Every developer knows the feeling. A task slips from Monday to Wednesday. Another one that was due Friday lands the following Tuesday. Nobody panics. The team says it'll catch up.
Dr. Atif Ansar says that's exactly how billion-dollar data center projects fall behind.
"A bunch of activities that are meant to start this week on Monday that started on the Wednesday seems like nothing, but it's actually a lot," said Ansar, executive chairman and co-founder of Foresight, when we spoke during Yotta 2026.
Ansar is an Oxford researcher who studies why megaprojects run late. Foresight, which raised a $25 million Series A led by Macquarie Capital Venture Capital in March, applies that research to data center construction through a predictive project delivery platform.
Foresight's research shows nine out of 10 large data center projects will miss their delivery targets. That's happening while capital is plentiful and vacancy rates sit below 2%.
"Demand is not the problem," Ansar said. Scale is. U.S. data center capacity has grown roughly 20-fold in five or six years. Supply chains haven't kept up, and neither have the contracts and delivery models the industry still relies on.
Earlier that day, Ansar spoke to a standing-room-only panel audience of CEOs, investors, and contractors. The idea that landed hardest was simple. Delays are the norm, and people can plan for them "without becoming defensive about them."
Ansar blames optimism bias, and experience doesn't cure it. A team that delivered a 10-megawatt project in 15 months may assume it can deliver a 100-megawatt project on the same timeline.
"More often than not, these biases don't go away just because of years of experience," he said. Historical data shows average overruns of about 30%, yet teams assume those delays happen to other people. Ansar says they should study past projects of the same type and plan mitigation before work starts.
I asked whether developers would recognize how a data center schedule slips. Ansar said the human side is the same. The difference is structure.
"Data centers represent a much more linear or serial sequence of events," he said. You have to pour the foundation before you build what sits on top of it. Software teams can often work on many parts of a system at once. Construction can't.
That makes small misses expensive. Ansar described a busbar installation where the team skipped a design review, a factory acceptance test, and a site acceptance plan. Nobody noticed until installation, when the parts didn't fit. That became a four-month delay.
For a developer, it's like skipping integration tests and finding the break in production. Except here, you can't roll back.
So what does Foresight actually look at? The schedule comes first, typically from tools like Primavera P6. The platform also takes in design information and contract documents that commit parties to specific dates. Much of it arrives as spreadsheets and semi-structured files. Foresight uses its own ontology to normalize the data into one view that owners, contractors, customers, and suppliers can share.
Ansar doesn't try to clean everything. He starts with the major milestones that align everyone, then uses them as scaffolding for the details.
"If you try to fit a jar with rocks, pebbles, and sand, and you start with the sand, you will miss the rocks and the pebbles," he said. "You've got to start with the rock first."
Chasing every five-day activity isn't worth it. Missing a critical milestone, like electrical equipment delivery, is.
The forecasting combines both approaches. Foresight's reference-class research on past megaprojects provides the training data. The machine learning models look at "both past and contemporaneous data," Ansar said, to estimate where a project is likely to land.
No model will ever show zero risk. Ansar compared it to driving from New York to Washington, D.C. You might arrive a few minutes early, but never in half the time, and a flat tire near the end can still make you late. Uncertainty narrows as a project moves forward, but it never goes away. "You're not done until you're done," he said.
The hardest part may be getting honest data from contractors, who could see the tool as a way to catch them. Ansar rejects that framing. "Our tool is to drive collaborative discipline with mutual accountability," he said.
He pointed to Terminal 5 at Heathrow, where everyone on the project got a Terminal 5 email address regardless of which company employed them. The goal was one team serving one project. Foresight then acts as the shared control tower where owners, contractors, and suppliers look at the same data.
Warnings then go to the right level. A slow foundation crew is handled with the contractor. A missed revenue milestone goes to the CFO or chief revenue officer, who can mitigate or renegotiate before penalties hit.
Culture matters too. Ansar says construction relies too heavily on the heroism of its workers, at a real burnout cost. Safety improved over 30 years through better processes and leadership. He thinks schedules can follow the same path.
Developers have been living with this for decades. The fix isn't more heroics or tighter deadlines. It's watching small slips before they compound, sharing uncomfortable data early, and building schedules that admit uncertainty up front. With hundreds of billions of dollars riding on AI infrastructure, the data center industry is learning the same lesson at a much larger scale.
