Panels on AI demand, cooling innovation, and gigawatt-scale design at Datacloud Global Congress 2026 reached a consistent and commercially significant finding: AI workloads are not displacing cloud demand but compounding it, and the architectural implications (for cooling, power density, rack design, and community engagement) are unlike anything the European market has previously absorbed.

The European data centre market has grown steadily for the better part of two decades on the back of cloud migration, digital transformation, and the proliferation of connected services. Annual cloud growth has run at 20% to 30%. That growth is now being overlaid by AI demand that panellists at DCGC 2026 described as categorically different in scale, density, and infrastructure requirements.

Over 80% of new data centre demand is now AI-related, driven by established hyperscalers, emerging AI companies, and Chinese cloud providers seeking European footholds. Critically, this demand does not replace cloud, it compounds it. An operator planning capacity on the basis of cloud growth projections alone is already planning short.

The scale shift: from megawatts to gigawatts

The most striking data point from the AI demand panel was the change in hyperscaler campus requirements. Where a large hyperscaler might previously have sought 50 to 100 megawatts of capacity, requirements now range from 300 to 700 megawatts, with some projects targeting 1 gigawatt.

This is a scale that Europe has not previously hosted. The panel noted a figure of 172 gigawatts as the AI-related energy optimisation potential identified for Europe, a number that underscores both the ambition and the infrastructural challenge of the continent’s AI trajectory.

Most large-scale AI training remains concentrated in the United States, but rapid scaling of inference workloads is expected across Europe within 18 to 36 months. Inference is less compute-intensive than training but far more distributed, latency-sensitive, and persistent, which creates a different set of facility requirements.

What gigawatt-scale AI means for facility design

The ‘Designing for Gigawatts’ session at DCGC 2026 provided the most technically specific account of what AI demand is doing to data centre architecture. The shift from air-cooled systems, which typically support up to 40 kilowatts per rack, to liquid-cooled systems capable of handling NVIDIA’s GB300 at 141 kilowatts per rack requires changes to cooling infrastructure, piping systems, and mechanical gallery design.

A single liquid cooling tech loop now costs between 600,000 and 1 million US dollars. Racks themselves are becoming heavier and more integrated, with some units weighing up to 22,000 pounds, creating floor loading, logistics, and insurance considerations that facility designers have not previously had to address. The panel also highlighted the need for flush and fill rooms to prevent contamination of liquid cooling loops when new servers are installed, an operational requirement that is reshaping facility layouts.

Supply chain, skills, and community: the three constraints

Beyond power and cooling, the AI demand panels at DCGC 2026 identified three intersecting constraints on European delivery capacity. Supply chain pressure on data centre equipment, particularly for cooling infrastructure and high-voltage electrical systems, means that lead times are a competitive differentiator.

The panel on AI demand noted that skilled labour is a bottleneck for delivery speed, with modular and prefabricated construction methods gaining traction specifically as a response to site-level workforce constraints. Community engagement was identified as the single most critical factor for project success in Europe, with the panel noting that early and sustained dialogue with local stakeholders is no longer an option but a prerequisite.

The trend toward private AI training tailored for specific enterprises was also flagged as requiring new facility designs distinct from hyperscaler campuses.

Key Takeaways

• AI demand is additive to existing cloud growth, not substitutional, operators planning on cloud projections alone are already underestimating future capacity requirements.
• Hyperscaler campus requirements in Europe have shifted from 50-100MW to 300MW-1GW, a scale the market has not previously accommodated, and which demands new approaches to land, power, and planning.
• Liquid cooling at rack densities of 141kW and beyond is no longer a niche requirement, it is the architectural direction for AI factory design, with significant implications for cooling infrastructure costs and facility layout.
• Skilled labour, not components, is now a delivery bottleneck in European data centre construction, driving adoption of modular and prefabricated build methods.
• European AI inference demand is expected to scale rapidly within 18 to 36 months, creating a narrow window for operators to build the facility types, supply chains, and community relationships required to meet it.

Europe’s data centre market is about to absorb a demand profile it has not been designed for. The operators who will benefit are those who start treating AI factory design as a distinct discipline from cloud data centre design now, not as a future consideration.

Cooling strategy, rack architecture, power density planning, and community engagement for gigawatt-scale projects are all different problems from their megawatt-scale equivalents. The 18-to-36-month window before European inference demand scales is not generous. It is, however, enough time for well-capitalised, well-organised operators to build a durable competitive position.