Data centre delays threaten to choke AI expansion
Almost 40% of such builds in US risk hold-ups, including projects tied to Microsoft and OpenAI
Delays to a swath of new US data centres threaten to slow the rollout of AI by the world’s biggest tech companies, with almost 40 per cent of all projects due this year at risk of falling behind schedule.
Major projects for Microsoft, OpenAI and other tech groups are likely to miss completion dates by more than three months, according to data shared with the FT by SynMax, a satellite and AI analytics group.
More than a dozen industry executives said campuses targeting hundreds of megawatts are being held up by permitting hurdles and chronic shortages of labour, power and equipment.
The bottlenecks are emerging as a key constraint on how quickly companies can turn vast spending on AI into revenue, raising concerns that billions in planned investment will take longer than expected to generate returns.
Hyperscalers are racing to build ever-larger data centres, pushing to bring facilities online that will draw at least 1 gigawatt of electricity — roughly a nuclear reactor’s output.
A handful of such projects are expected to complete this year, according to research group Epoch, including developments by Amazon Web Services, Meta and Elon Musk’s xAI.
But likely delays at many others highlight a growing gap between the scale of investment in AI and operators’ ability to deliver the infrastructure needed to support it.
“Financing at this scale is hard,” said Wes Cummins, chief executive of data centre operator Applied Digital. “Logistics are hard. Construction and operations are hard.”
To assess likely completion dates, SynMax uses satellite imagery to track construction progress from land clearing to foundation work. It cross-checks this against industry benchmarks compiled by IIR Energy, an industry research group, which tracks public statements, permit documents and conducts on-the-ground interviews.
One of the most prominent is a 1.4GW campus in Shackelford County, Texas, being developed for Oracle, which will in turn equip it with chips and provide computing capacity to OpenAI.
The 1,200-acre site is expected to host 10 buildings. Vantage Data Centers, the group constructing the site, in August said the first building was due for delivery in the second half of 2026.
Satellite imagery shows land cleared for six planned facilities, but as of early April only one showed signs of development. SynMax estimates the earliest possible delivery date for the first building is December, while a timetable in line with most comparable projects would push it out to late 2027.
Other OpenAI-linked campuses also appear to be progressing slowly. In Milam County, Texas, where Greg Brockman last month said a 1.2GW site was “taking shape”, satellite imagery shows construction has begun on only one facility. Of the group’s major Texas projects, only one in Abilene is expected to complete this year.
OpenAI said: “Our historic data centre build-out is on schedule and we will accelerate from here. In partnership with Oracle, SB Energy and a broader ecosystem of partners, we are delivering rapid progress in Abilene, Shackelford County and Milam County in Texas.”
Oracle said: “Each data centre we’re developing for OpenAI is moving forward on time, and construction is proceeding according to plan.”
SB Energy said: “The Milam County Data Center is on schedule and on pace to be one of the fastest data centres of its kind ever delivered.”
Two construction executives working on OpenAI-linked projects said there were not enough specialist workers, from electricians to pipe fitters, to meet demand across the build-out as companies race to construct clusters of increasingly large and complex facilities.
Strained grid capacity and shortages of equipment such as gas turbines and transformers are also causing delays. Remote locations are pushing labour costs up as much as 30 per cent, they added.
Doug O’Laughlin, president of SemiAnalysis, said the concentration of projects in some regions was intensifying competition even between providers serving the same end customer.
“OpenAI is [in effect] competing with OpenAI,” he said. “Workers are moving between projects in search of better pay cheques.”
“For those who stick around, even if they are doing double shifts, it’s going to be hard to meet these schedules,” he added.
Nebius, a so-called neocloud that builds computing capacity for clients, struck a deal with Microsoft last year to make a 300MW facility in Vineland, New Jersey, successfully completing the first tranche of capacity by the end of 2025.
Satellite imagery suggests the latest phases of the project are progressing more slowly, with structures in place but timelines slipping. Thermal imagery taken at the site indicates that equipment has yet to come online.
The Vineland project has faced permitting challenges and local opposition. A formal public comment process has also been requested, a step that typically extends timelines and signals growing community resistance.
Microsoft declined to comment.
Nebius said: “All capacity tranches under our agreement with Microsoft to date have been delivered on time, and we currently expect to deliver the remaining tranches on schedule. As of now we are not aware of issues that would materially affect this.”
SynMax estimates more than 60 per cent of projects scheduled for next year have yet to begin construction, adding to concerns over delays to the industry’s expansion pipeline.
“There’s a focus on speed and development brushing up against regulatory lag,” said Josh Price, an energy director at Capstone, a strategy firm. “The scale and complexity of these projects is necessarily going to lead to more scrutiny and potentially delays.”
Cummins at Applied Digital said: “We’re going to see a number of blow-ups and delays this year. My focus is on making sure it’s not us.”