Schneider Electric: Rising AI Demands Data Centre Retrofits

The AI revolution has created an infrastructure crisis for data centre operators.

As AI transitions from experimental technology to business-critical infrastructure, the industry faces a critical question: how can facilities meet surging demand for AI-ready infrastructure without the decade-long timelines and enormous capital requirements of greenfield construction?

For Steve Carlini, Chief Advocate for Data Centre and AI at Schneider Electric, the solution lies in the technological transformation of existing facilities rather than new construction.

With the right technical approach, traditional data centres can be transformed into AI-ready facilities capable of handling the intensive workloads that modern machine learning applications demand.

Technical requirements for power infrastructure

The technology infrastructure gap stems from fundamental differences between traditional computing and AI workloads.

While legacy data centres were designed for distributed workloads with modest power densities, AI applications concentrate unprecedented computing power in smaller footprints, creating thermal and electrical challenges that existing infrastructure cannot support.

According to Steve, brownfield retrofits offer compelling advantages beyond simple economics.

"Sites that exist will already be permitted as a data centre and can circumvent any lengthy permitting cycle," he explains.

"Additionally, many existing data centres may be strategically located close to the data sources or applications, which can have significant advantages in faster speed, lower latency and lower data transfer costs."

Not every existing facility makes a suitable candidate for AI transformation.

Steve outlines a clear hierarchy of technical priorities when evaluating potential retrofit projects, with power availability topping the list.

"Sites that have an abundance of utility power are pure gold as accelerated compute AI requires more power," he notes. "Second are sites where more utility capacity can be added quickly."

Beyond connectivity, location considerations include isolation from residential areas.

"Additional considerations, picking sites that are isolated from neighbourhoods and highly populated areas," Steve advises.

"These areas can be difficult for the new breed of data centres with more generators and chillers that can make quite a bit of noise and can trigger complaints."

The electrical distribution system represents the first major technical hurdle in retrofit projects.

"On the power side, the main issue is grid power," Steve states.

"The second issue will be the entire power distribution inside the data centre. Traditional data centres were designed for lower power densities or distributed workloads."

AI workloads demand concentrated power delivery, which may require upgrades to power distribution units (PDUs), medium-voltage switchgear, low-voltage switchgear, transformers, circuit breakers and busways or cabling.

Steve points to developments in direct current systems that could improve efficiency whilst reducing thermal loads.

"A new generation of 800VDC electrical power distribution to the AI servers will use less current and produce less heat in the future," he explains.

Implementing liquid cooling systems

The cooling challenge proves equally complex from a technical perspective.

"On the cooling side, most next-generation AI servers are natively liquid-cooled and come with integrated cold water inlet and hot water outlet connections," Steve explains.

"These are not optional – they are required for operation."

Required components include CDUs (Coolant Distribution Units), chillers or dry coolers compatible with liquid cooling, piping infrastructure for cold and hot water, heat exchangers and monitoring systems for temperature, flow rate and leak detection.

The implementation of liquid cooling systems requires careful planning and phased deployment.

Operators must assess their existing infrastructure to determine compatibility with liquid cooling technologies, including evaluating floor space, structural load-bearing capacity and access to water sources or dry cooling capabilities.

Integration with existing systems presents unique challenges.

Retrofit projects must often maintain operational continuity while installing new cooling infrastructure, requiring staged deployments that minimise disruption to existing workloads.

This phased approach allows operators to test and validate liquid cooling systems before full-scale implementation.

Crucially, liquid cooling does not eliminate traditional cooling entirely.

"Approximately 20% to 30% of the thermal load may still require air-based cooling for components like power supplies and memory," Steve notes.

This hybrid approach necessitates careful coordination between liquid and air-based cooling systems to ensure optimal performance across all infrastructure components.

Monetising AI inference workloads

The business case for retrofitting centres on AI's revenue potential as applications transition from experimental pilot projects to production systems.

"Many data centre operators would like to add accelerated compute AI and become and run applications for in-house application automation or offer pay for AI models," Steve explains. "The monetisation of AI working models or inference is the next big wave."

The convergence of several technological trends makes retrofitting an increasingly strategic priority.

"As production-ready AI inference applications start gaining momentum, companies improve their business process efficiency and they start to automate their business processes with agentic AI and eventually start Artificial General Intelligence (AGI), each progression will require significantly more computing horsepower enabled by data centres," Steve explains.

Market dynamics favour operators who can rapidly deploy AI-ready infrastructure.

The growing demand for inference workloads, particularly for distributed AI applications closer to data sources, creates opportunities for retrofitted facilities to capture market share without the lengthy deployment cycles associated with new construction.

Steve highlights sustainability improvements through advanced technologies.

"Closed loop liquid-cooling systems use much less water than traditional cooling," he points out.

For smaller and medium-sized operators, this creates an opportunity to compete in the distributed inference market while meeting increasingly stringent environmental requirements.