Vertiv & NVIDIA advance 800 VDC power for future AI factories

Vertiv has announced advances in its collaboration with NVIDIA to develop 800 VDC power architectures intended for the next generation of AI factories.

Infrastructure demands

As data centres move toward supporting large-scale artificial intelligence and high-performance computing operations, the industry faces changing requirements for power distribution at the rack level. Traditional 54 VDC in-rack distribution, which has supported kilowatt-scale racks, is not designed to meet the demands now being driven by AI workloads that can require megawatt-scale capacity.

To address this, Vertiv and NVIDIA are working together on 800 VDC power systems designed to provide scalable options and integrated energy storage, targeting the infrastructure needs of 'AI factories' handling highly synchronous and resource-intensive tasks. Vertiv's new systems aim to support the requirements of NVIDIA's future compute platforms, particularly the upcoming Rubin Ultra generation, by facilitating higher efficiency and energy capacity directly to the server racks.

Vertiv's 800 VDC platform is expected to be available in the second half of 2026 and is being timed to support the deployment plans that will begin in 2027 for NVIDIA's next-stage platforms. According to Vertiv, this transition from traditional rack-scale deployments to megawatt-scale system designs is a key step in enabling these new AI workloads.

Design and engineering progress

The company has moved from concept into what it describes as engineering readiness for these new architectures. Vertiv is now finalising component specifications for its platform, which will include centralised rectifiers, high-efficiency DC busways and rack-level DC-DC converters. These components will be designed to handle the larger power demands anticipated for AI-centric data centres.

"Larger AI workloads are reshaping every aspect of data centre design," said Scott Armul, Executive Vice President, Global Portfolio and Business Units at Vertiv. "Our systems-level expertise in both AC and DC-based power data centre architectures positions us uniquely to address the unprecedented power demands of AI workloads. With the development of our Vertiv 800 VDC platform designs, we're translating our extensive experience into next-generation solutions that will support the massive compute densities required for AI factories."

Armul's comments highlight the shift in data centre power architectures being shaped by the increasing computational demands from larger AI models and training workloads.

Strategic collaboration

"Powering the next generation of megawatt-scale AI factories requires a fundamental shift in power architectures," said Dion Harris, Senior Director, HPC, Cloud and AI Infrastructure at NVIDIA. "NVIDIA and Vertiv are working closely together to develop the scalable and efficient power foundation needed to unlock the full potential of next-generation AI infrastructure."

The partnership between Vertiv and NVIDIA is structured to address real-world requirements for growing data centre capacity. Vertiv reported its involvement in the early design phases of several large AI factory projects, using its 800 VDC architecture as the basis of design and scaling to meet gigawatt-scale demand profiles.

Service and support

Aside from platform development, Vertiv emphasised the importance of service readiness for these higher-voltage and more complex environments. Vertiv's service model, supported by more than 4,000 field service engineers worldwide, aims to provide operational confidence and reliability for mission-critical AI operations. The company underlined that safely servicing these complex data environments is essential for broad AI factory adoption.

Armul added, "We are engineering a holistic, scalable system where infrastructure parts interoperate as one - demonstrating Vertiv's role as a systems-level partner, moving from vision to readiness and enabling the infrastructure necessary to power future-ready AI factories."

Vertiv's collaboration with NVIDIA and shift toward 800 VDC architectures reflects ongoing changes across the data centre sector, driven by expanding AI workloads and the consequential demand for higher-capacity, more efficient power systems.