Vertiv has introduced a groundbreaking collaboration with NVIDIA and famend AI pioneer iGenius to deploy Colosseum, one of many world’s largest NVIDIA DGX AI supercomputers with NVIDIA Grace Blackwell Superchips. Set to deploy in 2025 in Italy, Colosseum will redefine the digital panorama via a first-of-its-kind sovereign AI knowledge centre for regulated workloads.
Designed to handle the calls for of extremely regulated industries similar to finance, healthcare and public administration, Colosseum will embody a fusion of transformative computational energy, power effectivity and knowledge sovereignty, whereas balancing stringent knowledge safety necessities.
Colosseum, a NVIDIA DGX SuperPOD, is the most recent development in a long-standing collaboration between Vertiv and NVIDIA. It’s strategically positioned in southern Italy to handle regional authorities necessities, marking a major milestone in Europe’s AI panorama.
“Harnessing the ability of NVIDIA’s cutting-edge accelerated computing and Vertiv’s modern infrastructure experience, Colosseum stands as a testomony to the transformative potential of sovereign AI,” mentioned Uljan Sharka, CEO of iGenius. “We’re demonstrating how modular programs and software-specific infrastructure allow a brand new period of mission-critical AI.”
Modular by Design. Engineered for Effectivity.
Colosseum combines Vertiv’s infrastructure administration experience, NVIDIA accelerated computing, and the NVIDIA Omniverse Blueprint for AI manufacturing facility design and operations. The deployment will leverage Vertiv’s 360AI reference structure infrastructure platform for knowledge centre energy and cooling that’s designed for the NVIDIA GB200 NVL72 , which was co-developed with NVIDIA and launched in late 2024.
This modular and scalable system positions iGenius to deploy one of many quickest hyperscale AI supercomputers, and one of many largest to help sovereign AI.
Vertiv has additionally prolonged its reference design library on its AI Hub with the co-developed knowledge centre energy and cooling design for NVIDIA GB300 NVL72. By staying one GPU era forward, Vertiv permits clients to plan infrastructure earlier than silicon lands, with deployment-ready designs that anticipate elevated rack energy densities and repeatable templates for AI factories at scale.
“The unit of compute is not the chip – it’s the system, the AI Manufacturing unit,” mentioned Karsten Winther, president of Vertiv, EMEA. “By way of our collaboration with NVIDIA and visionary AI participant iGenius, we’re proving the effectivity and system-level maturity of delivering the information centre as a unit of compute, unlocking fast adoption of AI-native energy and cooling infrastructure as a catalyst for AI at scale.”
Simulate with NVIDIA Omniverse. Ship with Velocity.
“AI is reshaping the information centre panorama, demanding new ranges of scale, effectivity and flexibility for world AI factories,” mentioned Charlie Boyle, vp of DGX platforms at NVIDIA. “With physically-based digital twins enabled by NVIDIA Omniverse applied sciences and Vertiv’s modular design for the iGenius DGX SuperPOD knowledge centre, Colosseum units a brand new normal for constructing supercomputers for the period of AI.”
Colosseum was co-designed as a bodily correct digital twin developed with NVIDIA Omniverse applied sciences, enabling real-time collaboration between Vertiv, iGenius and NVIDIA, to speed up system-level selections and compress the design-to-deploy cycle. The Omniverse Blueprint permits real-time simulations, permitting engineers to check and refine designs immediately, quite than ready for prolonged simulation processes, decreasing simulation instances from months to hours. Vertiv manufacturing and manufacturing facility integration processes scale back deployment time by as much as 50% in comparison with conventional knowledge centre builds.
This collaborative 3D design course of validated all the infrastructure stack, enabling predictive modelling of thermal load, electrical stream, and web site structure – for 132kW liquid-cooled racks to modular energy programs – earlier than a single module was constructed.
Designed with Intelligence. Unified by Software program.
Vertiv’s AI-ready prefabricated modular knowledge centre resolution is designed, manufactured, delivered, put in and commissioned by Vertiv. It consists of energy, cooling, administration, monitoring, service and upkeep choices, with energy and cooling capability supporting as much as 132kW/rack initially, with a capability to scale up as required for future designs. The constructing shell integrates prefabricated white house inside whereas deploying full modular gray house outdoors. This strategy gives distinctive scalability and power effectivity, reworking the way in which knowledge centres are constructed and deployed.
Colosseum will leverage NVIDIA Mission Management for knowledge centre operations and orchestration and Vertiv™ Unify to simplify and synchronise constructing administration for AI factories. Vertiv Unify supplies:
· Actual-time orchestration throughout energy, cooling, and compute
· Digital twin synchronisation for closed-loop optimisation
· AI-ready capabilities that help autonomous decision-making
By way of its integration of NVIDIA Omniverse applied sciences, Vertiv Unify permits real-time updates between bodily programs and digital fashions – permitting predictive upkeep, what-if simulations, and situation testing earlier than operational threat happens.
The Blueprint for AI Factories Globally
Colosseum is greater than a knowledge centre. It’s the template for scalable, repeatable, sovereign AI factories. By combining cloud-scale density, native knowledge management, and modular deployment, it alerts the subsequent part of AI: the place inference have to be safe, quick, compliant, and distributed.
This isn’t a one-off challenge – it’s a reference level. iGenius is constructing a blueprint with Colosseum designed to be repeated globally, with Vertiv and NVIDIA aligned on future platform help, together with DGX GB300 programs and past. The way forward for sovereign AI is not theoretical – it’s being constructed now.
