The European data centre conversation has been dominated for two years by one number: capacity. How many gigawatts can the continent build, how fast, and where. The question is real. The answer most of the market has reached for is wrong.
The default assumption has been that AI infrastructure scales by greenfield. Find a large piece of land, secure planning, queue for grid, build the facility, energise. The model worked when data centres were 20MW shells in a London suburb. It does not work for gigawatt-scale AI compute campuses competing against industrial heat, electric vehicle charging, residential electrification and offshore wind curtailment for the same constrained network.
The grid is the constraint. The brownfield is the answer.
The thesis
Across Europe sits a category of asset that has been quietly hiding in plain sight. Large industrial sites with existing or proximate grid connections, often in the 100MW to 500MW range. Decommissioned smelters. Closed refineries. Underutilised port facilities. Former petrochemical complexes. Sites that were originally built around heavy industrial load, where the grid infrastructure was sized for a use case that no longer exists.
These are not greenfield. They are not retail real estate. They are powered industrial assets with most of the hardest planning, environmental and grid groundwork already in place. Converting them into AI-ready data centre developments compresses the delivery timeline by two to four years compared to building from scratch.
The economics work because the constraint has moved. In 2020 the binding constraint was capital. In 2024 it was chips. In 2026 it is energised, qualified MW. The asset class that solves for that constraint, at scale, is the brownfield industrial conversion.
A handful of European platforms have started screening for these sites. The early portfolios suggest the addressable opportunity sits in the multiple gigawatts, across countries with deeper renewable generation than the south of England and faster connection processes than the UK currently offers. Northern Europe in particular, where stranded wind generation, available cooling water, and industrial decline have left a deep inventory of sites that match the profile.
Why most of the industry will miss it
The brownfield thesis sounds straightforward. In execution it is harder than greenfield, not easier. Three reasons.
First, origination. These sites do not appear on the open market. They sit on the balance sheets of legacy industrial operators who have not yet decided to sell, do not understand what the asset is now worth in a hyperscale context, and are often constrained by environmental liabilities that need to be untangled before a transaction is possible. Finding the inventory is a specialist function. Most developers will not do the work.
Second, structuring. A brownfield AI infrastructure development is not a single project. It is an integrated package: land, grid, energy supply, cooling, EPC, environmental remediation, planning consent, anchor tenancy. The platforms that work in this space have to coordinate five or six counterparty relationships in parallel and convert them into a single investable proposition. That is closer to project finance than to property development.
Third, and the one the market is most underestimating, delivery. A de-risked development on paper is still a delivery problem in execution. The brownfield gives you energised MW two to four years sooner. It does not give you a site that builds itself. The programme controls discipline, commissioning sequence, operational readiness and assurance frameworks required to convert a structured opportunity into an operating asset are precisely the same as for greenfield. They are just compressed into a tighter window with more legacy complexity in the ground.
This is where most of the new platforms in this space will struggle. Origination and structuring are deal-side capabilities. Delivery is an entirely different discipline, and it is the one that determines whether a structured opportunity becomes a producing asset on schedule or a write-down two years later.
The energy sector analogue
This pattern is not new. The European energy industry has been running brownfield conversions for forty years. Decommissioned coal plants converted to gas, then to biomass, then to grid-scale battery storage. Offshore platforms repurposed for CCS. Refineries pivoted into hydrogen and SAF. Each conversion sat on top of an existing site with existing infrastructure, accelerated by what was already there, but constrained by the same governance challenge: how do you deliver a complex multi-counterparty programme on a compressed timeline without losing control of cost, schedule or operational integrity?
The answer in energy was independent programme governance. Not the EPC, not the developer, not the investor. A separate control tower function whose only job is integrated programme assurance across cost, schedule, risk and operational readiness, with direct visibility for the capital provider. It is the discipline that allowed the European energy sector to deliver some of the most complex infrastructure programmes in the world, and it is the discipline that AI infrastructure now needs.
The data centre industry has not yet built this layer at scale. Most hyperscale programmes today still rely on the EPC's own programme function, the developer's in-house PM, or a tier 1 PMC carrying a contractor relationship. None of those models give the capital provider a genuinely independent view of delivery risk. For €500m to €2bn AI infrastructure assets, that gap is increasingly the largest unhedged variable on the balance sheet.
Where PMO Hive sits
PMO Hive is the independent control tower for hyperscale and energy infrastructure delivery. Our heritage is energy megaproject governance, four decades of FIDIC-grade discipline applied to oil, gas, LNG and offshore programmes. Our current engagements include a €1bn+ hyperscale AI compute campus in Northern Europe, where the same discipline is now being applied to the AI buildout.
For the development platforms now assembling brownfield-to-AI-ready portfolios, we sit alongside the deal team rather than replacing it. Origination and structuring stay with the platform. Independent assurance, programme controls, commissioning integration and operational readiness sit with us. The result is a development that arrives in front of investors with the delivery half of the risk profile already underwritten.
For the institutional and PE capital backing these platforms, the same logic applies in reverse. Origination is what closes the deal. Delivery is what produces the return. Underwriting one without underwriting the other is the largest avoidable risk in European AI infrastructure today.
The next twelve months
The brownfield thesis will not stay quiet for long. Several European platforms are already converting screening pipelines into pre-feasibility studies, and the first transactions will start moving from structured concept to construction within twelve to eighteen months. The platforms that win will be the ones that have built the delivery layer alongside the deal layer from the start, not the ones that try to retrofit it after the first programme slips.
Greenfield is not dead. But it is no longer where the smart money in European AI infrastructure is being deployed. The brownfield is the answer. Whether Europe converts that answer into operating gigawatts depends, as it always does in infrastructure, on who actually delivers.
