NVIDIA’s Vera Rubin and Rubin systems, as well as HGX and MGX-based configurations. This gives customers a common infrastructure approach across different AI system designs, subject to the power, thermal and configuration requirements of each platform.
Vertiv showcased SmartIT with an NVIDIA HGX Rubin NVL8 demonstration unit at NVIDIA GTC Berlin in October. The demonstration highlighted the company’s goal of enabling customers to start at a smaller scale and expand using a common infrastructure architecture.
However, support for a range of hardware platforms should not be interpreted as meaning every configuration can use identical power and cooling settings. Deployment teams must verify rack-level power budgets, coolant requirements, facility interfaces and operational limits for each supported system.
Deployment speed depends on the facility
Factory validation can shift integration work away from the data center floor and reduce the number of components that need to be assembled and tested on-site. A standardised design may also make multi-rack expansions more repeatable, particularly for enterprises and colocation operators building capacity in phases.
The actual time required to bring a system online will still depend on site readiness, utility capacity, cooling infrastructure, installation logistics and commissioning procedures. Vertiv’s announcement does not provide quantitative benchmarks for deployment-time reductions, energy efficiency or total cost of ownership.
For AI infrastructure buyers, those measurements are essential. Pre-engineered systems can reduce integration complexity, but the business case also depends on capital costs, serviceability, redundancy, power utilisation and how readily the solution fits the facility’s existing architecture.
SmartIT illustrates how infrastructure vendors are responding to AI’s growing physical requirements. As accelerator systems become more demanding, the ability to coordinate compute, power, liquid cooling and monitoring is becoming a key part of deployment planning. Vertiv’s approach is to package those elements into a repeatable system; customer results will determine how much that standardisation improves real-world deployment and operations.
Market Landscape
AI data center infrastructure is evolving as accelerator-heavy workloads increase the need for coordinated power delivery, liquid cooling and facility-level monitoring.
- High-density computing: AI accelerators create concentrated power and thermal demands that influence rack design and facility planning.
- Liquid cooling adoption: Direct-to-chip cooling is becoming increasingly important for managing heat from dense AI computing systems.
- Pre-engineered infrastructure: Integrated, factory-validated designs aim to reduce on-site integration work and make deployments more repeatable.
- Hardware compatibility: Support for multiple NVIDIA architectures gives infrastructure providers a way to address different AI deployment scales and configurations.
- Operational integration: Monitoring protocols and coordinated power and thermal controls help operators manage infrastructure across racks and facilities.
Top Insights
- Vertiv SmartIT combines rack infrastructure, DC power conversion, liquid cooling, cooling distribution and monitoring in a factory-validated architecture.
- The solution is designed for existing facilities and new data centers, with deployments scaling from one rack to multiple racks.
- SmartIT supports selected NVIDIA GB300 and Rubin architectures, alongside NVIDIA MGX and OCP ORV3 designs.
- Redfish, SNMP and Modbus support help connect infrastructure monitoring with broader data center management systems.
- Vertiv has not disclosed quantified deployment-time, energy-efficiency or total-cost-of-ownership improvements in the supplied announcement.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
