Vertiv announced alongside NVIDIA and AI specialist iGenius the upcoming deployment of Colosseum, an NVIDIA DGX SuperPOD powered by Grace Blackwell Superchips, in southern Italy. Slated for 2025, this sovereign AI data center integrates Vertiv’s 360AI modular power/cooling infrastructure, NVIDIA Omniverse digital twins, and software orchestration via Mission Control and Vertiv Unify to meet the stringent data security and residency needs of finance, healthcare, and public administration.
Strategic Imperatives for Sovereign AI
Europe’s regulated sectors demand AI solutions that ensure data sovereignty, compliance, and security. With the EU’s stringent data protection rules, enterprises face hurdles deploying cloud-based AI services hosted abroad. Colosseum addresses these challenges by localizing compute and embedding controls at every layer of the AI lifecycle, from model development to inference.
Colosseum Architecture & Capabilities
NVIDIA DGX SuperPOD & Grace Blackwell Superchips
-
Scale & Performance: Colosseum leverages dozens of NVIDIA GB200 NVL72 servers (each with 72 Grace Blackwell Superchips) to deliver hundreds of petaflops for real-time, trillion-parameter inference and training.
-
Modular Infrastructure: Built on Vertiv’s 360AI reference architecture, the data center supports 132 kW/rack liquid-cooled modules and scalable power/cooling systems, optimizing energy use and enabling rapid capacity expansion.
NVIDIA Omniverse Digital Twin
-
Design-to-Deploy Acceleration: Engineers use a physically accurate digital twin in Omniverse to simulate thermal loads, electrical flows, and site layout in real time—reducing simulation times from months to hours and cutting deployment schedules by up to 50%.
-
Collaborative Optimization: Real-time visualization and what-if analysis enable Vertiv, iGenius, and NVIDIA teams to refine infrastructure decisions before committing hardware, minimizing risk and rework.
Software Orchestration & Management
-
NVIDIA Mission Control: Provides centralized orchestration for GPU clusters, automating job scheduling and resource allocation to maximize throughput and availability.
-
Vertiv Unify: Delivers unified oversight of power, cooling, and compute, leveraging closed-loop optimization and predictive maintenance guided by digital twin data.
Business Impact & Use Cases
Regulated Workloads
Finance, healthcare, and public administration can now run sensitive AI workloads on-premises, ensuring data residency and compliance with GDPR and sector-specific regulations.
Cost Efficiency & Speed to Market
Modular design and pre-validated reference architectures reduce upfront engineering costs and accelerate time-to-value, enabling enterprises to launch AI services—like real-time risk analysis or personalized patient diagnostics—months faster than traditional builds.
Global Replicability
Colosseum serves as a blueprint for sovereign AI factories, with Vertiv and NVIDIA planning future expansions (e.g., DGX GB300 support). Organizations worldwide can replicate the model to achieve localized AI compute across multiple geographies.
Expert Perspectives
-
Uljan Sharka, CEO iGenius: “Modular systems and software-specific infrastructure enable a new era of mission-critical AI”.
-
Karsten Winther, President Vertiv EMEA: “The unit of compute is no longer the chip—it’s the system, the AI Factory”.
-
Charlie Boyle, VP DGX Platforms NVIDIA: “Colosseum sets a new standard for building supercomputers for the era of AI”.
Colosseum exemplifies the next generation of sovereign AI infrastructure—combining hyperscale compute, modular efficiency, and digital twin-driven agility—to meet the exacting requirements of regulated industries. For business and technology leaders, adopting this blueprint means achieving data sovereignty, rapid deployment, and scalable AI capabilities without compromising compliance or performance. As AI continues reshaping competitive landscapes, organizations should evaluate modular AI factories like Colosseum to future-proof operations, reduce time-to-market for AI-driven services, and maintain control over critical data assets.
