In The Arena by TechArena - Foresight’s Atif Ansar: AI Infra Execution & Delivery Risk

Episode Date: July 20, 2026

In this episode of Data Insights, Allyson Klein and Jeniece Wnorowski sit down with Atif Ansar, Co-Founder of Foresight Works, to explore the execution challenges behind large-scale AI infrastructure projects. The conversation reframes the perceived AI “bubble,” arguing that demand is not the issue. Instead, the ability to deliver projects on time and manage complex interdependencies is the true constraint.

Welcome to Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome in the arena. My name's Allison Kloin, and this is a Data Insights episode, which means I'm here with Janice Norowski. Janice, how are you doing? Hi, Alison. I'm doing well. How are you? I'm great. I'm really excited for our conversation today, and I know that you brought someone really special. with you. Why don't you introduce our guest and talk about the topic that we're going to be discussing today. Yeah, thank you. Today we have Tadip Ansar. Atib, thank you so much for joining

us today. I mean, Alison, we've had a lot of conversation on the topic of AI factories, neoc clouds. We've talked a lot about just the world of AI. What is the infrastructure doing? What are they up to? What does it look like? But it's very rare that we have the opportunity to need in organizations as helping to deliver on-time projects, really complex on-time projects for data center build-outs. I'm excited to have answer here and to talk a little bit more about what foresight is up to. Batif, welcome to the program. I just want to start with, can you introduce ForsightWorks? This is the first time where SiteWorks has been on Tech Arena. And tell us a little bit about what your leadership at ForsightWorks is responsible for. Thank you, Alison, Janice. It's a pleasure to

be here with you. So Foresight is a project delivery system. Power, of course, with AI, but also a proprietary data set and methods that I developed over 15 years of academic work at Oxford University. And the basic idea is projects are human systems, and they require a really unique way of delivering them to get them right, which is what we help people do. Projects are also a maturity journeys. So it's not just about getting good outcomes, like delivering them on time, but also about good process. So Fawcine is a way for people to create that process, almost like a fitness app, if you will, of how to not just produce good outcomes, but also deliver them with a maturity so you can build a repeatable, consistent battle. I'm one of the co-founders. So my role is in

providing the product vision, also making sure that our data science models are built really well, and then are leading our go-to-market ethics as well. Amazing. So Tith, a lot of the AI infrastructure conversation is framed around technical bottlenecks. But you've made the case for this bigger gap is more about execution and leadership. Why is the human element of all this becoming a limiting factor? Great question, Janice. So look, our mission is world-class project delivery. And like I mentioned earlier, world-class project delivery is both about outcomes, but also about the human process, but with which we go about it. What my research at Oxford University showed is that people bring tremendous skill, tremendous genius and energies to when they're building projects,

but we also bring our biases. And we have political biases as well as psychological biases. Psychological biases of the nature of being just overly optimistic. So even low-stakes tasks like grabbing a coffee, I might say to you, Janice, here, we back from a coffee in five minutes, but if you clock me, I might be back in 10 minutes or 20 minutes. It's very unlikely here back in back in five minutes. It's much more likely that I'm going to take a lot longer than that. And we don't necessarily think about it. And it's not like I'm being malicious in that. It's just simply we don't necessarily forecast all the impediments to what we may encounter

in even completing a small task like gathering coffee. And these problems magnify at a project level. The antidote to that, of course, is both better planning, which in the literature, Donald Kahneman won the Nobel Prize in economics for this in 2002. He calls that the inside view. So you have to develop a very detailed movie read of the future of how to, How is it that your project's going to go from achieving the funding released, ensuring that your architectural models are properly coordinated,

up to doing the procurement? All of that's got to be documented in a pretty clear database of some sorts. And second, then from executing against that in a very diligent manner. So Foresat is really a platform that enables people to both build those plans, what we call generative scheduling, but then also track against them, what we call generative control in being able to make sure that they are meeting their micro daily commitments, which then add up to the monthly and annual goals in terms of where the public's going to

end up. You know, Tia, in doing my research on the show, I came across some things that you'd said about the AI bubble. And less people posit that the AI bubble is about diminishing demand at some point. But you've argued that bubble risk is actually whether companies can deliver what's already been contracted. Can you explain that framing? So look, as we all know, there's just a huge amount of capital going into AI infrastructure right now.

There were about 500 megawatts worth of data centers being built five years ago. There's at least 10 gigawatts that have been announced or under construction in one way or another. So just the gap is massive. Now, many people fear that this massive supply of new capacity is not going to find in buyers and we will be stuck with extremely expensive AI factories that are unused. What I've argued previously is that's simply not the case for two reasons. First of all, we've seen the revenues of all of these AI companies accelerate really rapidly. So with fraud or chat, UBT, I'm sure you've seen their revenue numbers are through the roof.

And in our own daily use, I'm sure this podcast video is going to go through some form of AI to correct my arms and ours at some point. So all of that is consumed lots of... server space and some data center around the world. And demand is not an issue. And we see them the baby as well. A second kind of thing that's helping data centers, these are not speculative investments. They are built under contracts. Unlike the old telecom bubble of 2001 or 2000, when the telecom towers were being built speculatively, these data centers are being built under contract from very high-rated AIA companies like hyperscalers. And as a result, they have 15-year leases.

So the developers are not taking a huge amount of risk. And as a result, the investors also have the backing of a lease from Microsoft or Amazon Worked Services. So these are not speculative investments. These are ultimately a credit-worthy investments. So in all of that, the key variable that can potentially let the data center builders down is if they fail to meet their lease commitments. Because these leases are so strict and because of their finite nature of a 10 or 15 year lease commitment, every month's delay on average costs them $15 million or more if it's a 100 megabond data center and much more if it's a bigger data center.

And that early revenue, because of the way financial models work, is the most valuable revenue and it's never going to come back. So basically more than a three-month delay dissipates the net as in value of even a very, very large build. on top of the lost revenue, you have the issue of service credits or the SLA penalties because hypers themselves are so exposed to this. So they extract a pound flesh for you being late. And that basically means that the economics of these can become very adverse, not because the demand isn't there, but because there is demand. And the revenue of writing and as well as the penalties for missing your targets

becomes potentially insolvency causing for companies that are not one managed. You've drawn parallels to kind of seeing abandoned infrastructure projects growing up and not so much failures of funding but failures of planning. Where do you see that same dynamic showing up in today's data center buildouts? So look, there's a tendency in the market to introduce or announce quite a lot of what we call Braggawatts. So these are commitments that people make in the market, free funding. And people do that partly because they may have a piece of land. They're trying to improve the value of the land. they may not have secured the power.

They all kinds of sorts of reasons for announcing, also to scare their customers. There's a lot of kind of game theory in this very dynamic market that's going on. The key is people know who the mature developers are, and that's exhibited through their process maturity. So one thing we are noticing is that very large number of projects are getting canceled because of community opposition. And what I've argued is that communities have a pretty good sense. it's not just an envious, which is what it gets blamed on. Communities have a pretty good smell test for whether or not the developer is a sensible, reasonable citizen that is also going to create prosperity both for their own company and their community.

Or are they a cowboy that's trying to make a quick buck? And communities have a good smell test for that. And if they smell a cowboy, they don't want them in there. But if they smell a responsible citizen, they definitely want those. jobs. They definitely want those schools. And the process maturity that produces on-time delivery is also the same process maturity that produces better relationships with community. It's a way of thinking about the world and of being organized and orderly in your business conduct, which leads to these good outcomes, hence my emphasis on the strategic governance piece again and again.

The process that get cancer are the ones, there's nothing wrong with the project per se, but their preparation is so poor that they can't build the confidence in the communities and the investors, in the hypers, to be able to get them off the ground. There's a nuance here about process and process management as well. And one of the things that I've experienced a lot in terms of complex projects inside large organizations is that projects often look on track when they suddenly aren't. And I think that data center buildouts would be the Uber example of this. What has your experience been in the scope and how do you see large organizations navigating this challenge? Great question, Alison. So this issue of projects looking like they're on time and then suddenly

not is what we call the watermelon problem. So all projects green from the outside, people are suddenly turned red. And it's an infuriating problem causes a lot of both emotional as well as financial baggage, right? So often this arises because of the over optimism, that both the political as well as the psychological bias that I refer to. So people hope they'll catch up, but they doubt. The purchase orders that should have been issued for procurement that haven't been issuing. The problem with project shows they are very interdependent systems. So lots of things depend on each other.

So in order for you to get your equipment six months from today, you need to place the purchase order today. If somebody forgets to do that, you're not going to have that piece of equipment six months from now. But you will discover that delay six months from now unless you're tracking. it very vigilantly today. So I think that's precisely where we build foresight as this control tower that is able to both provide what needs to happen in real time, but then also get people the visibility that they can dive into the most minutest tasks and observe those variances as pretty as possible, given whichever level they might be at, because very small variances, just by thousand cuts,

as we call it, could really de-linked projects and cause these kind of surprise bad outcomes to happen all of a sudden. In one case, you've actually uncovered a five-month delay hidden beneath distorted reporting. How did these gaps emerge in practice? Look, there's lots of causes. When I was a younger academic, in my academic days, I was always very attracted to big, singular dramatic events that caused the delay, like a big flood or a big supply chain disruption or a utility not making equipment.

As I matured in my own thinking and looked at the data more and more closely, these big events matter. I'm not denying that they exist or they don't matter. But really, what lets people down is the accumulation of these very small variances. And then myself became very interested in skiing as an athletic hobby. And I realized the sports is the same, like really good athletes, obsessive of micro details, be it in tennis or in skiing or golf. And it's actually those micro details that make the difference between elite performance and average performance. And average performance, the type of thing. So same thing with projects. It's like really focusing on both the big macro picture, of course, which in sectors are clearly drawn to, but then also

having this handle on the micro issues and just being on top of it. And all of it basically needs to again, creating this very detailed plan. And I'm not necessarily talking about thousands and thousands of rows, but having a very, very clear idea of what's a design, procurement, physical works, any P works, etc. You need commissioning in order to get to a complete project. To give you like a quick tip, for example, one thing that we've discovered again and again is people love concrete. So there's a lot of construction, civil and structural activities

that are readily represented in the schedule. Mechanical, excellent and plumbing, because it's much more gnarly, but people are just going to put it in a very broad brushstroves. And because it's so gnarly, that's where you have a lot of delays. nearly 60 to 80% of a data center is MVP. So unless you really define the NEP footprint very clearly, you will have a delay. As much as it's not vivid and colorful like a precast concrete, the building envelope might be. Similarly, with commissioning, so people sort of put all of commissioning and the back end of a big data center

and think they're going to get it done in three months, what we found is the level one and level two commissioning, which is the factory acceptance test, the site acceptance test, have to happen way before the commissioning, phase really begins in earnest. And unless you've done that work, almost co-joined with your procurement and with your equipment deliveries, it's going to be very hard to commission in anything under six months, for example. So there's sort of a lot of things that people learn by experience as well. And that's why I'm most excited about the world of AI is it can take a lot of human heuristics, a lot of human experience of what's worked really well, and begin to build a library of knowledge to you, which helps to onboard younger employees,

just join or people who come from other sectors, a respective age, can onboard them really quickly into that particular trade or into that particular way of building. Now, you emphasize something called sequence and favor that over activity lists, if you will. Why is this so important when you're considering hyperscale builds? So this matters in all sorts of construction, but particularly in data centers, there's a kind of a path from building them.

And data centers are particularly linear or serial. forms of construction. In many ways, they're basically very large computers. And like how computers execute software programs, they have to sort of execute a chain of command of the sort of things that need to happen in order to build a data center. So what we've seen in the industry is a lot of people rely on external spreadsheets to basically create a very long list of activities that they want to do in order to get a project built. The reality is that these activities are interdependent. So unless, for example, you've done a design review of a particular piece of equipment, your PO that you're going to issue can't be quite right.

And unless that PO is issued, it's not going to get delivered on time, at which point it needs to have a site acceptance test, and only then it should be installed. So people, but, for example, rush into issuing the POs without doing the design reviews. They then receive the piece of equipment, they forget to do the site acceptance test. And then when they come to installing it, suddenly realize, oh gosh, those bus bars don't fit anymore. They're the wrong bus bars. We could have told you that eight months ago, based on those misses,

but it also means having a culture in which people listen to that without just railroading through these missions. Yeah. And the TAPE said he was really excited about AI being able to implement a lot of this, and you've actually done so. You've used AI to kind of stress test schedules before major capital is committed. And so with that, what does that surface that traditional planning kind of misses? Fantastic question, Janice.

So I think one of the key things that we find again and again is that when people build schedules, to Allison's earlier point as well, they contain a very large laundry list of activities they need to get done. It's a little bit like if you heard the analogy about the rock pebbles and sand, if you're trying to fill a jar with rocks, pebbles, and sand, if you start with the sand, you can't get any rocks and pebbles then. because a lot of the times these schedules are filled with sand. So they'll contain hundreds of activities about putting table or installing pipe. They'd be missing the rocks like electrical room complete or mechanical corridors complete. So when we're passing through this data with our AI, some of the things we're looking for is they may have the sandy detail, but does that sandy detail culminate into these rocks and pebbles that is an example? that as an executive can rely on to know that certain big elements of the project have been achieved, certain waypoints have been crossed.

So some of the ways in which we enrich that data is by analyzing that network, we help them look at what milestones they are likely missing or what gates they're likely missing and help them insert those back in the appropriate places and link them in the right places to make sure they've got a schedule that hangs together out of really well. And that that helps to not just align the front line, which tends to think in very tactical ways, but also the top line, which tends to think in these episodic brigates. So obviously pulling cable and installing pipe racks is heroic work, but none of that matters unless the electrical room is done.

And making sure that both those things are working in tandem are some ways in which AI helps to create that coordination between the good work that happens on site, as well as the way people think about projects in the H-Q. Now, as AI infrastructure scales from single sites to portfolio level programs, what leadership and management gaps become most exposed in your mind? So I think big projects and big portfolios need big people. And I think that's the biggest thing that gets exposed is your ability to zoom out and look at the whole system in its totality. I think a lot of people are technically extremely competent and adept, but that can also mean

that they can be very narrow in terms of their approach. and data centers are becoming so complex. They're almost as complex as all refineries these days. Just by being very good at one thing is no guarantee that the entire integrated system is going to hang together. So I think what's getting exposed is this need for people to be able to, almost like a T model, both be able to dive in deep into their specific domain of which they're an expert, but even more importantly to extract themselves out of that domain, and think in a much more horizontal, much more integrated system sort of way to make sure that

the thing hangs together. So to use an analogy, if you're trying to put Humpty Dumpty together, it's not just that you're worrying about the individual fingers, that they're head in place. You've got to put together the torso and the head and the limbs have all got to go in the right place. And I think what's becoming really crucial for people. And then this is, by the way, something only people can do is to take that system level view. where AI is really valuable is then surfacing aggregate insights that help people look up from their immediate contacts and take that big view. So doing out, you said execution problem is bigger than data centers, right?

It affects the energy transition and future prosperity. Why is delivery capacity becoming such a critical constraint? Look, first of all, there's just a lot going on in the world, right? So we're building a lot. the world is undergoing at least three massive revolutions all at the same time. One is the digital infrastructure, which is the AI build out and all the telecom and data infrastructure, digital infrastructure that goes with it. Second is the energy infrastructure not just to power the digital infrastructure,

but the climate change issue that we've had for a number of years in the energy transition that goes along with it. So that problem hasn't gone away. So people are still investing heavily in simply keeping pace. with the total volume of energy that needs to be built in the world, and total volume of clean energy that needs to be built in the world. And third, given the geopolitical dimension, a massive change in defense and defense-related spending.

So all of that's meant that the total project delivery needs are just more magnified today than they have been in previous decades. And that's just three sectors. You have evergreen sectors like pharmaceuticals, or even poverty, aviation, building schools, social infrastructure. None of that is going away. So the pressure is intense in terms of both the construction as well as delivery of it. So that's the demand side on one aspect. The second is supply of skills. We just don't have enough people in the world right now is still people to be able to keep

pace with this level of change. So give you an example, there's about one trillion dollars worth of data centers being built in the world. That's about 10% of the global construction market. Yet, the number of people who are working the data center industry is just 0.05% of the total construction labor force. So there's a massive mismatch between what they're trying to build in terms of construction volume and just number of people involved in the industry. So it's still a cartridge industry in terms of the footprint. So I think that those bottlenecks are, again, they need a lot of technology, automation and AI to simply keep pace. They also need education. So I'm still remain an educator at heart. And I think that we just also need to upskill people and train them

in the art of becoming better project managers, which is core to my presentation as we've been foresight. Now, I have one final question for you. For executives investing significant capital into AI infrastructure, you talked about the trillion dollar of investment collectively. What discipline or mindset shift matters most to ensure that these projects actually get from ideation to deploy it. Great. So I think the key mindset is what a colleague of mine at Oxford, Ben Flyberg, calls think slow, act fast.

The default tendency is very action-oriented, which I like. I'm an entrepreneur, so, you know, bias to action is a good thing. But when you're thinking in a strategic investment framework, you can't be fast-tracking your projects because trying to put shovels in the ground as quickly as possible is actually going to slow you down drastic. It's going to get a blowback.

from the communities, people once thought through investment ideas, not rushed ideas. Think platform, think repeatable, think consistent, think long term, think structured, think process in order to get those really good outcomes. And that's kind of what Fulcise is trying to create is this process of think in a very clearly orderly fashion right up front, build those systems right of front, put in place that strategic governance, and then you have a platform to be able to do repeatable projects again and again in our solid, orderly manner. It is fascinating, Atif. Where can our listeners go to follow you and get more information about your work?

So please follow me on LinkedIn at T. Fansar, Dr. Tfansar. And please also visit Foresight on www.f, www.f, we'd love to stay in touch with you. I see if that was so fascinating. Thank you so much for walking us through this. I would love to be back on the program some point. And, Janice, thank you so much for being on the show. Thank you, Alison. Thank you, again. I'd hope to. See you guys again.

Thanks for joining Tech Arena. Subscribe and engage at our website, Techorina.ai. All content is copyright by TechRena.