The retrofit fast path ends at the substation

Writing in Data Center Dynamics, Niklas Lindqvist, who runs Onnec's Nordic region, argues that upgrading the buildings already standing is the fastest way to add AI-capable capacity, because global data center capacity is expected to nearly double by 2030 while new sites sit behind grid constraints, planning timelines and supply-chain delays that stretch greenfield construction across several years. The power constraint is the part of that queue a retrofit cannot skip.

The mechanics are unglamorous: a retrofit frees stranded capacity by upgrading the systems that cap higher-density workloads—power distribution, cooling, containment, fiber pathways. Sequencing matters more than specification, because the work runs against live operations and has to follow a map of power, cooling, structural and network dependencies, and Lindqvist frames the question as whether a site can be upgraded safely and without compromising uptime. Phased upgrades can put capacity online in months rather than years, and 29 percent of operators lean toward retrofitting to expand AI infrastructure, even as global data center construction costs for traditional facilities rose roughly 5.5 percent in 2025.

Older halls, built for air cooling and cloud workloads of 5 to 10 kilowatts per rack, sit a long way from the 30 to 80 kilowatts or more that AI workloads demand, and feasibility turns on things you cannot see from the data hall floor—floor-loading limits, restricted ceiling heights, constrained plant space, insufficient electrical distribution. Power is usually the hard ceiling, and it compounds, because even where grid capacity is available, the switchgear, backup systems and rack-level distribution behind the meter often need substantial rework. The UK's total capacity stood at an estimated 2.9GW in early 2026 against a stated requirement of at least 6GW of AI-capable capacity by 2030.

Retrofitting buys a way around the planning queue and the construction schedule while doing nothing about the interconnection queue, because power is the binding constraint. A site with energized, permitted capacity and the wrong cooling can be brought forward in months; a site with empty floor space and no substation is a land bank with a roof, and the electrical room is where the retrofit budget actually goes. In that light, the 29 percent Lindqvist cites are the operators who already hold the scarce input.

The expected doubling assumes existing sites absorb a meaningful share of the load, and the ones that can are defined by their substations more than their square footage. Watch whether the operators leaning toward retrofit convert that intent into feasibility studies over the next year, and whether the ceiling they find is the slab or the switchgear.