European data-centre operators signed 420MW of AI-focused colocation capacity during the first half of 2026, almost five times the 89MW agreed during the same period last year, according to new research from CBRE. The figures show demand for artificial-intelligence infrastructure spreading beyond the continent’s established cloud markets as newer compute providers commit to large blocks of capacity. Two-thirds of the contracted power is expected to be delivered to so-called neocloud operators in the Nordic region.

Neoclouds are emerging infrastructure providers built around high-performance computing and AI workloads, often supplying access to scarce accelerator capacity without carrying the same legacy cloud portfolios as the largest hyperscalers. Their expansion has created a new class of data-centre tenant whose appetite for power can be substantial but whose financial history may be shorter than the operators, investors, and lenders financing the buildings they occupy. CBRE says the latest deals suggest providers have become more comfortable with that customer base than they were two years ago.

That comfort remains conditional because data-centre operators are using measures including rental deposits and letters of credit to reduce counterparty risk on some transactions. The market is therefore being shaped by two forms of scarcity at once: AI developers need access to physical compute capacity, while infrastructure owners need enough confidence in those customers to finance power-heavy sites whose useful life extends well beyond a typical AI product cycle. Record leasing volumes show those two sides are finding contractual structures that allow more projects to move forward.

The geographic concentration is equally revealing, as 66% of the AI-focused capacity identified by CBRE is expected to be delivered to neoclouds in the Nordics. Lower-cost power and abundant renewable generation have long made markets such as Sweden, Finland, Norway, and Denmark attractive for energy-intensive computing, but AI is increasing the value of those characteristics as individual deployments become larger. Europe’s data-centre map is consequently being pulled towards places where electricity, grid access, land, and cooling can be assembled economically rather than simply towards the cities with the largest concentrations of corporate customers.

Power is becoming a location strategy

Traditional European data-centre hubs are not becoming irrelevant, since latency, network density, enterprise demand, and cloud connectivity still support established markets. AI training and other compute-intensive workloads can, however, tolerate different location trade-offs from applications that require equipment to sit close to financial centres or dense pools of end users. When power becomes a larger share of both operating cost and development risk, a site hundreds of kilometres from a conventional hub can become commercially attractive if it offers a credible route to tens or hundreds of megawatts.

Competition for electricity is already changing how infrastructure projects are evaluated across Europe. As grid delays create a market around alternative power arrangements and connection strategies, the ability to secure electricity can become more valuable than having a theoretically ideal plot of land. Nordic markets benefit from a comparatively strong combination of renewable generation, cooler climates, and available sites, although any rapid increase in data-centre demand still has to coexist with local industry, households, and wider electrification plans.

CBRE’s numbers also show why a signed megawatt should not be treated as identical to an operating megawatt. Leasing agreements create future demand for buildings, electrical systems, grid capacity, financing, and equipment, but the infrastructure still has to be delivered. Record contracting can strengthen investment pipelines while simultaneously increasing pressure on utilities, developers, and local authorities that have to turn commitments into functioning campuses.

AI infrastructure acquires credit risk

As the customer mix changes, financial underwriting is becoming a more visible part of the AI infrastructure market. Large hyperscalers have spent years building balance sheets, investment-grade credit profiles, and long-term relationships with data-centre landlords, whereas many neocloud businesses are younger and expanding at a pace that requires substantial external capital. A provider may have strong demand for GPU capacity and credible customers of its own while still presenting a different financing profile from Microsoft, Google, or Amazon.

Deposits and letters of credit can help bridge that gap by giving landlords and their lenders additional protection if a tenant cannot meet its obligations. Their growing use also shows why cheap access to AI models does not translate automatically into cheap infrastructure: underneath software pricing sit power contracts, buildings, transformers, cooling systems, GPUs, debt, leases, and guarantees. AI economics therefore depend partly on whether infrastructure providers can turn fast-growing compute demand into contractual cash flows that banks and investors are willing to fund.

For Europe, that can shift infrastructure investment towards regions capable of combining electricity availability with credible financing and planning. The Nordics currently occupy a favourable position in that equation, but the distribution of future capacity will depend on grids, permitting, hardware supply, and the financial durability of the companies signing these leases. CBRE’s first-half figures show that neoclouds have moved far enough from the margins of the market to influence where Europe builds its next wave of AI infrastructure, even though the longer test will be how much of the contracted capacity reaches operation on the terms now being agreed.