March 26, 2026

A 3.2GW data centre campus in Poland isn’t a “project” so much as a power-and-permits statement. WBS Power’s planned 3.2GW Baltic Data Centre Campus in Lublewo, Pomerania is slated to arrive in four 800MW phases, with grid connection conditions already secured and preparatory work running through end‑2027. The subtext is hard to miss: Europe’s AI buildout is now being planned at the scale of generation and transmission—because that’s where the bottlenecks are.

The Big Stories

Europe’s newest mega-campus plan comes from WBS Power: 3.2GW in northern Poland, designed for AI/HPC and cloud, with renewables and battery storage in the early mix and nuclear power pencilled in for later. The phased approach (800MW at a time) is a practical admission that land, grid, and equipment timelines—not customer demand—set the pace. It also puts pressure on neighbouring markets: if Poland can package grid access and expansion runway, it starts to compete with the Nordics and the usual FLAP-D gravitational pull.

That competition is already visible in the Nordics, where AI demand is driving what CBRE calls Europe’s fastest growth in data centre capacity. WIRED’s look at the Nordic boom highlights the usual advantages—hydropower, wind, lower electricity costs, and cool climates—plus an atNorth facility planned in Sweden. The interesting shift is that “cheap and green” is no longer just a marketing line; it’s becoming the default site-selection filter as rack densities climb and power procurement becomes the limiting reagent.

In the UK, the sovereign-control pendulum is swinging in a way that should worry US hyperscalers—even if capex headlines keep coming. Research cited in an Asanti report says 52% of UK IT decision-makers plan to reduce reliance on US-based cloud providers, with 45% looking to limit exposure to US jurisdiction, following the 2024 designation of data centres as Critical National Infrastructure. The near-term outcome likely isn’t “de-clouding” so much as more hybrid: cloud plus colocation plus on‑prem, with procurement and architecture decisions increasingly framed by legal control, not just latency and cost.

India is pushing on a different lever: making “green” power for data centres dispatchable, not intermittent. NTPC Green Energy Limited (NGEL) and Nxtra’s MoU targets renewable energy–based round-the-clock (RE‑RTC) supply across Nxtra’s Indian footprint, using combinations of solar, wind, and storage/balancing mechanisms. Nxtra’s stated ambition—to expand to over 400MW in three years and reach net-zero by 2031—reads like a direct challenge to the market to deliver firm power products at scale, not just headline PPAs.

On the connectivity side, Brussels is trying to remove some friction from the infrastructure that actually makes distributed compute work. The European Commission’s Digital Networks Act aims to harmonise telecom regulation to accelerate 5G/6G, fibre, and “resilient digital infrastructure,” with Analysys Mason modelling AI-driven interconnect traffic increases of up to 50 percentage points in high-growth scenarios. For data centres, this is the underappreciated constraint: you can build the megawatts, but if fibre rollout and cross-border rules stay patchy, you end up with stranded capacity or expensive workarounds.

Behind the Headlines

The money quietly moving into grids is the tell for where the real constraint sits. The EIB’s €200 million loan to A2A to modernise and reinforce Milan’s electricity distribution network (including renewal of roughly 450km of medium-voltage and 140km of low-voltage lines, plus nearly 800 substations) is framed as decarbonisation and resilience—with an estimated ~6.9 kilotonnes CO2 reduction annually. For the data centre industry, it’s also a reminder that urban and peri-urban capacity upgrades are increasingly financial-engineering problems: long-lived regulated assets, staged upgrades, and public lenders stepping in where timelines matter as much as returns.

Liquid cooling is moving from “specialty” to baseline design, and the market numbers are starting to reflect that. A Network World deep dive on liquid cooling notes operators including Google, Microsoft, and Meta deploying or committing to liquid-cooled designs, and points to Nvidia’s Vera Rubin platform supporting 45°C warm-water cooling with availability expected in H2 2026. The key implication isn’t just higher rack density—it’s operational normalisation: new failure modes, new monitoring requirements, and new vendor dependencies that look more like industrial cooling systems than classic white-space HVAC.

Finally, the network is trying to catch up to the workload. Tata Communications’ IZO Data Centre Dynamic Connectivity pitches software-defined, self-healing connectivity with deterministic multi-path routing, AI-based predictive insights, and a >99.99% availability target across five continents, alongside consumption-based pricing. Whether or not the “up to 30% op cost savings” claim holds in the messy real world, the direction is right: as AI workloads sprawl across regions (because power and GPUs don’t show up where you want them), buyers are going to pay for predictable performance and automated failover rather than overbuilding idle backup paths.