Data Centre LIVE: AI-Ready Infrastructure Panel

At Data Centre LIVE, leaders from ebm-papst, VIRTUS Data Centre, atNorth and Nscale debated the future of data centre design for AI with insights into power, cooling and speed. In a fast-paced panel session, leading voices in data centre strategy came together to explore the growing challenge of building AI-ready infrastructure. 

Panellists included:

• Stefan Bastian, Managing Director at ebm-papst
• David Watkins, Solutions Director at VIRTUS Data Centres
• Karl Havard, Chief Commercial Officer at Nscale
• Eyjólfur Magnús Kristinsson, CEO at atNorth

AI workloads are redefining infrastructure needs

Kicking off the conversation, the panellists agreed that AI isn’t just adding more compute – it’s changing the rules entirely. 

“We used to work with 5–15kW per rack,” Stefan said. “Now AI workloads are pushing 60–120kW, and Nvidia has announced hardware up to 600kW per rack.”

This surge in density demands a rethink across all infrastructure layers: cooling, power delivery, rack layout and even physical footprint. 

Karl referenced Nvidia’s Grace Blackwell release as a turning point, noting, “We’re now in a two-year refresh cycle, but contracts last seven to ten years. That disconnect creates real friction.”

Magnus highlighted the scale of deployments now expected: “Clusters are no longer 1 or 2MW—they’re 20 to 50MW in a single hub. That changes everything.”

David added that traditional large data centres of 30–40MW are now dwarfed by incoming requests. 

“We’re talking 300, 400, even 500MW campuses,” he said. “And the grid can’t keep up with delivery.”

Power, heat and flexibility: three critical design pressures

Much of the discussion focused on how to adapt existing facilities. Rack density is only part of the challenge—so too is power availability. 

“We have cities in Germany where it’s no longer possible to build due to grid constraints,” said Stefan. 

Carl noted that power price, not just access, is also driving decisions: “We’re seeing a clear migration to the Nordics.”

The consensus – future infrastructure must be modular, flexible and prepared for hybrid demands. 

“Cooling will be both liquid and air,” said Magnus. “In our white space, it’s already close to 50/50.” He stressed that residual heat from liquid-cooled systems must still be managed efficiently.

The group also stressed the importance of “dual-mode” designs. 

David explained: “AI and traditional workloads will co-exist for some time. You need a data centre that can pivot between them without sacrificing efficiency.”

Carl summarised the commercial challenge: “Public cloud providers have to serve both worlds. That’s harder than building for pure AI, like we do at Nscale.”

Edge, sovereignty and sustainability

The panellists turned to edge computing and the geopolitical implications of AI. Carl pointed out that edge isn’t just about latency. 

“It’s also about sovereignty. Countries and companies want AI workloads under their own legal control,” he said. 

Modular deployments and containerised solutions are now in demand from both enterprise and government.

Stefan agreed: “We’ve used edge computing in manufacturing for years – it’s now vital for privacy and security, especially when compressing and filtering before sending data to hyperscale centres.”

The sustainability angle was never far away. Magnus warned that AI data centres could consume up to 10% of global electricity. 

“We need better reuse of waste heat and long-term district heating strategies,” he said. “Otherwise, we’ll face public resistance.”

Carl added that some form of industry cap on power draw may be necessary. “If we keep doubling every 18 months, legacy sites will be obsolete within five years.”

Data storage, retrofits and the role of AI in design

AI workloads also shift the storage paradigm. 

“Previously, storage was 20% of the infrastructure,” said David. “Now we’re seeing much higher requirements and the need for proximity to compute makes design more complex and expensive.”

Magnus confirmed the growing interest in storage-centric planning: “Customers used to focus only on GPU density. Now it’s 50/50 between compute and storage.”

The group agreed that retrofitting legacy sites is possible – but not always viable. 

“We’ve retrofitted liquid cooling successfully,” said David. “But some 10-year-old sites just won’t be fit for the future.”

Stefan and Karl both emphasised the value of AI in infrastructure management itself. 

“We’re using AI to optimise cooling set points in real-time,” said Stefan. 

Karl added that AI must also be used to automate predictive maintenance and real-time control systems.

Cyber risk, governance and the future outlook

Security concerns were also addressed. 

Stefan and David noted the need for full separation between building control systems and customer-facing compute, keeping infrastructure secure. 

Karl warned that ‘Know Your Customer’ processes must improve: “You don’t want to sell infrastructure to a bad actor building unethical models.”

Magnus made the case for data centre operators to stay alert. 

“There’s a geopolitical race for AI, and infrastructure is now a target,” he said. “Security must be baked into every layer.”

In closing, the panellists looked ahead. 

Karl predicted shake-ups as older AI deployments become commercially unviable: “There’s a lot of sunk investment in tech that’s already out of date.” 

Stefan and Magnus reiterated that waste heat reuse and energy integration will be key differentiators. 

David added a final point: “We need to educate the public and government. AI data centres aren’t optional—they’re part of the digital future.”

Essential diary dates for 2025

Discover the essential diary dates for Data Centre Magazine, as well as its sister publications – Mobile Magazine and Technology Magazine.

To enter for the Global Tech & AI Awards, click here.

• Tech & AI LIVE London + Cyber LIVE London | 14-15 May 
• The Global Tech & AI Awards | 14 May
• Tech & AI LIVE Singapore | 4 November
• Tech & AI LIVE New York | 18 November

Explore the latest edition of Data Centre Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.

Discover all our upcoming events and secure your tickets today. 

Data Centre Magazine is a BizClik brand