Data centres built for AI are running into limits on power, water, materials and public patience, according to analyst firm GlobalData. Its Strategic Intelligence report, "The Constraints Reshaping Data Center Development", expects annual data centre power consumption to rise from 590 terawatt-hours in 2025 to 1,520TWh in 2030, or 4.5% of global electricity use, up from 2.1% in 2025.
Most announced projects remain stuck in planning. Those that broke ground between Q3 2025 and Q3 2026 waited just over two years from announcement to construction start, against about 18 months for those that began between Q1 2020 and Q2 2025.
Grid access is the biggest bottleneck. The report cites the IEA's 2026 Electricity Outlook, which counts more than 2,500GW of renewable, large-load and storage projects in connection queues worldwide, with waits of seven to 10 years in Europe's main data centre hubs. New grid infrastructure takes five to 15 years to build, against one to three for a data centre, which is pushing operators towards on-site gas power.
Opposition is hardening. Martina Raveni, senior analyst on GlobalData's Strategic Intelligence team, said: "Data centers are increasingly located in water-stressed areas, and their construction, along with the energy-intensive training and running of large language models (LLMs) are adding new sources of emissions." Citing Data Center Watch, GlobalData says at least 75 US projects worth roughly $130 billion were blocked or delayed in the first quarter of 2026, the largest single-quarter total on record, and similar backlash is emerging in the UK, Chile, South Korea and Japan.
Raveni pointed to the supply chain. "A single hyperscale facility can require up to 50,000 tons of copper, while AI data centers need 10 to 20 times more fiber than traditional builds, at a time when fiber manufacturers are already at capacity," she said. A shortage of electricians and other skilled trades is adding delay.
Operators are responding with closed-loop cooling, water stewardship commitments, renewables, battery storage, fuel cells and longer equipment lifespans, though GlobalData describes progress as uneven. Raveni's conclusion: "These constraints are already translating into real delays for AI infrastructure." Compute becomes dearer and harder to get, she said, pushing AI developers towards smaller models and narrower use cases.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
