Tariffs and trade turmoil driving up cost and build times for datacenters

Biz needs AI infra for training ever larger models, but something's gotta give

World War Fee Datacenter operators in Northern Europe say US tariffs and growing global geopolitical instability are inflating costs and causing delays to construction projects.

Demand for rackspace has shot up in the past year or two, driven by AI infrastructure for training ever larger models.

A survey by datacenter specialist Onnec finds that AI workloads have increased by an average of 42 percent over the past twelve months, with nearly two-thirds of operators saying that requirements to support AI were higher than expected, stretching the capacity of both existing and planned sites.

The report is based on a survey of 250 senior decision makers from the UK, Ireland and the Nordic countries who say they are being hampered by skills shortages, geopolitical turbulence and design complexity.

The news could worry UK premier Keir Starmer, whose government wants to speed up the building of datacenters all over the country to drive its AI ambitions, but perhaps he has more pressing concerns at the moment.

As far as tariffs go, 69 percent of respondents cited Donald Trump's US administration's on-and-off vacillating tariff dance in the US, along with other geopolitical issues, as further driving up costs and increasing delays for them.

Back in April, system builders warned that tariffs would likely lead to a hike in server prices, while  financial analyst Jefferies said the uncertainty was leaving the tech industry in limbo, because equipment is often made up of components from multiple sources, causing headaches in the supply chain.

"The AI race is global, but protectionist policies are forcing operators to make cost-based decisions that will limit the quality of infrastructure they can deliver," claimed Onnec Nordics General Manager Niklas Lindqvist.

The company's report finds that more than half of operators expect the AI craze to shorten the lifespan of current datacenters. This is because the most demanding AI workloads are rapidly outgrowing existing power, cooling and bandwidth capacities.

Nearly three-quarters (74 percent) of operators say they have had to to rethink power, cooling and location strategies for AI-ready facility builds.

This has already affected some global players. Earlier this year, reports surfaced of Microsoft cancelling leases on some bit barn capacity in the US, which sparked questions over whether it had overestimated the AI infrastructure it would need. In reality, it seems the cloud giant needed to reassess which sites could meet the power and cooling needs.

"With AI evolving so fast, operators need to make sure datacenters are built to last, ensuring future sites have dynamic capacity when they're up and running," the report notes.

According to Savills, a global real estate provider, construction costs for datacenters builds in Europe were already on the up in 2024, rising 6.5 percent on average to $9.1 million per MW. Zurich, London and Frankfurt were the most expensive cities to build in.

Land, building and powered shell, electrical systems, HVAC, mechanical and cooling were the most costly elements.

Meanwhile, 79 percent of respondents in Onnec's survey pointed to skills shortages as another factor likely to delay all those construction projects.

"With a shortage of specialist engineers, operators may be forced to scale back or delay their ambitions, turning what were intended as cutting-edge facilities into compromised, less efficient builds. Meeting these challenges and easing cost pressures will be key to ensuring AI infrastructure is built to support long-term needs," Lindqvist said.

Onnec is also keen to call out cabling as an oft-overlooked weak link in datacenter infrastructure. The report says 70 percent of operators agree that poor-quality cabling can compromise performance and reduce resilience, with 27 percent citing a shortage of skilled personnel for installation and maintenance as a challenge. ®