Vertiv: Powering Infrastructure for the AI-Driven Edge

The acceleration of AI is forcing a fundamental redesign of data centre infrastructure.

Andrea Ferro, VP Power and IT Systems EMEA at Vertiv, explains that the industry is grappling with huge changes in resource use and power demands, caused by the speed of AI adoption.

Vertiv is responding to forecasts showing AI-ready data centre capacity growing 33% annually to 2030, according to McKinsey.

In parallel, Goldman Sachs projects AI could lead to a 165% increase in data centre power consumption by the same year.

Rethinking power for next-generation AI racks

The power consumption of AI systems requires new approaches to infrastructure design, as traditional methods are insufficient for next-gen AI power densities.

“Today’s high-end racks already consume up to 120-132kW or more of power," Andrea explains. "But next-generation systems launching in 2027 and later are estimated to be up to 600kW per rack – and potentially beyond 1MW."

He stresses this requires rethinking power delivery and thermal management, not just scaling existing infrastructure. 

Edge AI growth and the move to inference

A critical transformation is the migration of workloads from centralised training to distributed inference at the edge.

Andrea sees 2025 as a key year for this change, moving away from the large-scale training clusters that have dominated planning.

“Training workloads can tolerate higher latency," Andrea says. "Inference, particularly for agentic AI applications, demands microsecond response times with consistent performance."

This change supports real-time applications like robotics and autonomous systems.

Reflecting this, the edge AI market is projected to grow from US$20.78bn in 2024 to US$66.47bn by 2030.

“Enterprises need local processing for speed, control and efficiency," Andrea insists. He notes edge data centres also support regional resilience, data sovereignty and security.

This isn't about abandoning the cloud but creating a hybrid model.

“AI needs a hybrid approach that leverages both centralised and distributed capacity,” he adds, believing that the key is matching the workload to the right infrastructure, with real-time inference requiring edge deployment.

Integrated solutions for AI infrastructure challenges

Power density is the primary infrastructure challenge.

The jump from 20kW racks to systems using more than 120kW creates numerous technical issues.

“Power delivery requires 33kW DC power shelves and 1400A busbars – infrastructure not imagined for pre-AI data centres,” Andrea says.

Thermal management also demands advanced liquid cooling systems, like Vertiv’s CoolChip CDU 100, to handle heat loads that overwhelm traditional air cooling.

Vertiv addresses these needs with integrated solutions where power, cooling and racking work together seamlessly. For example, poor integration can lead to thermal hotspots or inefficient heat transfer.

To simplify this, Vertiv has developed reference architectures.

“Our Vertiv 360AI reference architectures can be deployed as integrated systems that work within existing facility constraints," Andrea says.

This approach can reduce deployment time by up to 50%.

He emphasises that adaptability is a core design principle: "The main principle is designing for adaptability and scalability, not just current requirements."

"The decisions we make today about edge infrastructure will enable or constrain the next decade of AI innovation."