0:00
kidding.
Search this video
0:00
Right, show me that bed again.
0:00
It looks like if you lie down you're gonna kick your feet through the wall, you six foot three little mongoloid dude. Like dude.
0:04
Yeah, like if if I like if I have a n You there's no way you're fitting a nut.
0:08
It if I have a night tear my foot goes right through the fucking wall. Yeah, Ahem. Looks good.
0:15
Take a look at the take a look at the pole in the middle of the room. Right right there, man. Hell yeah.
0:23
Reyk's gonna be Reyk's gonna be dreaming about Brad Gerstner tonight, just like fighting the wall and kicking the wall and Yeah, put his picture in there.
0:31
My room is fine, it's just that I got two single beds, which was kind of weird, but besides that, Like that what was it again?
0:34
Room is, look at that carpet.
0:38
That you know that room's nice, you yeah.
0:40
They just said, Yeah, give break the bucket shitty one.
0:44
I'm just I know, talk to Michelle.
0:51
Yeah, I gotta talk to someone. Okay.
0:53
You guys ready to rip into it? You guys ready to start? Yeah. You Ha ha ha.
0:57
Everybody get in bed for the cold open. Let me jump in. Everyone.
1:08
Welcome back to another episode of SemiAnalysis Weekly.
1:11
In this episode, I'm joined by Jeremie and Reyk, our experts from the data center model.
1:15
We're gonna talk about their article, Data Centers 2027, the year of acceleration, fifty gigawatts in a single year.
1:22
This one comes from a bit of a backdrop.
1:25
We just the other week, Brad Gerstner from Altimeter was showing our data on the All-In podcast and then his BG2 interview with Jensen and they were talking about how our forecast of forty-three gigawatts in twenty-seven is too aggressive.
1:37
Well, we're gonna dig into why that's not the case.
1:40
This in this and we're actually increasing the forecast.
1:43
So when people think that only about twenty-five gigawatts are gonna get built, we're gonna go through exactly, you know, what the FUD is about.
1:50
They talk about moratoriums and other political restrictions, they talk about power generation, they talk, you know, about all sorts of you know reasons why this might not be the case.
2:00
But guys, I don't want to walk through it.
2:02
Let's talk through it with you guys. Welcome.
2:06
Nah, thank you, thank you. Thanks for having us. Yeah.
2:07
We got a lot to lot to talk about today. This is a good one.
2:12
Okay, so before we get into the actual content of the article, how about we do lay the land and you guys give me a little bit of history on your track record with the data center model?
2:20
Jeremie, can you tell me a little bit about the history of this model, when you launched it, and just talk about some of the forecasts that you've made in the past and how that's played out in terms of like fidelity of the model?
2:32
Sure, so the model launched nearly three years ago now.
2:35
We published an article actually in March of 2024.
2:40
To discuss publicly some of these results that we got from building the first iteration of that.
2:44
So that article is available today.
2:49
It's AI Datacenter Energy Dilemma, Race for AI Datacenter Space, whatever.
2:54
And so we started this article by sort of showing other market forecasts out there that we're putting around, which was materially ahead of many other firms like McKinsey or whoever else was making forecasts these days.
3:10
But that has largely, of course, been proven right.
3:13
We've increased the forecast a bit since then.
3:17
But overall, the orders of magnitude are there.
3:20
We were expecting, back in early 2024, again, a material acceleration in 2025, and then in 2026.
3:25
The way we do this is always by combining several data points.
3:30
So we use data from the chip side.
3:32
We use data from the data center side, from the energy side.
3:34
We always triangulate as much as we can through everything we can follow.
3:40
Another thing I could point you to is our article, a few months ago, it will stop saying half of US data centers are all delayed, where we show this chart of our 2027 number and how it has been revised over the different versions of our data center model.
3:57
And really, it has barely moved.
3:59
So we're always expecting.
3:59
Sorry, was it sorry, what it was the twenty twenty six one though.
4:06
Yeah, whatever, 26, 27, people can go look at it, same thing. It's all right, Jeremie.
4:12
It doesn't matter which year it is. Yeah, yeah.
4:13
You can pick any year, but it doesn't matter. Alright, 2026, sure. Yeah, exactly.
4:19
So Let's resume in the try and see how it's been pretty extremely close.
4:23
Again, like we've been doing this for nearly three years now.
4:27
So the track record has been there now.
4:30
The methodology is actually not that complicated.
4:33
You just got to be thorough and comprehensive and make sure you got everything.
4:39
And so we obviously spent a lot of time and built a whole team to ensure that.
4:44
But that's what got us there is now sort of we have extremely strong submittals pointing to 27 being a gigantic here, many different subcomponents that we can talk about in depth.
4:54
so there we are after Dear Brad called us out publicly and cut our forecast in half. Didn't know.
5:04
This is actually at least some number we made up.
5:09
There is a lot of thought and a lot of work that went into it.
5:12
And we are firmly behind it and stand by it.
5:17
So let let's try and walk through where some of this opposition comes from, Reyk.
5:20
Like when you hear people who take our number and then cut it in half based on a hunch, what's the hunch actually coming from?
5:28
Like is this purely moratoriums?
5:30
Is this the supply chain? Is it capital? Like labor? Yeah. Yeah.
5:32
I mean, it's a number of things, right?
5:36
Like I think when you read a lot of the news articles or you read a lot of the like kind of sell side reports or anything like this, they all kind of hinge around the same, I don't know, like FUD, I guess, or like fear inducing topic where it's like The moratoriums are going to, you get rid of all the data center build out in the US, like it's all over, they're canceled, right?
5:55
And then you have like other reports coming out where it's like people not totally covering every data center in the US, but rather taking headline numbers and saying, well, this headline number was there before, but now this got canceled from a moratorium.
6:09
And I mean it's like that.
6:09
And then the energy side of things where there's no grid connections anymore.
6:14
Sure, like, you know, these data centers are never going to get power.
6:19
There's there's a number of things, labor, you know, like the actual build-out itself, millions of things here.
6:23
But the kind of issue is that nobody digs as granularly into it as we do, right?
6:30
Like Jeremie brought up that it's granular, but the what that literally means is we go site by site, building by building, right?
6:36
Like not even campus level, but Building by building, permit by permit, power allocation by power allocation, buying satellite photos.
6:44
We go through estimate starts of construction, see when the whole pro how the project is progressing, when it should come online.
6:51
It's hard to refute any of this. Right.
6:55
When when I'm physically looking at all of these data centers under construction, I don't know how I could haircut it. Right.
7:03
These are the numbers that are adding up when you count all of this.
7:07
And so we start, we have that.
7:09
And then we also address these same problems that people are worried about. Right.
7:13
With the moratoriums, we said, okay, then let's, you know, also count all of them and map all of them.
7:19
And we went to like, you know, over 400 moratoriums, combined it with every data center that we had.
7:23
Or like, you know, mapped into the region or the city, whatever jurisdiction may apply, and found, you know, like 10% of it was actually being affected out of all 400.
7:32
Energy, same story, you know, grid headroom, and then you lead into like behind the meter solutions where you know, maybe you don't have to have grid headroom in the first place.
7:41
I can just build a on-site generation project, right?
7:44
Number of ways around this.
7:44
Point being is that these topics are broadly true, but they don't apply to every project.
7:52
And when you dig into the like appropriate projects, the ones that are under construction, they all pass this bar, right?
7:59
All of them pass the bar.
7:59
And that's kind of why our methodology can give us these numbers and why we have such confidence in them.
8:06
Yeah, and and take me through the individual projects.
8:09
Like one of the great charts in the article was just this from dirt to data halls one where you're tracking three individual projects from, you know, when they broke ground back in 2021 or 2022 to when they're fully built out and the data halls are available.
8:25
Clearly, these projects that we're talking about that are targeted for 2027, maybe moratoriums cover cancellations and to a smaller extent delays.
8:35
Can you talk to me about the chance that half of these construction projects slip by a year, two quarters enough to push them into twenty twenty eight or something?
8:46
What's the risk of small delays resulting in a change from twenty seven to twenty eight or something like that?
8:53
Looking at all the data centers that we have before, right?
8:56
Like we have thousands in the model.
8:58
We've reviewed the thousands over and over.
9:00
You get a sense of what looks like a delay, right?
9:03
Or what might constitute a delay, or it's gonna be, you know Maybe there's a winter storm, right?
9:08
Like in northeastern US during like the TeraWulf build-out or like Nebius's build-out.
9:11
Maybe there's, I don't know, like a equipment problem, right?
9:18
Like the transformer blowing up at the Core Scientific site. Right.
9:20
There's a number of things that you can kind of see, or you can kind of gauge from like public reports, as well as, you know, overly aggressive timelines, where You start to understand what the build out actually looks like on a timeline perspective, right?
9:34
And so it's in our view, you have timelines at 12 months, are like pretty aggressive, right?
9:40
Like these are quick timelines, especially when you compare it to the historical industry.
9:44
These are quick timelines.
9:44
And so when anybody's guiding like sooner than a 12-month timeline, it already rings a red flag.
9:50
And then if it's a 12-month timeline, you know, from a developer that just came out of nowhere, and like some of the permits are kind of like obfuscated, power source is obfuscated, maybe we can't identify it perfectly.
10:02
Maybe there's nobody signing or being the tenant of the project.
10:05
All of it kind of leads up into enough of a question where it's like, I don't totally know how certain we can be on this timeline. Right. So it's not to say that.
10:16
These things catch like don't catch us off guard.
10:19
They can catch you off guard, right?
10:20
Like I I didn't predict a winter storm, right? Like it's not my job.
10:23
but but the point is is we can get close enough to where it's like, well, you know, we have enough of a filter to not let these longer, these longer duration projects become 12 month projects in our methodology.
10:34
Yeah, you're already considering the the potential for this. So Yeah.
10:41
And just to build on that super quick, this is also why it's somewhat counterintuitive to apply a haircut to our numbers.
10:51
We're admittedly we're being a bit conservative on some things, right?
10:55
Like there are sites under construction right now and they don't add all their megawatts in 27, right?
11:00
Or maybe they don't add megawatts in 27 at all in the case of some international instances, right?
11:05
We're not being overly aggressive either, right?
11:08
We're taking all of these factors into account and taking a reasonable timeline.
11:12
And so adding a haircut on top of that is doing a bit of like double counting almost, or you know, doing the same work over again where we've already applied it.
11:22
And so you end up with a I guess not our number, but we we like to think our numbers pretty good.
11:28
I mean, look, the other thing is we're obviously quite a lot on the ground.
11:33
We travel the world going to conferences, talking to all of the folks actually building these facilities.
11:39
And so we also get sort of the on the ground insights from folks that are actually doing it.
11:43
They will tell you for the most part these days that it's very challenging to build these things, that they're facing a lot of community pushback, that they're struggling to get labor, and so on and so forth.
11:53
But on the other hand, the timelines that they're targeting are getting shorter and shorter, right?
11:58
And so in many cases, folks tell us, hey, I'm expecting to have this ready by mid-2027.
12:05
And so in our number, we purposely add delays of like one to three quarters, depending on whatever we're observing, just because we expect stuff essentially to be delayed by default because of all of these challenges. Yeah. And yeah.
12:18
Okay, so so my favorite chart from the article actually went site by site across forty sites. It's really impressive.
12:26
it's kind of clear from this directly that these are massive sites with a ton of buildings.
12:33
They're all under construction, they're all counting towards the twenty-seven, maybe not all of their megawatts as you're saying in twenty-seven, but they are contributing towards the total forecast.
12:44
A couple of people have commented on a few of these sites.
12:47
Maybe first just give me a reaction to the size and scope that you're referring to of all of the forecasts and how the even just these 40 is still not even the full complement of data centers that are under construction worldwide right now.
13:03
It's it's still under 50% of the data centers that globally that are shown on screen here.
13:06
But also that there are, you know.
13:11
Individual sites in here that some people know better than others that maybe have some high profile delays and you can talk through the difference between an outright cancellation and a delay and why things are still gonna come in in twenty twenty seven and not be around twenty five gigawatts and closer to fifty, as you guys are saying.
13:32
Yeah, look, there's two things.
13:34
One is, when is the first phase of the site going to be delivered?
13:38
How fast will the site ramp to full capacity?
13:41
Some of the projects that we show here in the Spectre, we have so much more conservative ramps than others.
13:47
Example, Oracle New Mexico.
13:47
We were the first in the market to talk about all the issues that they're facing on that site.
13:53
And obviously, that has been proven right.
13:55
All the media is all over it now.
13:57
And they are calling force majeure and whatnot.
13:59
So they're obviously facing a tremendous challenge to get power to the facility.
14:02
And that's why we assume a ramp that is much slower.
14:09
So if you look at a picture, you see that essentially the whole campus is under construction over a gigawatt of IT capacity under construction right now.
14:17
But our forecast is more conservative than that because of the challenges we've identified.
14:24
So the share of megawatts that goes into our 27 number for this specific campus is actually fairly low.
14:28
Another one, some Twitter.
14:32
Accounts commented on our ramp on Crusoe in Childress.
14:37
Folks saying that we were assuming there's a whole gigawatt that is going to be, that is included in our forecast and what we spent is wrong.
14:43
We have under 200 megawatts as of the latest version published.
14:48
And so, yeah, don't know, folks just making up stuff.
14:52
It's not actually accurate.
14:52
We assume these things ramp over time because you're not going to go from zero to 300 megawatts.
15:00
Some of these buildings are 300 megawatts, 200.
15:01
400, you don't go from zero to 400 stuff gets ramped over time.
15:07
If you've ever been to data center, you know that, which we do a lot.
15:12
So we actually know how that stuff works. Yeah. Makes sense.
15:14
Okay, talk to me a little bit about the leading indicators, right?
15:17
Because this is actually not like a forecast that you guys just created two weeks ago.
15:21
I mean, this has been pretty consistent for a number of months, if not years, at this point, considering that the model is years old and you guys have been tracking twenty seven as it comes into fruition.
15:31
Like if we're if we're forecasting for twenty thirty two right now, this depends on leasing and pre-leasing activity that comes right now and we expect it to increase, not decrease because like there's only so much you can track when it's that early stage.
15:43
But right now, like if something's gonna be built in twenty seven, there is construction activity happening.
15:48
So talk me through like how you started a forecast for twenty seven maybe a couple of years ago and then how it how you track it through to how it plays out right now.
15:55
Yeah, and I want to say that that's also the reason why we have different models to track the whole ecosystem.
16:01
So for example, around the data set topic, we have the energy model, we have the industrials model.
16:07
There's plenty of suppliers that are exposed to projects at different phases of that project.
16:11
And we sort of track every single one of these companies and have been extremely successful in breaking the strength and order for some of these guys.
16:19
Like one example, we published a big, dive on our industrials model of this company called Sterling.
16:24
Sterling, do early stage construction, They did the ground breaking, they did the site work.
16:28
So obviously that happens first.
16:30
It's the first thing that happens when you see these, when one of these projects starts getting built is you do some of the early site work.
16:36
So this company, can track again, like at the project level, essentially.
16:41
And so we've been pretty, we're actually good at predicting that their orders were gonna surge, which they ended, which is exactly what happened.
16:49
Then there's many other companies down the supply chain.
16:53
So for example, all the equipment, electrical equipment, and cooling equipment, we also track that very closely. That happens afterwards.
16:58
And that can also impact the company both ways.
17:05
Sometimes they're delivering ahead of schedule, and so these guys see strength, and you see their revenue is strong in a cup and a quarter.
17:11
We've also had times where, to our industrial model clients, we talked about potential supply chain delays, meaning that some of these sites are late.
17:20
And we've also flagged several sites in our model that have been delayed by ship equipment.
17:25
Equipment shipping challenges and supply chain challenges.
17:27
And so that's another thing we can track.
17:29
Energy also, of course, very long lead times.
17:32
Everyone knows the gas turbines are multiple years out.
17:35
Now there's a bunch of other solutions for reciprocating engines, notably, that have shorter lead times.
17:40
And that's another indicator we can track to understand the timeline of these projects.
17:43
So there's essentially this waterfall of things that we can track that are more or less binding.
17:49
And we look at all of that to ensure our timelines are accurate.
17:53
Now would say the two ones we we presented most strongly in our article.
17:58
Self-built construction starts pretty trivial.
18:01
You just gotta track every single building and when they're construction of the building.
18:04
And you can see if you, I don't know if you've showed that chart before, but how the number goes up quarter after quarter.
18:11
If you understand the timeline at which this stuff is getting built, is a strong indicator as to how large 27 is going to be relative to 2026.
18:19
The other one is pre-leasing, which means hyperscalers, neoclouds, AI labs, signing lease contracts with data center operators.
18:29
This one is actually even earlier, because typically you sign a lease, then you raise debt.
18:35
The timing can vary to be clear.
18:36
Leases can be signed at different moments of the lifecycle of a data center project.
18:41
But generally speaking, it's a bit earlier and tracking that extremely closely as we do.
18:45
And every quarter, we sort of forecast that before whatever is published by folks.
18:48
Generally, it would be extremely accurate as well.
18:52
It helps us understand ahead of time what's going to get built under these binding contracts that cannot be canceled.
18:58
Yeah, and this is just a really big number, Jeremie.
19:00
Just just double double click on the over one trillion of data center leasing that's been signed that drives a lot of the forecast from the you know, credit, capital and markets model that Dan and the guys are working on.
19:15
We we just talked about it a week ago and we're gonna be doing more of this, but like this has been a pretty big story in the news recently, just how big the scale of some of this is.
19:26
Yeah, look, there's a bunch of indicators.
19:28
You've got to track at the building level to get to that number.
19:32
Things like what is the pricing, what is the contractor, escalator, that kind of stuff.
19:37
Some of these deals are public.
19:41
So if you look at all of the crypto miners, for example, companies like Applied Digital, Core Scientific, you can see the speed at which their backlog has been increasing and is now in the tens of billions of dollars.
19:51
But the thing is there's plenty of these companies that are private.
19:56
The public markets don't have a good sense of all the contracts that being signed between companies like QTS or EdgeConneX or Aligned and so on and so forth, Vantage, with all of these leading hyperscalers or AI labs, right?
20:09
And that's stuff that we track extremely closely.
20:12
And by having a good sense of all of these private companies and tracking them, every single one of them one by one, then we can essentially gather the data and put out these numbers that maybe other folks were missing.
20:26
Or I think a crystal clear indication of the strength to build out that we're going to see. Awesome.
20:30
Okay, Reyk, tell me about logistics, man.
20:33
Like, when we talk about modular data centers, we wrote a whole article on Lego data centers and how these things go up.
20:40
You're the guy who stares at a lot of satellite images and watches these things.
20:44
The trend seems to be that the the change in the satellite images is going faster, not slower, right?
20:50
I've got on screen one Yeah, yeah.
20:50
Of the charts that shows Google self-build and how they have change their design for the electrical side, Yeah.
20:59
Put it up faster for the modular sites.
21:02
Can you talk me through that this is going faster, not slower over time?
21:06
Yeah, I think this is one part of the build out that's a bit overlooked, to be honest. Right?
21:10
Where if this were like traditional data centers, maybe labor would be called into question, maybe timeline would be called into question.
21:18
But for us in our mind, the arrival or not even the arrival, but like the larger scale utilization of modular deployment is gonna be the what's the word I'm looking for?
21:32
Like The reason that this works and the reason that this happens, right?
21:37
When you look at traditional timelines of 18 months compared now to say like DayOne's timelines, where they can go down to nine months-ish for you know any number of hundred megawatts if you'd like.
21:48
You can look at bright ray also down there in Southeast Asia, which can build in like six months, right?
21:54
You can look at now the Anthropic reference designs also, which are going to be employing modular, right?
22:01
Modular is basically becoming more and more widespread across the industry.
22:06
You're gonna see build times drop.
22:08
You're gonna look at the Meta tents, you're gonna look at the new Amazon buildings, you're gonna see these buildings basically just pop up out of nowhere, right? These guys.
22:15
This is the this is the, you know, the Project Houdini program, right?
22:22
Where it's I think it's six.
22:22
Six gigawatts a year for the next like two or three years was the stated goal, like maybe six months ago by them.
22:31
These are the reason that this build-out works, right?
22:34
Without this, you'd have a lot more labor-intensive like build out, right?
22:39
We have a, I think we did the math in one part of the article where you see that if you have your MEP, like your more electrical side of the fit out within the data hall, if you have this more modularized, right?
22:51
For example, Say I take the power hall, right?
22:55
And I just imagine like a literal shipping container and I just slot it into the actual hall, like the shell of the data center, and I plug in a few things.
23:02
This demands much less labor than you know having 20 electricians in there trying to like actually coordinate everything and quote unquote stick build the entire hall.
23:12
And so you can imagine this just on a much grander scale.
23:16
This is the reason why this works, right?
23:18
Of course, right now we're working with the industrials team to write and do a competent letter, a competent newsletter on the labor for the 27 build-out.
23:27
But so far, I mean, everything points to the fact that modular is what's making this work.
23:32
And you can see it across the industry, really.
23:34
Yeah, I add again that if you actually visit data centers, you know, and some of the big ones as well as some of the two or three year old ones, as well as some of the smaller ones, you can clearly see the difference in the way how many of these new builds are fully designed and fully optimized to go as fast as possible on the build time, on the ramp time, the commissioning time, absolutely everything I get thought of for an accelerated timeline.
23:58
And again, like many folks tell us, they are going to pull this off in six months.
24:02
We purposefully haircut a lot of these assumptions.
24:05
But it's still extremely impressive when you actually visit these sites and see what folks are doing to accelerate while keeping the bar high in quality and and certainty of delivery and so on and so forth. Yeah.
24:16
And I mean, I I I think if people aren't I think the whole kind of point here is somebody might say, you know, like, you know, how do you measure modular adoption in this case?
24:27
Well it's like, well, we have it on the industrial tracker, but the other thing is that if you pr if you don't really follow suit in the modular build out or like this kind of design strategy.
24:38
You will probably not be allocated the labor that you need, nor will you be competitive within the build out market, right?
24:44
If I can offer you a data center in 12 to 14 months, I'm really not worth as much as the guy they can offer to you in eight to nine months.
24:50
And so it's kind of going to be able to, it's going to start forcing people's hands in this direction.
24:54
And we're already watching, you know, we're watching people do interesting things, right?
24:59
Like I'm watching Bright Ray is talking to multiple neoclouds right now to start selling just entire data center modules, right?
25:07
Not the like, you know, one to two megawatt crates, but the like actual hall that that you can ship and install into your core and shell if you'd like.
25:15
Or I mean with the core and shell from Bright Ray.
25:18
But people are going to keep coming to these folks that can provide this equipment.
25:21
People are going to keep buying it.
25:23
I think this build out is exceedingly led by the modular aspect. Yeah. Makes sense.
25:30
How about the the balance?
25:35
I mean we put this behind the paywall in there to be clear, but it seems like generally speaking, people are trying to call this shot whether you're gonna have accelerators and overbuild of data centers or whether you're going to be constrained on data centers and have a oversupply of accelerators or whether it's gonna be a perfect match.
25:53
There's a Maybe one comment to make there that we know about is just how how many people overdeploy GPUs and then use power capping.
26:03
And it seems like power, like the power envelope is still a constraint for a lot of these, whether it's grid connected or behind the meter, there's just only so much power you can bring on, even if you have the physical space and the chips available.
26:13
So you'll be power capping GPUs to fit in the envelope.
26:15
With how much the construction timelines are reducing in the modular sites, maybe you guys can just say high level, like You think there's gonna be a lot more data centers built in the next couple of years, right?
26:24
I mean, look, it's obviously tough.
26:27
I would say, generally speaking, the forecast for data centers is slightly easier in the sense that it's based on infrastructure signals that you can track ahead of time.
26:38
Chips can be slightly more volatile.
26:41
But overall, also have a lot of very strong indicators pointing to what's going to happen in 2026 and 2027.
26:47
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.
27:00
We're already in sort of oversupply of chips today as we speak.
27:06
It's always one always has to be slightly bigger constraint than the other.
27:11
And if you want to deploy something, if you want something like right now, it's easier to get, I would say, GPUs than data center megawatts.
27:18
But when we compare how these different supply chains are expanding into next year, and especially in 2028, we do see a risk for 2028, I would say, flipping and potential data center oversupply.
27:32
So that's going to be the thing we watch the most closely, just how fast can the chip supply chain expand.
27:39
And I would say that's obviously a strong positive to things like CapEx in terms of investing in new semiconductor fabs, buying new equipment.
27:46
And that's the type of stuff we've been flagging our clients for several months already.
27:53
So let's wrap on this one, guys.
27:55
I th great job taking us through all of this, super helpful.
27:57
we've obviously covered all all sorts of aspects of this from the moratoriums to the labor to the electrical equipment to the cooling equipment to the actual construction progress.
28:07
Like there's all sorts of things that play into a data center arriving on time that we think we cover in in great detail. I know you guys do.
28:13
let me get your direct reaction to people like Brad Gerstner showing our chart with an old forecast to Jensen.
28:22
Which basically says like you're gonna build too many chips and not enough have enough places to put And Jensen kind of just like taking it, acknowledging it, yeah, yeah, and moving on in the conversation.
28:31
Like where do you think the sentiment comes from?
28:33
And maybe you can just respond directly to that. I mean, I think it's.
28:35
I I think I think it's easier to feel Go ahead. I remember.
28:37
Like what's happening right now versus know how that feeling is going to evolve into like, you know, end of next year, so 12, 15 months from now.
28:46
Right now it's crystal clear that there is more supply of chips than supply of data centers.
28:53
Again, you talk to anyone that looks to deploy AI infrastructure, they can get access to GPUs, they cannot get access to land, power, shell in the next three, six months.
29:03
So we think the constraint right now is crystal clear.
29:07
Obviously, we want to make sure the situation is solved.
29:11
And so I think that's why Jensen absolutely wants to make sure this doesn't happen.
29:15
Also, mean, Jensen has a long history of being kind of a supply chain, you know, extremely smart about supply chain and he can be aggressive as well.
29:26
And so we've seen him securing tremendous amount of supply like, you 2021.
29:31
To secure GPUs for the crypto mining build outs which obviously crashed so there can be a bit of a cyclical aspect but you know I think overall it makes sense it makes perhaps the company slightly cyclical but the whole sector of semi is cyclical anyways so it doesn't change
29:45
too much his company is in extremely good financial health and I think he's just like writing these massive checks to secure supply from the memory supplier from the logic suppliers and he's just really good at securing supply from these guys
29:59
and I think he he wants to make sure he can build as much as he can and have no constraints. And so, yeah, it makes sense to pressure everyone in the supply chain to build more. And I think the the other thing I'd add on this too is I think I think a lot
30:04
And so, yeah, it makes sense to pressure everyone in the supply chain to build more.
30:04
And I think the the other thing I'd add on this too is I think I think a lot of it comes from people maybe operating on the not not the most solid foundation of like information, right?
30:22
It's one thing for, you know, it's one thing to understand that like, you know, yes, moratoriums are like slowing data center build out. Yes, it's true.
30:30
Grid connections are as well, labor is as well, equipment supply chains as well. All of it's true.
30:35
but being able to measure that like accurately is a different story, right?
30:41
There's Thank I don't know if there's anybody else out there that's really, I think Just like, as an aside, Jeremie, did you say that we were like looking at oversupply in twenty twenty eight before I continue?
30:57
I didn't say formally, I'd say it's the main risk as we get into 2028.
31:01
I was gonna I was gonna start talking about that but I don't know.
31:04
Yeah, let's just move on on that.
31:06
But yeah, no, so I'll just finish my thought.
31:10
Yeah, I think the other thing here is that a lot of people are operating on not the most solid foundation of information when they come to these conclusions.
31:21
Like Yeah, I I love Wall Street.
31:25
All of Wall Street sort of tells us, you you guys are playing wrong.
31:29
And obviously some of the great investors out there like Brad, but he's obviously not alone.
31:31
He just did it on a big podcast, but many other folks tell us, you guys are way too high.
31:38
And there's a lot of data sources out there that have data center forecasts that are materially smaller than what we project.
31:46
And we're familiar with a bunch of these and, you know, like a lot of these Claude-coded forecasts that track data centers and that, I don't know, just prompted Claude and be like, try a lot these sites, there's like three, four, five of them that have numbers that are 2x smaller than ours.
32:00
And some of our clients tell us, these guys are lower.
32:03
And just tell them, we didn't prompt.
32:05
We didn't prompt or water.
32:11
There's more that goes behind that.
32:14
So would say, generally speaking, a lot of sources that probably aren't really based on accurate information, unfortunately.
32:19
And obviously, every single CEO also just telling folks that, right now the biggest constraint is, again, like, land, power, shell go back to sort of the sentiment driven by what's happening right now.
32:31
Yeah, and maybe if I could just say something directly.
32:33
What set me off on this one is that like a month ago, the guy gets on stage and presents to a YouTube audience now of three hundred and thirty thousand thousand people, followed by this one from four days ago to two hundred and sixty-five thousand people in counting, cites our data and then just says it's wrong.
32:50
Says yeah, I don't think so.
32:53
Cite other stuff and say it's different, sure.
32:55
cite our data, provide specific critiques to it.
32:59
Great, we can always learn on offhand. This is the forecast. We think it's wrong. Seems a little strange.
33:05
Glad you guys could you know, respond directly to some of that in this article.
33:10
And I look forward to walking through twenty seven and then continuing to see if the track record that you guys have on the data center model continues to hold true.
33:17
Yeah, I mean, just to be clear, the goal was not to respond to one person or whatever the goal is.
33:23
It's a huge topic that obviously people ask us all the time.
33:27
Our institutional clients obviously have been aware of how we do things and what we've been projecting.
33:33
I think we just wanted to give a piece of that to our broader audience and put the content out there and see how folks would react.
33:40
Yeah, again, the point is we're extremely Mm. Yeah.
33:43
Bullish in terms of how big the infrastructure for Delta will be in 2027.
33:48
We think we have a rock solid indicators to about that. Yeah.
33:51
And I mean it's you know, it's not the it's not the first time we've gotten the feedback that, hey, your guys' numbers are insane.
33:57
Like we don't know what to do with these numbers.
34:00
Nobody agrees with these numbers.
34:01
So it's not the first time.
34:01
So we wanted to like directly address that whole thing, honestly. Awesome guys. Okay. Well great job today.
34:09
Appreciate you taking the time to educate everybody as always on what you what you spend all your days doing and how accurate it is. Appreciate it. All Cheers, Jordan. Cheers Jordan. Right, we're done now.
34:20
Can you smile for the camera quickly? Smile for the algorithm.
34:23
Come on, with some teeth, Reyk, buddy.
34:25
Smile Can you what, for the algorithm, Jeremie.
34:28
Look at the camera and He wants you to smile. Smile.
34:30
You want that on the fucking thumbnail?
34:30
So I we have to see each other. Okay, good job. I'm stopping.