Ep. 039 - [EMERGENCY] Brad Gerstner Cut Our 2027 Forecast in Half. We Raised It. (Datacenters)
Jordan NanosJeremie OntiverosReyk Knuhtsen
- SemiAnalysis raises its 2027 data center forecast from 43 GW to 50 GW amid public challenges to its scale. Against expectations closer to 25 GW, the team points to construction already underway, tracked “not even campus level, but building by building,” alongside permits, power allocations and supplier activity.
- The forecast already incorporates construction delays and gradual capacity ramps, making another blanket haircut potentially duplicative. Jeremie says the team adds one to three quarters where warranted. At the discussed Crusoe site, the model includes under 200 MW rather than the full gigawatt critics assumed; Jordan notes that an additional haircut could “double count” adjustments already made.
- Jordan points to more than $1 trillion of signed data center leases as support for the construction outlook. Jeremie emphasizes “binding contracts that cannot be cancelled,” with private operators contributing commitments that public-company disclosures miss. Early construction orders, electrical equipment and cooling deliveries provide additional checks on when those commitments become usable capacity.
- Modular construction is the team’s central explanation for delivering more capacity despite labor constraints. Reyk contrasts traditional 18-month builds with DayOne’s roughly nine-month timelines. His physical example is a shipping-container-like power hall slotted into a building, replacing extensive on-site electrical work: “modular is what’s making this work.”
- Power remains the immediate constraint even when GPUs and physical space are available. Jordan highlights operators installing more GPUs and power-capping them to fit the available electrical envelope. Jeremie says buyers can obtain GPUs today but struggle to secure land, power and shells within three to six months, helping explain prevailing skepticism.
- Data center oversupply entering 2028 is a risk Jeremie watches, rather than a formal forecast. If facilities expand faster than chip supply, today’s bottleneck could reverse; he sees that possibility as relevant to semiconductor fab and equipment investment. When Jordan checks how firmly he is making the call, Jeremie clarifies: “I didn’t say formally. I’d say it’s the main risk as we get into 2028.”
1. The 50 GW forecast rests on buildings already taking shape
Jordan opens with the disagreement: Brad Gerstner treated SemiAnalysis’s previous 43 GW forecast for 2027 as “too aggressive,” while skeptics expect roughly 25 GW. The team has instead raised its forecast to 50 GW. Its evidence includes 40 pictured construction sites representing less than half of the global data centers shown.
Jeremie traces the model’s record to a March 2024 outlook anticipating acceleration in 2025 and 2026. Jordan corrects one detail: the chart showing limited forecast revisions concerns 2026. Jeremie’s methodological point survives the correction—the team triangulates chip, data center and energy signals rather than relying on a single industry forecast.
Reyk describes tracking “not even campus level, but building by building,” using permits, power allocations and satellite photos. Mapping more than 400 moratoriums against tracked locations produces an affected share of roughly 10%, he says. Grid constraints also require project-level assessment: some sites can use on-site generation. His claim about the relevant projects under construction is that “all of them pass this bar.”
2. Delays change delivery schedules without erasing entire campuses
Jordan’s pushback is about timing: even without cancellations, could enough projects slip by two quarters or a year to move capacity into 2028? Reyk says 12-month timelines are already aggressive. Faster promises, new developers, obscured permits or power sources, and no identified tenant all raise questions. These factors help the team avoid treating longer-duration projects as 12-month builds.
The uncertainty remains real. Reyk cites a transformer blowing up at a Core Scientific site as the kind of disruption that can surprise the team: “they can catch you off guard.” Jeremie says developers targeting delivery around mid-2027 may receive another one to three quarters in the model, depending on observed conditions.
Jeremie separates a site’s first delivery from its subsequent ramp to full capacity; a site under construction does not mean all its capacity will be counted for 2027. Oracle’s New Mexico project has over a gigawatt of IT capacity under construction, but substantial power challenges mean only a relatively small contribution enters the 2027 forecast. Jeremie says SemiAnalysis flagged those problems before broader coverage and assumes a much slower ramp.
For the discussed Crusoe project, Jeremie rejects claims that the forecast includes a whole gigawatt: “we have under 200 megawatts” in the latest version. Buildings containing 200–400 MW do not immediately become fully operational: “you don’t go from zero to 400.” Jordan’s objection to another blanket discount is that it risks “double counting almost.”
3. Contracts and supplier orders provide earlier checks on delivery
Jeremie describes a sequence of observable commitments. Pre-leasing often precedes financing and construction, although contracts can be signed at different project stages. Hyperscalers, neoclouds and AI labs sign what he calls “binding contracts that cannot be cancelled.” Rising quarterly self-build construction starts then provide a more physical indicator of how much larger 2027 could be than 2026.
Jordan points to more than $1 trillion in signed data center leasing. Jeremie says estimating that total requires building-level pricing, contract duration and escalators. Public crypto-mining companies such as Core Scientific reveal backlogs reaching tens of billions of dollars, while private operators—including QTS, EdgeConneX, Aligned and Vantage—require separate tracking to capture commitments public-market investors may otherwise miss.
Sterling supplies the concrete leading-indicator example: its groundbreaking and site-work orders arrive early, and Jeremie says the team anticipated their surge. Electrical and cooling equipment follow; accelerated deliveries can strengthen supplier revenue, while shipping problems can delay facilities. Energy equipment provides another timing check: gas turbines have multiyear lead times, while reciprocating engines offer shorter alternatives.
4. Modular construction changes the labor requirement and competitive timetable
Reyk contrasts traditional 18-month construction with DayOne timelines of around nine months for 100-MW sites, and cites Bridge Data Centres in Southeast Asia building in six months. Anthropic reference designs, Meta’s tents and Amazon’s new buildings illustrate wider adoption. He also recalls Project Houdini’s stated goal as, “I think,” 6 GW annually for the next two or three years.
The mechanism is physical: imagine a power hall resembling “a literal shipping container,” slotted into the data center shell and connected. That demands much less site labor than “20 electricians” coordinating a conventional installation. Reyk’s argument is that repeated modular deployment changes the labor intensity of the whole buildout, rather than merely accelerating an unchanged construction process.
Competition reinforces adoption. A developer offering 12–14 months loses value against one offering eight or nine; Reyk also thinks firms avoiding modular designs may struggle to obtain labor. Jeremie says newer facilities optimize construction, ramp and commissioning together. The team is preparing a fuller labor analysis, but Reyk’s current conclusion is that “modular is what’s making this work.”
5. The bottleneck could reverse as data center supply catches up
Jordan preserves a constraint that faster construction alone does not settle: operators can overdeploy GPUs and then power-cap them within an available electrical envelope. Jeremie agrees that GPUs are currently easier to obtain than data center megawatts. Buyers seeking immediate deployment can access chips but struggle to secure “land, power, shell” within three to six months.
Looking further out, Jeremie is slightly more worried about data center supply exceeding chip supply than the reverse, especially entering 2028. Jordan checks whether that was a formal oversupply forecast; Jeremie narrows it explicitly: “I didn’t say formally. I’d say it’s the main risk as we get into 2028.” The distinction matters to the strength of the call.
The resulting question is how quickly semiconductor supply can expand. Jeremie says that expansion would support new fabs and equipment investment. He interprets Jensen’s large memory and logic supply commitments as efforts to remove constraints, while acknowledging cyclicality: Jensen also secured substantial GPU supply for the 2021 crypto-mining boom, which subsequently crashed.
Jeremie’s explanation for the disagreement is temporal: “it’s easier to feel what’s happening right now” than anticipate conditions 12–15 months ahead. Jordan’s framing is that the obstacles are real but hard to measure accurately; Reyk says many lower forecasts may not be based on accurate information.
Full transcript
Reyk, show me that rundown. It looks like if you lie down, you’re going to kick your feet through the wall. Like, dude. If I have a night terror, my foot goes right through the wall.
Yeah. Take a look at the pole in the middle of the room.
Reyk’s dreaming about Brad Gerstner tonight, just fighting the wall and kicking it. There’s an inspector in there.
My room is fine. I just got two twin beds, which was kind of weird, but besides that, the room was okay.
That room is—look at that carpet. That room’s nice. They just said, “Yeah, you break the shitty one.”
I know. Talk to Michelle.
Yeah, I’ve got to talk to someone. Okay, you guys ready to rip into it? You guys ready to start?
Yeah.
Everybody get in bed for the cold open.
1. The 2027 Forecast Debate
Let me jump in. Hello everyone. Welcome back to another episode of Semi Analysis Weekly. In this episode, I’m joined by Jeremie and Reyk, our experts on the data center model. We’re going to talk about their article, “Data Centers 2027: The Year of Acceleration—50 GW in a Single Year.” This comes against the backdrop of Brad Gerstner from Altimeter showing our data on the All-In Podcast the other week, and then in his BG2 interview with Jensen. They were talking about how our forecast of 43 GW in 2027 is too aggressive.
We’re going to dig into why that’s not the case, and we’re actually increasing the forecast. When people think only about 25 GW are going to get built, we’re going to go through exactly what the FUD is about: moratoriums and other political restrictions, power generation, and all sorts of reasons why this might not be the case. But I don’t want to walk through it. Let’s talk through it with you guys. Welcome.
Man, thank you.
Thanks for having us.
Yeah, we’ve got a lot to talk about today. It was a good one.
2. Model Track Record
Before we get into the actual content of the article, let’s lay the land. Can you give me a little history of your track record with the data center model? Jeremie, can you tell me when you launched it, some of the forecasts you’ve made in the past, and how those have played out in terms of the model’s accuracy?
Sure. The model launched nearly 3 years ago now. We published an article in March 2024 to discuss some of the results we got from building the first iteration of it. That article is still available today. It was called “The AI Energy Dilemma.”
We start the article by showing other market forecasts and then putting our own alongside them. Our forecast was materially ahead of many other firms, like McKinsey or whoever else was making forecasts at the time. That has largely, of course, been proven right. We’ve increased the forecast a bit since then, but overall, the orders of magnitude were right. We were expecting a material acceleration in 2025 and then in 2026.
The way we do this is by combining several data points. We use data from the chip side, the data center side, and the energy side. We triangulate as much as we can through everything we can follow. Another thing I could point you to is our article from a few months ago called “Stop Saying Half of U.S. Data Centers Are Delayed,” where we showed this chart of our 2027 number and how it has been revised across different versions of our data center model. The reality is that it has barely moved.
Sorry, what was it—the 2026 one, though?
Yeah, whatever, 2026 or 2027. People can go, “Okay, same thing.”
It’s all right, Jeremie. It doesn’t matter which year it is.
Yeah, you can pick any year. It doesn’t matter.
All right, it’s 2026. Sure.
Yeah, exactly. It’s been extremely close. We’ve been doing this for nearly 3 years now, so the track record is there. The methodology is actually not that complicated. You just have to be thorough and comprehensive and make sure you’ve got everything. We’ve spent a lot of time and built a whole team to ensure that.
That’s what got us here. We have extremely strong signals pointing to 2027 being a gigantic year, with many different subcomponents that we can talk about in depth. So there we are, after Brad called us out publicly and cut our forecast in half. This isn’t just some number we made up; there’s a lot of work behind it, and we stand by it.
3. Where the Pushback Comes From
Let’s walk through where some of this opposition comes from. Reyk, when you hear people take our number and cut it in half based on a hunch, what’s the hunch actually coming from? Is it purely moratoriums? Is it the supply chain? Is it capital? Labor?
It’s a number of things. When you read a lot of the news articles or sell-side reports, they tend to hinge on the same fear-inducing topic: moratoriums are going to get rid of all the data center build-out in the U.S. It’s all over; they’re canceled. Then you have other reports that don’t cover every data center in the U.S. but take headline numbers and say, “Well, this headline number was there before, but now this got canceled because of a moratorium.”
Then there’s the energy side of things, where people say there are no grid connections anymore, so these data centers are never going to get power. There are a number of concerns: labor, the actual build-out, and many other things. The issue is that nobody digs as granularly into it as we do. When we say granular, we mean site by site, building by building—not even just at the campus level. We look at permits, power allocations, and satellite photos. We estimate when construction starts, see how each project is progressing, and work out when it should come online.
It’s hard to refute this when I’m physically looking at all of these data centers under construction. I don’t know how I could haircut the numbers; these are the numbers that add up when you count everything. We also address the same concerns people have. With moratoriums, we count and map them all. We looked at over 400 moratoriums, combined them with every data center we had mapped into the relevant region, city, or jurisdiction, and found that only about 10% of the projects were actually affected.
It’s the same story with energy. You look at grid headroom, and then at behind-the-meter solutions, where maybe you don’t need grid headroom in the first place. You can just build an on-site generation project. There are a number of ways around this. These topics are broadly true, but they don’t apply to every project. When you dig into the relevant projects—the ones under construction—they all pass this bar. That’s why our methodology gives us these numbers and why we have such confidence in them.
4. Delays and Haircuts
Yeah. Take me through the individual projects. One of the great charts in the article was “From Dirt to Data Halls,” where you track 3 individual projects from when they broke ground in 2021 or 2022 to when they’re fully built out and the data halls are available.
These projects are targeted for 2027. Maybe moratoriums cover cancellations and, to a smaller extent, delays. Can you talk about the chance that half of these construction projects slip by a year—two quarters—enough to push them into 2028? What’s the risk that small delays result in a change from 2027 to 2028?
Looking at all the data centers we’ve seen before, we have thousands in the model, and we’ve reviewed them over and over. You get a sense of what looks like a delay, or what might constitute one. Maybe there’s a winter storm in the northeastern U.S. during the TeraWulf or Nebius build-out. Maybe there’s an equipment problem, like the transformer blowing up at the Core Scientific site.
There are a number of things you can see or gauge from public reports, as well as overly aggressive timelines. You start to understand what the build-out actually looks like from a timeline perspective. In our view, 12-month timelines are pretty aggressive. These are quick timelines, especially compared to the historical industry. When anybody is guiding to less than a 12-month timeline, it already raises a red flag.
If it’s a 12-month timeline from a developer that just came out of nowhere, and some of the permits are obscured, the power source is obscured, or we can’t identify it perfectly, maybe there’s nobody signing on as the tenant for the project.
All of it leads to enough of a question that I don’t totally know how certain we can be, right? It’s not to say these things don’t catch us off guard; they can. I didn’t predict a winter storm. It’s not my job. But the point is, we can get close enough to have a filter that keeps these longer-duration projects from becoming 12-month projects in our methodology.
Yeah, you’re already considering the potential for this.
Yeah.
Just to build on that super quick, this is also why it’s somewhat counterintuitive to apply a haircut to our numbers. Admittedly, we’re being a bit conservative on some things. There are sites under construction right now that won’t add all their megawatts in 2027, or maybe won’t add megawatts in 2027 at all, in some international cases. We’re not being overly aggressive either. We’re taking all of these factors into account and using a reasonable timeline.
Adding a haircut on top of that is almost double-counting, or doing the same work over again when we’ve already applied it. We like to dig into our numbers pretty thoroughly.
Yeah. I mean, look, the other thing is we’re obviously on the ground quite a lot. We travel the world, go to conferences, and talk to the people actually building these facilities, so we get insights from the folks doing the work. For the most part, they’ll tell you these days that it’s very challenging to build these things. They’re facing a lot of community pushback, struggling to get labor, and so on.
On the other hand, the timelines they’re targeting are getting shorter and shorter. In many cases, folks tell us, “Hey, I’m expecting to have this ready by mid-2027.” In our number, we purposely add delays of 1 to 3 quarters, depending on what we’re observing, because we expect things to be delayed by default, given all these challenges.
5. The Top 40 Sites
Yeah. Okay. My favorite chart from the article went site by site across 40 sites. It’s really impressive. It’s clear from this that these are massive sites with a ton of buildings. They’re all under construction and all count toward 2027, even if, as you’re saying, they may not add all their megawatts in 2027. They are contributing to the total forecast.
A couple of people have commented on a few of these sites. First, give me your reaction to the size and scope of the forecast, and how even these 40 sites aren’t the full complement of data centers under construction worldwide right now. They’re still less than 50% of the data centers globally that are shown on screen here. But there are individual sites that some people know better than others, and some may have high-profile delays. Can you talk through the difference between an outright cancellation and a delay, and why things are still going to come in in 2027—not around 25 gigawatts, but closer to 50, as you guys are saying?
Yeah, look, there are two things. One is when the first phase of a site is going to be delivered; the other is how fast the site will ramp to full capacity. For some of the projects we show here in the picture, we assume much more conservative ramps than for others. For example, Oracle New Mexico: we were the first in the market to talk about all the issues they’re facing on that site, and that’s been proven right. The media are all over it now, and they’re calling force majeure and whatnot.
They’re facing tremendous challenges getting power to that facility, which is why we assume a much slower ramp. If you look at the picture, you see the whole campus is under construction, with over a gigawatt of IT capacity under construction right now. But our forecast is more conservative because of the challenges we’ve identified. So the share of megawatts that goes into our 2027 number for this specific campus is actually fairly low.
Some Twitter accounts commented on our ramp for Crusoe in Childress, saying we were assuming a whole gigawatt was included in our forecast. That’s wrong. We have under 200 megawatts in the latest version we published. People are just making things up; that’s not accurate. We assume these things ramp over time, because you’re not going to go from zero to 300 megawatts. Some of these buildings are 300, 200, or 400 megawatts. You don’t go from zero to 400; things get ramped over time. If you’ve ever been to a data center—which we visit a lot—you know how that stuff works.
6. Leading Indicators
Yeah, makes sense. Talk to me a little bit about the leading indicators, because this isn’t a forecast you just created two weeks ago. It’s been pretty consistent for a number of months, if not years, given that the model is years old and you’ve been tracking 2027 as it comes to fruition.
If we’re forecasting for 2032 right now, that depends on leasing and pre-leasing activity happening now, and we expect it to increase, not decrease, because there’s only so much you can track at that early stage. But right now, if something’s going to be built in 2027, there’s construction activity happening. Talk me through how you started a forecast for 2027 a couple of years ago, and how you’ve tracked it through to how it’s playing out right now.
Yeah, and I want to say that’s also why we have different models to track the whole ecosystem. Around data centers, we have the energy model and the industrials model. Plenty of suppliers are exposed to projects at different phases, and we try to track every one of these companies. We’ve been extremely successful at predicting the strength in orders for some of them.
For example, we published a big deep dive on our industrials model. There’s a company called Sterling Infrastructure. They do early-stage construction: groundbreaking and site work. That happens first; it’s the first thing that happens when one of these projects starts getting built. We can track this company at the project level, essentially. We’ve been really good at predicting that their orders were going to surge, which is exactly what happened.
There are many other companies down the supply chain. We track electrical equipment and cooling equipment very closely; that happens afterward. It can affect companies in both directions. Sometimes they’re delivering ahead of schedule, and these companies see strength in their revenue during the quarter. We’ve also talked to clients of our industrials model about potential supply-chain delays, meaning some of these sites are late. We’ve flagged several sites in our model that have been delayed by equipment-shipping and supply-chain challenges. That’s another thing we can track.
Energy also has very long lead times. Everyone knows gas turbines are multiple years out. There are other solutions for power generation, notably reciprocating engines, that have shorter lead times. That’s another indicator you can track to understand the timelines of these projects. There’s essentially a waterfall of things you can track that are more or less binding, and we look at all of that to ensure our timelines are accurate.
The two indicators we presented most strongly in our article are self-build construction starts and pre-leasing. For self-build construction starts, it’s pretty trivial: you track every single building and when construction starts. You can see how the number goes up quarter after quarter. If you understand the timeline for building these things, that’s a strong indicator of how large 2027 is going to be relative to 2026.
The other indicator is pre-leasing, which means hyperscalers, neoclouds, and AI labs signing lease contracts with data center operators. This one is even earlier, because typically you sign a lease and then raise the debt. The timing can vary, too; leases can be signed at different moments in the life cycle of a data center project. We track that extremely closely, and every quarter we forecast these things before they’re published by others. We’re generally extremely accurate, which helps us understand ahead of time what’s going to get built under these binding contracts that cannot be canceled.
Yeah, and this is just a really big number, Jeremie. Just double-click on the over $1 trillion of data center leasing that’s been signed, which drives a lot of the forecast from the Credit Capital Markets model that Dan and the guys are working on. We just talked about it a week ago, and we’re going to be doing more of this, but this has been a pretty big story in the news recently: just how big the scale of some of this is.
Yeah. Look, there are a bunch of indicators you have to track at the building level to get to that number: pricing, contract term, escalators, that kind of thing. Some of these deals are public. If you look at the crypto miners, for example, companies like Core Scientific, you can see how quickly their backlog has been increasing. It’s now in the tens of billions of dollars.
But plenty of these companies are private, right? The public markets don’t have a good sense of all the contracts being signed between companies like QTS, EdgeConneX, Aligned, Vantage, and others, and all these leading hyperscalers or AI labs. That’s stuff we track extremely closely. By tracking every one of these private companies, we can gather the data and put out numbers that other folks may have been missing, but that I think are a crystal-clear indication of the strength of the buildout we’re going to see.
7. Modular Builds
Awesome. Okay, Reyk, tell me about logistics, man. When we talk about modular data centers, we wrote a whole article on Lego data centers and how these things go up. You’re the guy who stares at a lot of satellite images and watches these things. The trend seems to be that the changes in the satellite images are happening faster, not slower, right?
One of the charts shows Google’s self-build and how they’ve changed their design on the electrical side to get the modular sites out faster. Can you talk me through why this is going faster, not slower, over time?
Yeah, I think this is one part of the buildout that’s a bit overlooked, to be honest. If these were traditional data centers, maybe labor would be called into question, and maybe timelines would be called into question. But in our minds, the larger-scale utilization of modular deployment is the reason this works and the reason this happens.
When you look at traditional timelines of 18 months compared with, say, DayOne timelines, where they can get down to around 9 months for any number of 100-megawatt sites, you can also look at Bridge Data Centres down there in Southeast Asia, which can build in 6 months. You can look at Anthropic’s reference designs, which are also going to employ modular. Modular is becoming more widespread across the industry, so you’re going to see build times drop.
You’re going to look at the Meta tents and the new Amazon buildings, and see these buildings basically pop up out of nowhere. This is the Project Houdini program. I think the stated goal was 6 gigawatts a year for the next 2 or 3 years, maybe 6 months ago. These are the reason this buildout works.
We did the math in one part of the article: if you have your MEP—the more electrical side of the fit-out within the data hall—more modularized, you can imagine taking the power hall, like a literal shipping container, slotting it into the shell of the data center, and plugging in a few things. That demands much less labor than having 20 electricians in there trying to coordinate everything and, quote-unquote, stick-build the entire hall. You can imagine that on a much grander scale.
We’re working with the industrials team to write a comprehensive letter and newsletter on the labor for the 2027 buildout. So far, everything points to modular being what’s making this work, and you can see it across the industry.
Yeah, I would add that if you actually visit data centers, both the big ones and the two- or three-year-old ones, as well as some of the smaller ones, you can clearly see the difference. Many of these new builds are fully designed and optimized to go as fast as possible on build time, ramp time, commissioning time—absolutely everything has been thought through for an accelerated timeline.
Many folks tell us, “I’m going to pull this off in 6 months.” We purposefully haircut a lot of these assumptions, but it’s still extremely impressive when you visit these sites and see what people are doing to accelerate while keeping the bar high on quality and certainty of delivery.
Yeah. And I think the whole point here is that somebody might ask, “How do you measure modular adoption in this case?” Well, we have it on the industrials tracker. If you don’t follow suit in the modular buildout, or this kind of design strategy, you probably won’t be allocated the labor you need, nor will you be competitive in the buildout market.
If I can offer you a data center in 12 to 14 months, I’m not worth as much as the guy who can offer it in 8 to 9 months. It’s going to start forcing people’s hands in this direction. We’re already watching people do interesting things. BrightRay is talking to multiple neoclouds right now about selling entire data center modules—not the 1- to 2-megawatt crates, but the actual hall that you can ship and install into your core and shell, or with the core and shell from BrightRay.
People are going to keep coming to these folks who can provide this equipment, and they’re going to keep buying it. I think this buildout is exceedingly led by the modular aspect.
8. Chips vs Datacenters
Yeah, makes sense. How about the balance? We put this behind the paywall, to be clear, but it seems like generally speaking, people are trying to call this shot: whether you’re going to have accelerators and an overbuild of data centers, whether you’re going to be constrained on data centers and have an oversupply of accelerators, or whether it’s going to be a perfect match.
There’s maybe one comment to make there: we know how many people overdeployed GPUs and then used power capping. It seems like the power envelope is still a constraint for a lot of these, whether they’re grid-connected or behind the meter. There’s only so much power you can bring on, even if you have the physical space and the chips available, so you’ll be power-capping GPUs to fit within the envelope.
Given how much construction timelines are reducing at modular sites, maybe you guys can say at a high level whether you think there are going to be a lot more data centers built in the next couple of years.
I mean, look, it’s obviously tough. I would say, generally speaking, our forecast for data centers is slightly easier, in the sense that it’s based on infrastructure signals you can track ahead of time. Chips can be slightly more volatile. But overall, we also have a lot of very strong indicators pointing to what’s going to happen in 2026 and 2027.
I would say, generally speaking, we’re a bit more worried about the risk of oversupply of data centers relative to chips than the other way around. We’re already sort of oversupplied on chips today, as we speak. One always has to be a slightly bigger constraint than the other, and if you want to deploy something right now, it’s easier to get GPUs than data center megawatts.
9. Responding to Brad Gerstner
But when we compare how these different supply chains are expanding into next year, and especially into 2028, we do see a risk of the situation flipping, with potential data center oversupply in 2028. That’s what we’ll watch most closely: how fast the chip supply chain can expand. That’s obviously a strong positive for things like capex, investing in new semiconductor fabs, and buying new equipment. That’s the type of stuff we’ve been flagging to all clients for several months already.
So let’s wrap on this one, guys. Great job taking us through all of this. Super helpful. We’ve obviously covered all sorts of aspects, from the moratoriums to the labor, the electrical equipment, the cooling equipment, and the actual construction progress. There are all sorts of things that play into a data center arriving on time, and we think we cover them in great detail. I know you guys do.
Let me get your direct reaction to people like Brad Gerstner showing our chart with an old forecast to Jensen, basically saying, “You’re going to build too many chips and not have enough places to put them.” Jensen kind of just took it, acknowledged it, and moved on in the conversation. Where do you think the sentiment comes from? Maybe you can respond directly to that.
I mean, I think it’s—
I mean, I think—
I think it’s easier to feel what’s happening right now than to know how that feeling is going to evolve by the end of next year, 12 to 15 months from now. Right now, it’s crystal clear that there’s more supply of chips than supply of data centers.
Again, you talk to anyone who wants to deploy AI infrastructure, and they can get access to GPUs, but they cannot get access to land, power, or shell space in the next 3 to 6 months. I think the constraint right now is crystal clear, so obviously we want to make sure the situation is solved. That’s why Jensen absolutely wants to make sure this doesn’t happen.
Jensen has a long history of being extremely smart about supply chains, and he can be aggressive as well. We’ve seen him secure a tremendous amount of supply, like in 2021, to secure GPUs for the crypto-mining boom, which obviously crashed. There can be a bit of a cyclical aspect, but overall it makes sense. It perhaps makes the company slightly cyclical, but the whole semiconductor sector is cyclical anyway, so it doesn’t change too much. His company is in extremely good financial health, and I think he’s writing these massive checks to secure supply from memory suppliers and logic suppliers. He’s really good at securing supply from these guys, and he wants to make sure he can build as much as he can with no constraints. So, yeah, it makes sense to pressure everyone in the supply chain to build more.
I think the other thing I’d add on this, too, is that a lot of it comes from people operating on not the most solid foundation of information. It’s one thing to understand that moratoriums are slowing data center buildout, grid connections are, too, labor is, too, and equipment supply chains are, too. All of that’s true, but being able to measure it accurately is a different story. As an aside, Jeremie, did you say that we were looking at oversupply in 2028 before I continue?
I didn’t say that formally. I’d say it’s the main risk as we get into 2028.
I was going to start talking about that, but I don’t know.
Yeah, let’s just move on from that.
But, yeah, no. To finish my thought, I think the other thing here is that a lot of people are operating on a not very solid foundation of information when they come to these conclusions.
Yeah, I mean, look, all of Wall Street tells us, “You guys are plainly wrong.” Some of the great investors out there, like Brad—he’s obviously not alone—have said it on a big podcast, but many other folks tell us, “You guys are way too high.” There are a lot of data sources out there with data center forecasts that are materially smaller than what we project. We’re familiar with a bunch of them, including cloud capex forecasts that track data centers. There are 3, 4, or 5 of them with numbers that are 2x more than ours. Some of our clients tell us, “Hey, these guys are lower,” and we tell them, “We didn’t just prompt Claude and write this. There’s more that goes into it.” Generally speaking, a lot of sources probably aren’t based on accurate information. And obviously every CEO is telling folks that the biggest constraint right now is power. It goes back to the sentiment driven by what’s happening right now.
Yeah. And maybe if I could just say something directly: what set me off on this one is that a month ago, the guy got on stage and presented to a YouTube audience that’s now 330,000 people, followed by this one from 4 days ago, with 265,000 people and counting. He cites our data and then just says it’s wrong.
He says, “Yeah, I don’t think so,” and says other stuff. That’s different. Sure, cite our data and provide specific critiques. Great—we can always learn from that. But offhandedly saying, “This is the forecast; we think it’s wrong,” seems a little strange. I’m glad you guys could respond directly to some of that in this article, and I look forward to walking through 2027 and continuing to see if the track record you guys have on the data center model holds true.
Yeah. I mean, just to be clear, the goal isn’t to respond to one person. It’s a huge topic that people ask us about all the time. Our institutional clients have been aware of how we do things and what we’ve been projecting, and I think we just wanted to give a piece of that to our broader audience, put the content out there, and see how folks react. But, yeah, again, we’re extremely bullish on how big the infrastructure buildout will be in 2027, and we think we have rock-solid indicators of that.
Yeah. And it’s not the first time we’ve gotten the feedback that our numbers are insane, that people don’t know what to do with them, and that nobody agrees with them. So we wanted to directly address that whole thing, honestly.
Awesome, guys. Okay, well, great job today. Appreciate you taking the time to educate everybody, as always, on what you spend all your days doing and how accurate it is. Appreciate it.
Cheers, Jordan. All right, we’re done now. Can you smile for the camera quickly? Smile for the algorithm. Come on, with some teeth, Reyk, buddy.
Can you what, sir?
Smile for the algorithm. Smile. Look at the camera and smile.
On the fucking thumbnail. Okay, good job.