Nvidia, Vertiv, iGenius, partner on AI supercomputer

Nvidia, Vertiv, and iGenius are teaming up to launch Colosseum, one of the world’s largest AI data centre supercomputers. The facility, powered by Nvidia’s DGX SuperPOD architecture and the latest Grace Blackwell Superchips, aims to set a new global benchmark for secure, sovereign AI infrastructure, particularly for industries that deal with sensitive data – including finance, healthcare, and government.

Sovereign AI refers to AI systems that operate entirely within a country’s borders, ensuring data is processed, stored, and managed locally. This approach supports compliance with national regulations, enhances data privacy, and gives governments and organisations greater control over sensitive information.

Colosseum, a NVIDIA DGX SuperPOD, is the latest advancement in a long-standing collaboration between Vertiv and Nvidia. It is strategically positioned in southern Italy to address regional government requirements, marking a significant milestone in Europe’s AI landscape. Colosseum is set to deploy this year (2025).

“Harnessing the power of NVIDIA’s cutting-edge accelerated computing and Vertiv’s innovative infrastructure expertise, Colosseum stands as a testament to the transformative potential of sovereign AI,” says Uljan Sharka, CEO of iGenius. “We’re demonstrating how modular systems and software-specific infrastructure enable a new era of mission-critical AI.”

Modular by design, engineered for efficiency

Colosseum combines Vertiv’s infrastructure management expertise, Nvidia accelerated computing, and the Nvidia Omniverse Blueprint for AI factory design and operations. The deployment will leverage Vertiv’s 360AI reference architecture infrastructure platform for data centre power and cooling that is designed for the Nvidia GB200 NVL72 , which was co-developed with Nvidia and released in late 2024.

This modular and scalable system positions iGenius to deploy one of the fastest hyperscale AI supercomputers, and one of the largest to support sovereign AI. 

Vertiv has extended its reference design library on its AI Hub with the co-developed data centre power and cooling design for Nvidia GB300 NVL72. By staying one GPU generation ahead, Vertiv enables customers to plan infrastructure before silicon lands, with deployment-ready designs that anticipate increased rack power densities and repeatable templates for AI factories at scale. 

Karsten Winther, president of Vertiv, EMEA, says: “The unit of compute is no longer the chip – it’s the system, the AI Factory. Through our collaboration with Nvidia and visionary AI player iGenius, we are proving the efficiency and system-level maturity of delivering the data centre as a unit of compute, unlocking rapid adoption of AI-native power and cooling infrastructure as a catalyst for AI at scale.”

Simulate with Nvidia Omniverse, deliver with speed

Charlie Boyle, VP of DGX platforms at Nvidia, says: “AI is reshaping the data centre landscape, demanding new levels of scale, efficiency and adaptability for global AI factories. With physically-based digital twins enabled by Nvidia Omniverse technologies and Vertiv’s modular design for the iGenius DGX SuperPOD data centre, Colosseum sets a new standard for building supercomputers for the era of AI.”

Colosseum was co-designed as a physically accurate digital twin developed with Nvidia Omniverse technologies, enabling real-time collaboration between Vertiv, iGenius and Nvidia, to accelerate system-level decisions and compress the design-to-deploy cycle.

The Omniverse Blueprint enables real-time simulations, allowing engineers to test and refine designs instantly, rather than waiting for lengthy simulation processes, reducing simulation times from months to hours. Vertiv manufacturing and factory integration processes reduce deployment time by up to 50% compared to traditional data centre builds.

This collaborative 3D design process validated the entire infrastructure stack, enabling predictive modelling of thermal load, electrical flow, and site layout – for 132kW liquid-cooled racks to modular power systems – before a single module was built.

Designed with intelligence, unified by software

Vertiv’s AI-ready prefabricated modular data centre solution is designed, manufactured, delivered, installed and commissioned by Vertiv.It includes power, cooling, management, monitoring, service and maintenance offerings, with power and cooling capacity supporting up to 132kW/rack initially, with an ability to scale up as required for future designs.

The building shell integrates prefabricated white space inside while deploying full modular grey space outside. Vertiv says this approach offers exceptional scalability and energy efficiency, transforming the way data centres are built and deployed. 

Colosseum will leverage Nvidia Mission Control for data centre operations and orchestration and Vertiv Unify to simplify and synchronise building management for AI factories.

Vertiv Unify provides:

• Real-time orchestration across power, cooling, and compute.
• Digital twin synchronisation for closed-loop optimisation.
• AI-ready capabilities that support autonomous decision-making.

Through its integration of Nvidia Omniverse technologies, Vertiv Unify enables real-time updates between physical systems and digital models – allowing predictive maintenance, what-if simulations, and scenario testing before operational risk occurs.

The blueprint for AI factories globally

Colosseum is intended to serve as a model for scalable and repeatable sovereign AI infrastructure. By integrating high-density computing, localised data control, and modular deployment, it represents a shift toward AI systems that are secure, efficient, compliant with regulations, and geographically distributed.

Rather than being a standalone project, Colosseum is designed as a reference architecture. iGenius plans to replicate the approach globally, with Vertiv and Nvidia supporting future platform generations, including the upcoming DGX GB300 systems. The project reflects the ongoing move from conceptual discussions about sovereign AI to real-world implementation.