Dr. Atif Ansar on why 30% of data center assets are heading for the disaster zone — and what separates the ones that aren't.
There's a persistent myth in the data center industry that these are simple projects. Sophisticated people — senior executives, experienced engineers, seasoned developers — repeat it constantly: data centers are like warehouses, the builds are predictable, we've done this before.
They haven't. Not at this scale.
A 100 megawatt data center requires $10 to $18 million per megawatt just to get to shell — land, mechanical and electrical systems, cooling, redundancy. Add the compute and you're at a minimum of $3 billion, potentially $6 to $8 billion for a single asset. By the Oxford definition, a megaproject is anything above $1 billion. The data center industry isn't adjacent to the megaprojects world. It is that world.
That's the framing Dr. Atif Ansar, Executive Chairman and co-founder of Foresight Works, brought to a recent conversation on the Data Center Go-To-Market Podcast with host Joshua Feinberg. The full episode is embedded below. What follows are the arguments worth sitting with.
The constraint isn't demand. It's execution.
The current data center boom has one genuinely unusual characteristic: capacity is being built under contract, not speculatively. Unlike the Telecom boom, where capital chased anticipated demand, today's hyperscale and co-location builds are largely pre-contracted with customers. The demand is real. The revenue is waiting.
Which means the only thing standing between a developer and a decade of contracted revenue is their ability to actually deliver the asset on time.
This is where the industry is failing. Every month of delay on a 10 to 15-year fixed-term contract is lost revenue that never comes back. The earliest revenue in any long-term contract is the most valuable — any project finance professional will tell you that. Delay doesn't just push revenue back. It destroys it permanently on a net present value basis.
Why optimism bias is structural, not accidental
When people project a 100 MW data center to be delivered in 18 months, they're not lying — most of the time. They're experiencing what Nobel Prize-winning research by Daniel Kahneman and the behavioral economics literature has documented extensively: human beings systematically underestimate how long things take, even for low-stakes tasks.
In the data center industry, this psychological optimism is compounded by a second layer: political bias. Deals get won on bold timelines. A build that realistically takes 24 months gets sold at 18, or 16, or 11. Everyone in the room knows the number is aggressive. The commitment gets made anyway. Delay is effectively baked in at the point of sale.
And expertise doesn't protect against this. Atif's research suggests it can actually amplify it. He uses the image of an optical illusion — two lines of equal length that look different depending on the arrows at their ends. You can know intellectually that the lines are the same length and still see them as different. The only correction is measurement. A ruler. External reference data that overrides your intuitions.
The quadrant: 30% disaster, 20% happy
Across Atif's research on megaprojects — originally developed from US defense project data and replicated for data centers — a consistent pattern emerges when you plot time overrun against revenue performance.
Roughly 30% of assets land in what he calls the disaster zone: significantly delayed and underperforming on revenue forecasts. These are value-destroying outcomes that in some cases threaten solvency. Another 20% land in the happy quadrant: ahead of schedule and above revenue projections.
The variance analysis between those two groups points to two dimensions. The first is cognitive orientation — whether the leadership team is genuinely open to what the data is telling them or closed off behind years of industry experience. The second is realism vs. magical thinking in how they approach forecasting. The distinction isn't ambition — the top performers are ambitious. It's whether ambition is grounded in evidence or detached from it.
Data centers aren't warehouses. They're refineries.
Part of what sustains the "it's a simple build" myth is the physical form of early data centers — relatively modest co-location facilities that sat comfortably inside real estate portfolios. That framing has never been updated to match what these assets have become.
Today's hyperscale data centers are closer to oil refineries in complexity and capital intensity than to cold storage or grocery distribution. Rack densities have moved from 2.5 kilowatts to upward of 133 kilowatts. The cooling requirements, power redundancy, and mechanical systems involved are orders of magnitude more complex than they were a decade ago. And the speed of change — construction volume is up 10 to 20x since 2020 — has outpaced the industry's ability to develop the labor pool, supply chain, and project management discipline to match it.
The companies that are getting this right are treating these assets with the rigor they deserve: clear-eyed about reference class data, open-minded about new approaches, and willing to measure rather than assume.
What this means for go-to-market
One of the sharper moments in the conversation was Atif's answer to what better GTM looks like for companies selling into this industry. His answer: stop being horizontal. Every company that tries to serve every segment of the data center market — operators, hyperscalers, co-location providers, hardware vendors, software platforms — ends up serving none of them particularly well.
The highest-performing companies have extremely narrow segmentation: a clear sense of who they are, what they're genuinely excellent at, and which customer profiles are the closest match. Not the biggest TAM. The tightest fit.
And when it comes to selling: move away from activity metrics. The instinct to fill long enterprise sales cycles with outbound volume, conference appearances, and automation sequences is understandable. It's also a trap. One conversation that generates genuine understanding and moves a customer's thinking is worth more than a thousand emails that produce no response.
That's the difference between input fixation and outcome fixation — and it's the same underlying discipline problem that separates the data center projects that deliver from the ones that don't.
Foresight Works builds AI-powered project delivery software for data center owners and developers. If you're thinking about delivery risk in your current pipeline, get in touch or subscribe to the FastTrack for Dr. Ansar's ongoing research and thinking.