WEBVTT
00:00:00.560 --> 00:00:03.839
Next, it's the installed data center podcast.
00:00:23.199 --> 00:00:25.679
It is your host, Andy David.
00:00:28.079 --> 00:00:30.399
Welcome to the Inside Data Center Podcast.
00:00:30.559 --> 00:00:37.920
We are back again, joined by Ben Vallieri for the second edition of Data Center News and GPUs.
00:00:38.159 --> 00:00:39.119
Afternoon, Ben.
00:00:40.159 --> 00:00:41.280
Good afternoon, Andy.
00:00:41.359 --> 00:00:44.000
Yeah, we've uh it's been a little while since the last one, isn't it?
00:00:44.240 --> 00:00:45.920
What's happened in the past couple of months?
00:00:46.240 --> 00:00:46.640
Exactly.
00:00:46.719 --> 00:00:47.039
I know.
00:00:47.119 --> 00:00:48.159
We said we do one a month.
00:00:48.240 --> 00:00:54.079
We're on about one every six weeks at the moment, which is uh make making it difficult to catch up with the news, but there you go.
00:00:54.159 --> 00:00:57.679
But you know, you've been out at um GTC and obviously we've been busy travelling around.
00:00:57.759 --> 00:00:59.840
So but still, good to get it in.
00:00:59.920 --> 00:01:02.719
We're still stuck with the name, Data Center News and GPUs.
00:01:02.880 --> 00:01:07.200
Haven't had any suggestions for better ones, so that's what we're uh what we're going with now.
00:01:07.359 --> 00:01:18.640
We're gonna give you another whistle stop run through of some of the biggest news or the most traction news that we've had, um whether it's GPU or or data center related.
00:01:18.959 --> 00:01:20.400
So now we've started this, Ben.
00:01:20.480 --> 00:01:21.040
Do you want to start?
00:01:21.120 --> 00:01:24.319
Ben's fresh off a plane from the US, so there you go.
00:01:24.959 --> 00:01:27.280
Yeah, the uh the joys of a Heathrow fire.
00:01:27.359 --> 00:01:31.680
Um I think the the biggest one has got to be GTC, um, to be honest.
00:01:31.920 --> 00:01:34.959
The GTC was super interesting.
00:01:35.040 --> 00:01:36.959
Um, I was out there for the week.
00:01:37.280 --> 00:01:53.439
The keynote, I mean, aside from feeling like we're in a bit of a cult, to be honest, because when you're queuing at half past seven in the morning for a 20,000-person stadium and you're a long way at the back of the queue, and then it's sold out and people are being turned away to see a tech CEO speak.
00:01:53.599 --> 00:02:00.719
I mean, it's it's a little bit culty, but on the other hand, super, super interesting based on the content.
00:02:00.799 --> 00:02:04.400
And two things really stood out to me from it.
00:02:04.719 --> 00:02:12.560
One being there was no mention of training anywhere in the keynote, no mention of model training at all.
00:02:12.639 --> 00:02:50.479
Um, the focus was almost entirely on inference, um, and this arbitrary measure that Jensen seems to have introduced for the AI factory concept, which is tokens per second per megawatt, which I think is quite an interesting um nod as to where NVIDIA seems to be trying to guide the market, um, which from where I'm sat and from the massive amount of financial modelling work that have done over the course of the past six months, months or so, makes perfect sense because training, you're constrained really, in terms of how much revenue you can make.
00:02:50.560 --> 00:02:51.759
It's a race to the bottom.
00:02:51.919 --> 00:03:09.439
Someone is always going to be able to undercut you on GPU per hour charges, but inference, where you've got the Xerox model, where it's copies basically, you deploy the infrastructure, you get your customers to come in, um, and their charges are coming out of OpEx, the cash flow profiles are fundamentally different.
00:03:09.599 --> 00:03:14.000
So, from a hardware owner standpoint, really interesting that that's where things are going.
00:03:14.639 --> 00:03:27.919
From a business standpoint as well, and then actually creating enterprise value with all of this wonderful training that has been done and these models that are now reaching these lofty heights by various benchmarks.
00:03:28.960 --> 00:03:31.199
Inference is where the value is going to be created.
00:03:31.280 --> 00:03:36.719
Now, what that value creation looks like, I think that's still the question that needs to be asked.
00:03:37.039 --> 00:03:48.319
But it seems like that's where the industry is going to be going probably pretty quickly over the course of the past over the past next 12 months or so.
00:03:48.400 --> 00:03:51.759
Um, which, as you know, how this industry moves, right?
00:03:51.840 --> 00:03:54.479
It's like dog years, 12 months is going to be seven years.
00:03:54.560 --> 00:03:56.319
Um, so we'll see where we end up.
00:03:56.800 --> 00:03:57.520
Yeah, yeah.
00:03:57.599 --> 00:03:58.800
I watched the Jensen keynote.
00:03:58.879 --> 00:04:08.960
Obviously, I wasn't, I wasn't there, I wasn't in the queue, but yeah, he he kept talking about it being an inflection point, didn't he, for AI and trillion dollar data center industry, et cetera, et cetera.
00:04:09.039 --> 00:04:18.560
And then since then you've seen quite a lot of shifts in um the hyperscalers kind of strategy, I guess, is for want of putting it a better way.
00:04:18.639 --> 00:04:26.079
So there's definitely a a period of um thought going on at the moment as to what's next, what do we do next?
00:04:26.160 --> 00:04:32.800
You know, what I think someone said the other day, you don't want to build a data center that is totally useless because that's kind of where they're getting to now.
00:04:32.959 --> 00:04:35.279
So I think inflection point is a good way to look at it.
00:04:35.360 --> 00:04:43.839
And like you say, moving away from training and moving into the actual can we make money out of AI, I suppose, is is where most people are looking at it now.
00:04:43.920 --> 00:04:44.720
Like, where's the money?
00:04:44.800 --> 00:04:45.439
Where's the cash?
00:04:45.519 --> 00:04:46.720
Where's the return of investment?
00:04:46.800 --> 00:04:48.240
Because it's a big investment.
00:04:48.720 --> 00:04:49.600
Yeah, absolutely.
00:04:49.680 --> 00:04:53.360
And I think that Deep Seek has been really interesting in that regard.
00:04:53.439 --> 00:05:00.240
Um, because obviously the market reaction was everything's fallen apart, um, that's it, it's the end of days.
00:05:00.399 --> 00:05:25.360
And then from what was said in the keynote, because Deep Seek and the reasoning models that have proliferated off the back of it due to the fact that DeepSeek open source an awful lot of their work, reasoning models chew through a huge amount of tokens, um, which again, from a hardware owner's perspective, is super interesting.
00:05:25.600 --> 00:05:39.920
Um, I think the statistic that Jensen used was it requires like a hundred times more compute than expected, um, which is great from both the data center standpoint and also from a compute deployment standpoint, because we're gonna need a hell of a lot more capacity.
00:05:40.160 --> 00:05:45.519
But again, that then feeds into it's all well and good needing the capacity, but what is that capacity going to be used for?
00:05:45.759 --> 00:05:54.800
So I think use cases um are gonna be the uh the most important thing going forward, especially the second part of the keynote.
00:05:55.040 --> 00:05:59.920
Um if you are having to deploy 600 kilowatt racks or above, right?
00:06:00.160 --> 00:06:03.839
Which from a data center standpoint sounds like a bit of a headache.
00:06:03.920 --> 00:06:06.879
Um, I'm sure I'm sure you'll have some perspective there.
00:06:07.680 --> 00:06:08.879
Yeah, I think it's crazy.
00:06:09.519 --> 00:06:15.120
The the rack side of it, from a non-technical, it's just crazy where it's going and where it can get to.
00:06:15.199 --> 00:06:18.720
And I think it's one of the biggest challenges, it's what density, where does it end?
00:06:18.800 --> 00:06:20.879
You know, what is the density we're we're building for?
00:06:20.959 --> 00:06:27.120
And I think that is why we've seen a lot of this, as I say, like fluctuation in the markets and a bit of uncertainty.
00:06:27.199 --> 00:06:37.279
And you can totally understand why if you're investing tens and tens of billions of dollars into this industry in a period of such uncertainty, it's it is a big gamble.
00:06:37.360 --> 00:06:42.720
And at some point, the money behind it does have to say, hang on a minute, like you know, are we doing the right thing here?
00:06:42.879 --> 00:06:47.279
They might pause and go again, but they need to re-evaluate and ensure they're doing the right thing.
00:06:47.360 --> 00:06:53.360
And you you've kind of got to re-evaluate every every couple of weeks at the moment, and that's again that that is a challenge.
00:06:53.439 --> 00:06:56.959
And you know, NVIDIA are one of others as well, aren't they?
00:06:57.040 --> 00:07:01.439
So there's there's all this research going on into the background, and yeah, it's a it's a crazy time.
00:07:01.519 --> 00:07:05.040
And I I liked the bit on the the humanoid, the the robots as well.
00:07:05.120 --> 00:07:06.079
Like, you know, that was cool.
00:07:06.160 --> 00:07:13.120
And I think when you start talking about data and capacity and demands in that in that area, you you're in another world.
00:07:13.199 --> 00:07:16.480
And I think, yeah, I know that's not here now, but that's coming.
00:07:16.560 --> 00:07:19.680
And every year he does a keynote as a better robot.
00:07:19.759 --> 00:07:23.680
So I don't think it'll be long until those uh those robots are walking among us.
00:07:23.759 --> 00:07:25.279
And if they're not already, you know, who knows?
00:07:25.360 --> 00:07:25.920
They may be already.
00:07:26.160 --> 00:07:26.800
Yeah, exactly.
00:07:26.879 --> 00:07:30.399
I mean, I think the other interesting piece there to that point too.
00:07:30.560 --> 00:07:34.959
One robotics is the the I mean, no one knows what that's going to look like.
00:07:35.040 --> 00:07:39.519
But I mean, if you've spent enough time reading sci-fi, you've got probably got a pretty good understanding of what's going on.
00:07:39.600 --> 00:07:43.519
Didn't he say it'll be yeah, didn't he say it'll be the biggest industry ever?
00:07:43.680 --> 00:07:44.879
Something along those lines.
00:07:45.199 --> 00:07:46.000
Yeah, right.
00:07:46.480 --> 00:07:57.680
Like everything that is everything that moves could be potentially automated, and everything that moves could therefore, by extension, potentially require some sort of thinking per se, right?
00:07:57.759 --> 00:08:06.160
Which means that it might be a model, it might be whatever the next iteration of AI is, and that's probably going to need some compute of some sort.
00:08:06.319 --> 00:08:13.680
Um, I think the other interesting thing though was that there was no real mention of the hyperscalers in like a partnership context, right?
00:08:13.759 --> 00:08:23.199
Because if you look back at previous keynotes, there was generally a nod towards like partnerships with Meta or partnership with AWS or Azure or something along those lines.
00:08:23.360 --> 00:08:25.360
But those guys are making their own chips now, right?
00:08:25.759 --> 00:08:28.160
That's eating into NVIDIA's market share.
00:08:28.399 --> 00:08:39.840
Um Lepton AI was a purchase that NVIDIA is apparently considering that cropped up recently, and Lepton offers GPUs per hour, right?
00:08:40.000 --> 00:08:41.679
Which is kind of cloud service.
00:08:41.919 --> 00:08:55.360
So wouldn't be surprised based on what was not said and the relative degree of complexity that is going to be, I mean, that is required to deploy a 600 kilowatt rack, right?
00:08:55.440 --> 00:09:00.080
But wouldn't be surprised if we start seeing NVIDIA going a little bit more into the cloud side as well.
00:09:00.159 --> 00:09:01.279
Because I mean it makes sense, right?
00:09:01.360 --> 00:09:05.360
If your competitors are building chips, why not compete with them at their game too?
00:09:05.840 --> 00:09:06.720
Yeah, exactly.
00:09:06.879 --> 00:09:10.320
It is a it is a battle for for business at the end of the day, I think.
00:09:10.399 --> 00:09:11.440
You have to remember that.
00:09:11.679 --> 00:09:31.440
And it's similar on the on my side, the biggest news was was the TD Cowan and Microsoft's news of cancelling leases, which ties in again with the moving away from partnerships because it was all linked to the allegedly it was linked to the Open AI partnership, which now we believe OpenAI are going to look to build, develop their own infrastructure, which again makes sense.
00:09:31.519 --> 00:09:34.559
You know, if you want to own the infrastructure, it's probably the right way to do it.
00:09:34.720 --> 00:09:37.279
Whether that's true or not, we'll have to we'll have to wait and see.
00:09:37.360 --> 00:09:40.879
But yeah, Microsoft cancelling um a number of leases.
00:09:41.039 --> 00:09:44.399
I think this last week it was two gigawatt across the US and Europe.
00:09:45.279 --> 00:09:53.919
Previously, we saw projects start re-evaluation of strategy, and we know there's a strategy review happening here, but again, it ties into that.
00:09:54.240 --> 00:09:55.279
What is the strategy?
00:09:55.360 --> 00:09:59.919
And I don't mean with Microsoft, I mean generally, like what do we do going forward here?
00:10:00.080 --> 00:10:02.320
How how do we capitalise on this?
00:10:02.480 --> 00:10:04.240
Are we moving away from training?
00:10:04.320 --> 00:10:07.759
You know, are these data centers being built for the right purposes?
00:10:07.840 --> 00:10:09.919
Are they in the right places for for what they require?
00:10:10.080 --> 00:10:18.480
I'm sure there's a big review going on somewhere with a lot a lot more intelligent people than myself looking into it and trying to work out where best to spend this capex.
00:10:18.720 --> 00:10:28.799
I never see it as a as an industry challenge, I just see it as a another re-evaluation point where everyone's kind of like, oh no, you know, but what do we do?
00:10:28.960 --> 00:10:32.480
We've just announced$100 billion of CapEx for 2025.
00:10:32.639 --> 00:10:35.279
Are we spending$100 billion in the right place?
00:10:35.440 --> 00:10:37.919
And I think we will see that probably across this year.
00:10:38.000 --> 00:10:40.720
I think we'll continue to see it with um some of the other hyperscalers as well.
00:10:40.960 --> 00:10:46.240
Won't just be uh Microsoft that are that are reviewing where and when to spend.
00:10:46.799 --> 00:10:47.600
Oh yeah, definitely.
00:10:47.679 --> 00:11:00.320
And I I think to the um the Microsoft point with OpenAI, um OpenAI have also just um announced a partnership with or a very large deal with Courtweave, right?
00:11:00.399 --> 00:11:02.639
Who've also just very recently gone public too.
00:11:02.720 --> 00:11:08.159
And I think that amounted to like$12 billion or something, which is absolutely astronomical.
00:11:08.639 --> 00:11:24.960
Um that is kind of from my perspective and from what I'm seeing with other people, contingent on OpenAI being able to successfully close their next funding round, which apparently they're very close to doing with SoftBank it being like a 40 billion round or something absurd like that.
00:11:25.039 --> 00:11:31.120
But then that's also contingent on OpenAI being able to transition to a fully full profit model as well, right?
00:11:31.360 --> 00:11:39.440
And whether or not they're going to be able to do that from a legal challenge standpoint kind of remains to be seen.
00:11:39.679 --> 00:11:50.480
Um, and then also the two gigawatts of cancellations I've read this morning that while Microsoft has pulled back on those, apparently those same sites were scooped up by Meta and Google very, very quickly.
00:11:50.799 --> 00:11:57.759
So I think this kind of taps into what we were speaking about in the the first one that we did of all of those six weeks ago, right?
00:11:57.919 --> 00:12:10.480
In that I think there's a little bit of brinkmanship going on, in that I think things probably started a little bit before people really wanted them to, but you can't not do it in case your competitors have something that you want, right?
00:12:10.559 --> 00:12:15.120
It's better to have something and give it up and pay a little bit of money for it, right?
00:12:15.279 --> 00:12:19.679
As opposed to need something that you can't get because your competition's already got hold of it.
00:12:20.000 --> 00:12:21.440
Yeah, totally, definitely seeing it.
00:12:21.519 --> 00:12:31.759
And yeah, I've had I've had similar as well, which again is why I don't ever see the I I I laugh when I post uh these kind of announcements because you get the panic stations from the from everybody.
00:12:32.399 --> 00:12:40.720
You know, don't read too much into it, is what I always say to people, you know, like do your research, understand the market, talk to people, and you can actually gauge what is actually really happening.
00:12:40.799 --> 00:12:47.840
You'll never see what's really happening in the public, but you know, pick the phone up, have conversations, and you'll and you'll quickly you'll quickly see what's out there.
00:12:48.799 --> 00:12:51.679
Ask people questions and they tend to tell you the answers, right?
00:12:51.919 --> 00:12:52.320
That's it.
00:12:52.480 --> 00:12:53.440
Especially in this industry.
00:12:53.600 --> 00:12:56.240
Can't you can't tell anyone anything, but you can tell anyone everything.
00:12:56.320 --> 00:12:59.120
It's uh look at it.
00:12:59.279 --> 00:13:04.480
Um I also I quickly scanned through some of the projects that broke as well in uh in the last few weeks.
00:13:04.639 --> 00:13:15.200
And again, you know, when you look at investment, you're continuing to see this wave of hundred-acre sites, multi-billion dollar investments, new funds raising.
00:13:15.279 --> 00:13:17.440
It's still happening regularly, regularly.
00:13:17.519 --> 00:13:26.320
So um we had a£1 billion data center and technology part in Anglesey and Wales, which I think I raised a lot of eyebrows with regards to who's the customer for that one.
00:13:26.399 --> 00:13:28.399
You know, it's a it's an interesting location.
00:13:28.639 --> 00:13:38.000
UAE obviously continued to invest significantly, 30 to 50 billion euro AI data center in France, big, big range there.
00:13:38.080 --> 00:13:47.360
Um Stone Shield, 500 megawatt in in Spain, Azora, 2 billion development in Zaragoza against Spain, Evrock, 96 megawatt in France.
00:13:47.600 --> 00:13:55.440
QTS obviously got approval for the big site in the north northeast of England, uh, which is approximately 10 billion dollar investment.
00:13:55.600 --> 00:13:56.799
And it goes on and on.
00:13:56.879 --> 00:14:00.480
So I think you're continuing to see the scale of investments.
00:14:00.559 --> 00:14:20.320
Um, one thing I'm also starting to see now, and or have more conversations with, is actually in the um whether you call it edge or colo or metro edge, whatever you want to call it, but there's certainly a lot of investment now starting to look back at the smaller edge data centers, so five to twenty megawatt city-based.
00:14:20.480 --> 00:14:28.080
I think an area that's been forgotten because you're seeing all these big numbers, but actually is a massive, massive market.
00:14:28.159 --> 00:14:39.120
And I think somebody, some company at some point will take huge advantage of that and you'll see um you'll see somebody do very well out of it while everyone else has got their eyes on the half a gigawatt sites wherever they may be.
00:14:39.360 --> 00:14:40.000
Yeah, right.
00:14:40.240 --> 00:14:46.799
And also like the the absurdity of calling a five to twenty megawatt data center small is kind of funny as well, right?
00:14:47.600 --> 00:14:49.440
I I don't even know what edge is anymore.
00:14:49.519 --> 00:14:57.440
Like I know I used to I remember recording podcasts like three years ago, and I'd ask anyone that said they were at edge what edge means to them, and you've got a different definition every single time.
00:14:57.519 --> 00:14:58.799
And I'm still at that point.
00:14:58.960 --> 00:15:03.919
I don't know what small is, I don't know what edge is, I don't even know what big is anymore.
00:15:04.000 --> 00:15:06.960
It's like what what actually how do we scale any of this?
00:15:07.039 --> 00:15:09.200
It's it's like when people say, What does hyperscale mean?
00:15:09.600 --> 00:15:13.440
Hyperscale is a definition, means like continuing to grow or something along those lines.
00:15:13.519 --> 00:15:19.679
But but actually, we don't ever have any sort of market for these words that we use all the time.
00:15:19.759 --> 00:15:21.039
It's just but there you go, yeah.
00:15:21.120 --> 00:15:33.279
The smaller data centers than the bigger ones, there's definitely a huge uh a huge market for that as you start to join the dots of these major campuses across whatever countries they're in and continents as well.
00:15:33.759 --> 00:15:42.240
Yeah, yeah, and I I think also like the France is an interesting, very, very interesting area at the moment, also from a neo-cloud GPU as a service standpoint.
00:15:42.399 --> 00:15:47.120
Um, one other announcement um to add to your list as well.
00:15:47.360 --> 00:15:51.600
E and I, G42, um, one gigawatt in Italy as well.
00:15:51.840 --> 00:16:01.440
So that I mean the fact that it's like, right, we're gonna have a gigawatt here, we're gonna have a gigawatt there, as opposed to an 100 megawatt or a 200 megawatt, I think that's probably a pretty good indication of the direction of travel.
00:16:01.519 --> 00:16:08.240
But I think France is interesting in that regard specifically, because you've got the the G42 and the AMD deal with Data One, I think it was.
00:16:08.399 --> 00:16:12.559
Yeah, they are um, which is massive in and of itself.
00:16:12.720 --> 00:16:26.399
But then at the same time, same week, this is off the back of the um the AI summit in Paris, Fluid Stack announcing a massive, massive partnership with um Mistral to develop a huge amount of infrastructure in France as well.
00:16:26.559 --> 00:16:29.519
Same with Cesurs as well, another gigawatt facility.
00:16:29.759 --> 00:16:41.360
Um and then ENI, not ENI, sorry, EDF, opening up, I think, two gigawatts of land parcels uh close to various nuclear power generation facilities around the country as well.
00:16:41.519 --> 00:16:58.080
Um France is interesting because if you look at the the energy mix of France relative to every single other European country, aside from the Nordics, they're really the only ones who can support the kinds of developments at the speed that they need to happen right now.
00:16:58.240 --> 00:16:58.879
Why is that?
00:16:59.039 --> 00:17:02.240
They've got a huge amount of nuclear power generation capacity.
00:17:02.399 --> 00:17:16.960
Um, and I think we'll probably see increasing discussion around the need to develop more of that fast in non-traditional markets for nuclear, to be honest, because I mean France has shown that it can work, right?
00:17:17.039 --> 00:17:19.359
They wouldn't be getting these deals if it wasn't viable.
00:17:19.519 --> 00:17:24.559
Um, I wouldn't be surprised if we see some projects rushed through well, relatively speaking, rushed for nuclear.
00:17:24.640 --> 00:17:27.279
So we're probably still, what, five, ten years out, to be honest.
00:17:27.359 --> 00:17:29.440
But I think that's where things are gonna go.
00:17:29.680 --> 00:17:31.519
Yeah, telly house ended last week as well.
00:17:31.599 --> 00:17:41.279
I think a billion euro investment in France, include and it I think it was a uh 90, I might be wrong on this, but around 96 megawatt in Marseille was part of that.
00:17:41.440 --> 00:17:43.200
So yeah, exactly.
00:17:44.480 --> 00:17:47.920
One of the smaller edge facilities in Marseille.
00:17:48.000 --> 00:17:50.079
But Marseille's an interesting again, France is interesting.
00:17:50.240 --> 00:17:51.920
Marseille, a lot of development coming up there.
00:17:52.000 --> 00:17:54.480
Um, it's uh you know, connected very well.
00:17:54.640 --> 00:17:57.440
Obviously, Paris is a big, big challenge for a number of reasons.
00:17:57.680 --> 00:18:00.960
A talent is one of the worst places to to recruit for.
00:18:01.039 --> 00:18:04.799
Um, I can vouch for that with with definite certainty.
00:18:04.880 --> 00:18:06.720
Um, we always try our best, but it's difficult.
00:18:06.799 --> 00:18:10.720
Um, but yeah, southern France, yeah, good, good areas, a lot of a lot of development.
00:18:10.880 --> 00:18:18.480
The other um campus or development that I find quite interesting is old Kevin O'Leary's Wonder Valley in Alberta in Canada.
00:18:18.559 --> 00:18:20.079
I don't know if anyone's seen the video.
00:18:20.240 --> 00:18:22.480
It reminds me of a Centre Parks video.
00:18:22.640 --> 00:18:26.480
Um, every time I watch it, I I see the deers running around in the background.
00:18:26.559 --> 00:18:34.559
I think it's it's like me back at Centre Parks, similar price to a week in centre parks as well, 70 billion dollars to anyone, anyone that's ever been in the UK.
00:18:34.640 --> 00:18:36.000
But it's an interesting development.
00:18:36.160 --> 00:18:44.400
I think he um I think there was a video probably a few weeks ago of him needs to raise the first 10 billion or something for that development.
00:18:44.480 --> 00:18:56.480
But he's um interesting character, he's quite um you know, like for those that that don't know who he is, he's he's kind of similar to Dragon's Den type type character, um, very public, big investor.
00:18:56.640 --> 00:19:00.559
Um, whether it'll happen or not, I think is uh is up for debate, but it's it's worth a look.
00:19:00.640 --> 00:19:01.519
Wonder Valley, it's cool.
00:19:01.599 --> 00:19:03.599
If anyone hasn't seen it, yeah, go and go and watch the video.
00:19:03.680 --> 00:19:06.079
And if you've been centre parts, you'll know exactly what I mean.
00:19:06.160 --> 00:19:10.160
But if he does do it, I'm definitely going because it's uh it looks cool.
00:19:10.400 --> 00:19:11.920
Oh, yeah, it looks super interesting.
00:19:12.000 --> 00:19:16.960
And what's the total power capacities for seven and a half gigs?
00:19:17.200 --> 00:19:18.559
I was gonna say what's stuck in my head.
00:19:18.799 --> 00:19:21.440
Yeah, seven gigs in my ridiculous.
00:19:21.839 --> 00:19:25.680
Um, but then like to that point, it's in Canada, right?
00:19:25.759 --> 00:19:44.079
And they're gonna be targeting US customers, and that's potentially a difficult proposition right now, given uh the ongoing geopolitical struggles, I think would be the uh the most delicate way to put uh a description of the relationship between the US and Canada right now.
00:19:44.319 --> 00:19:47.200
Um I think it's super interesting.
00:19:47.359 --> 00:19:53.759
I think Canada's a really, really interesting proposition at the moment because of the amount of power and the amount of energy that they have available.
00:19:54.319 --> 00:19:59.599
Um it's just if things kind of carry on as they are, who's the off take?
00:19:59.759 --> 00:20:01.359
Because that's a hyperscale facility.
00:20:01.440 --> 00:20:03.279
That has to be a hyperscale facility.
00:20:03.440 --> 00:20:06.000
Um, but all of the hyperscalers are American.
00:20:07.359 --> 00:20:08.880
How do you square that circle, right?
00:20:09.039 --> 00:20:09.440
Exactly.
00:20:09.519 --> 00:20:15.759
And you are seeing you know, geopolitical uh is coming into strategy now, 100% um definitely seeing it.
00:20:15.920 --> 00:20:27.359
Another point I thought of while you were talking is actually, which again, you'll be quite aligned, haven't written it down, but the move away from the likes of Crusoe and Applied and that moving away from the Bitcoin into the into AI infrastructure.
00:20:27.440 --> 00:20:30.799
I think you've seen um obviously it's been happening for a while.
00:20:30.880 --> 00:20:44.319
You know, I've been working with funds that have been doing this for a good few years, but um you've seen a few more public announcements now of we are no longer, you know, in the Bitcoin world, we are now an AI infrastructure developer.
00:20:44.480 --> 00:20:52.880
So that's quite an interesting shift as you start to, especially with the AI, and if we are moving away from learning, etc., where does all that fit in?
00:20:52.960 --> 00:20:57.920
Because a lot of these are huge scale facilities, but in fairly remote locations.
00:20:58.000 --> 00:20:59.920
So it's going to be interesting to see where that market goes.
00:21:00.000 --> 00:21:01.440
But in the US, you're seeing it now.
00:21:01.519 --> 00:21:08.960
There's a very, very clear shift from we're a bit above to we're definitely sitting on the AI side of the fence now.
00:21:09.680 --> 00:21:21.119
Yeah, and I mean, I think that it's an interesting move that makes sense, really, because if you look at what Bitcoin mining is, it's massive amounts of compute deployment incredibly quickly.
00:21:22.000 --> 00:21:27.039
Pretty good experience if you're looking at what's going on in the AI space now, what's required.
00:21:27.200 --> 00:21:32.720
Those facilities tend to have massive power capacity available.
00:21:33.039 --> 00:21:48.960
Um depending on where the where the mining is taking place, that does tend to influence how socially, geopolitically, morally, whatever you want to call it acceptable Bitcoin mining is seen to be.
00:21:49.200 --> 00:21:59.759
Um, I think that perceptions will change because there's some really interesting work going on in the Bitcoin mining space around deployment in support of.
00:22:00.240 --> 00:22:09.839
Renewable energy, for example, like microgrids, for example, and Daniel Batten of C4 Capital does some really, really interesting work in that space.
00:22:10.079 --> 00:22:21.279
Um, but I think the shift away from Bitcoin mining into specific AI compute, GPU compute, AI infrastructure makes sense for a couple of reasons.
00:22:21.440 --> 00:22:28.160
One, as I said before, these companies have the experience in doing it, like um core weave, for example, core scientific.
00:22:29.200 --> 00:22:31.680
It's that's what they were doing first.
00:22:31.759 --> 00:22:34.480
Now they've gone public with Core Weav, right?
00:22:34.799 --> 00:22:42.079
Um Bitdigital has also spanned off a new entity recently, white fiber, again for the same kind of reason.
00:22:42.160 --> 00:22:47.519
And there's also probably a pretty compelling argument that the reason that's being done is from an exit multiple standpoint.
00:22:47.680 --> 00:22:52.880
If you've got any Bitcoin mining on your balance sheet, you're going to be probably capped at maybe a 5x EBITDA multiple.
00:22:53.119 --> 00:22:57.519
If you are AI infrastructure, that's significantly higher, like 15x plus, right?
00:22:57.599 --> 00:23:03.119
So it's significantly more valuable as a business, even though it's basically just servers doing math, right?
00:23:03.759 --> 00:23:08.400
What's the fundamental difference in terms of the activity that is being done?
00:23:08.480 --> 00:23:10.160
It's not really that different.
00:23:10.240 --> 00:23:18.960
It's hardware using huge amounts of power in a big shed somewhere with a huge amount of power available to it that's going to be difficult to cool depending on how much is in there.
00:23:19.119 --> 00:23:21.039
The output is slightly different.
00:23:21.279 --> 00:23:29.920
Um, depending on who you ask, you could probably make a pretty compelling argument that Bitcoin mining is creating more value at the moment than all of the training.
00:23:30.079 --> 00:23:44.240
Um, but I think that's probably a trend that we'll see accelerate because I mean, even if the companies don't shift into hosting and deploying the infrastructure themselves, they've got the PPAs, they've got the facilities, they've got the power available to them.
00:23:44.400 --> 00:23:48.799
And that is the one thing that AI as a whole is going to need going forward.
00:23:48.880 --> 00:23:54.079
So going back to the keynote, first thing that Jensen said, we're a power-constrained industry.
00:23:54.240 --> 00:23:56.559
It's not compute anymore, it's power availability.
00:23:56.640 --> 00:24:06.799
And if you've got 600 kilowatt racks coming in 2027, realistically, we could probably see a meg not long after that, maybe two meg, not too long after that.
00:24:07.279 --> 00:24:08.559
That's a lot of power.
00:24:08.640 --> 00:24:10.000
Um, where's it going to come from?
00:24:10.079 --> 00:24:11.359
Who's already got access to it?
00:24:11.599 --> 00:24:13.759
The miners have got a lot of it right now in the short term.
00:24:14.079 --> 00:24:19.920
Yeah, and I and I regularly get asked, what's the what's the current price for powered land?
00:24:20.000 --> 00:24:36.799
You know, like what is the what is the value of, let's say, or whatever the whatever the power is, but you know, powered land is is certainly going to increase in value, I would say, over the over the coming years, and they'll probably come a point where, yeah, is it better to develop it or flip it?
00:24:36.960 --> 00:24:41.440
And and I guess there'll be the conversations that the the people in those positions will have.
00:24:42.400 --> 00:24:49.200
The other thing I wanted to talk about was um just events, but data centre world London.
00:24:49.359 --> 00:24:51.119
I don't know, did you go or not?
00:24:51.440 --> 00:24:52.640
Yeah, we were on the stand.
00:24:52.880 --> 00:24:57.519
Um, so Pete and I, um Panchayah Limited, we were exhibiting with PMY.
00:24:57.680 --> 00:24:58.720
Um so yeah, we were there.
00:24:58.799 --> 00:25:00.079
It was uh interesting event.
00:25:00.240 --> 00:25:02.079
I just wanted slightly different.
00:25:02.400 --> 00:25:03.839
I just wonder how busy it was.
00:25:04.000 --> 00:25:04.960
That was all I like.
00:25:05.279 --> 00:25:08.240
I just couldn't believe the first day was mad.
00:25:08.319 --> 00:25:10.319
Like absolutely manic.
00:25:11.119 --> 00:25:13.440
Um it was non-stop.
00:25:13.519 --> 00:25:20.880
Like we got onto the sand shortly before nine o'clock, started talking at about half past nine, and then didn't finish talking until about half past four.
00:25:20.960 --> 00:25:23.519
So really, really solid footfall.
00:25:23.839 --> 00:25:37.599
Um, in terms of demographics and the types of conversations that were being had, last year felt a little bit more like people were there and they were ready to deploy, right?
00:25:37.759 --> 00:25:49.039
So they were instead of tire kicking, it was more of right, we're we've got a project, we've got a pilot project, we want to understand this pilot project, maybe we want to deploy some hardware, maybe we want to continue the conversation afterwards.
00:25:49.200 --> 00:25:53.599
Um, and that was also more general, so it wasn't just AI last year.
00:25:53.759 --> 00:25:57.359
Um, there was a bit around visualization, there was a bit around render.
00:25:57.519 --> 00:26:17.759
Um, this year, pretty much only AI, but also a little bit more hesitation around what's going on, how best to do things, um which makes sense given current market conditions and the speed with which everything is moving, but otherwise, really, really good event.
00:26:18.079 --> 00:26:23.680
Um I was just thinking when I was there of yeah, exactly.
00:26:23.839 --> 00:26:32.480
I was just thinking when I was there of I don't know how many years I've been going to that event, and that's why I use it as a comparator because I've been going for let's say 10 years.
00:26:32.720 --> 00:26:43.440
I've no idea how many years I've been going, but every year it just gets busier and busier and busier, and I think it it's a good way to just understand what's happened to the industry, isn't it?
00:26:43.599 --> 00:26:48.799
And I don't mean how many stands are there, I just mean how many people like in general are actually there.
00:26:48.880 --> 00:26:57.920
I'm sure the event organizers will give us stats around you know, there's more stands or more football or whatever, or more registrations, but it's just the the amount of the mass of people.
00:26:58.160 --> 00:27:09.359
I was you'd kind of walk to walk into the XL and you'd get collared 25 times within about a hundred metres of people that you know, and then you go into the event and it just gets getting more and more and more.
00:27:09.440 --> 00:27:12.960
And I think this industry is is mainstream now, isn't it?
00:27:13.039 --> 00:27:15.279
There's no there's no hiding away anymore.
00:27:15.519 --> 00:27:16.400
Yeah, right.
00:27:16.640 --> 00:27:24.720
Yeah, it's yeah, I mean, seeing data centers being I say that as someone as if I'm like someone who understands how this industry works, right?
00:27:24.799 --> 00:27:26.960
I've been in here for about a year, to be honest.
00:27:27.279 --> 00:27:35.359
Um, but public conversation around it is something that does seem to have picked up, even in just the past 12 months, right?
00:27:35.440 --> 00:27:44.799
So I'm sure you've seen the the long tail of the growth and how we've kind of ended up in this position, but it does seem like things have really accelerated over the course of the past year.
00:27:44.960 --> 00:27:54.559
Now, coming from the crypto industry, the fact that everyone is talking about it now does make me a little bit worried because that tends to imply that there's a top signal and everything's about to fall apart.
00:27:54.799 --> 00:27:57.279
But hey, maybe this time is different.
00:27:57.519 --> 00:27:58.319
Yeah, exactly.
00:27:58.480 --> 00:27:59.119
You just don't know.
00:27:59.200 --> 00:28:05.359
But I do I just found it yeah, really I was a couple of colleagues there and I was just saying, you know, how how busy it is now.
00:28:05.440 --> 00:28:10.720
And we held a data center marketing club event the night before the actual thing.
00:28:10.799 --> 00:28:17.839
And even that, you know, we had I don't know how many people like just did it off the cuff, like 60 to 100 people turn up and talk about marketing.
00:28:17.920 --> 00:28:30.559
And it's just crazy now you you can get all these people in a room from just little silos within the industry all all trying to understand where we're at and what's happening because the change is so quick, and I think that's um that's what it is.
00:28:30.640 --> 00:28:41.200
It it changes before your eyes, but you don't really see it, you don't notice it until you're at these points where you're like, I was here last year and it wasn't this busy, or I wasn't having this many conversations last year.
00:28:41.279 --> 00:28:45.039
And I I just find it find it fascinating when you go to the major events.
00:28:45.119 --> 00:28:49.599
And I know I've I know people that have been obviously I went to Kickstar this year, it was the busiest year they've ever had.
00:28:49.759 --> 00:28:51.279
I'm sure MIPIM was crazy.
00:28:51.359 --> 00:28:52.960
I didn't go, but I'm sure it was crazy.
00:28:53.039 --> 00:29:02.799
Obviously, Europe, DTC, you've got Data Center Nation in Milan, I think, this week, which I'm sure will be a busy event because Milan is off the scale with regards to investments.
00:29:02.960 --> 00:29:09.359
You've got all the platform events that are coming up, you've got Data Cloud Can, who I know are planning for a record event.
00:29:09.519 --> 00:29:10.799
It just goes on and on and on.
00:29:11.200 --> 00:29:12.720
And then you add in the US as well.
00:29:12.799 --> 00:29:15.920
You know, I haven't been out to the US for a while, but you add in all the events out there.
00:29:16.160 --> 00:29:20.640
DCD New York was last week, spoke to a couple of people who just said it was the busiest it's ever been.
00:29:20.720 --> 00:29:23.839
So it's just um just getting busier and busier and busier.
00:29:24.079 --> 00:29:31.519
And I think it's um it's important that we we you know remain in the industry and focused on what we're doing because there is so much noise now, so much happening outside.
00:29:32.000 --> 00:29:33.359
Yeah, yeah, massively.
00:29:33.440 --> 00:29:39.119
So and I think the increased public attention is good on the one hand, right?
00:29:39.279 --> 00:29:42.880
Because it means that people are talking about it and more people are aware of the problems, right?
00:29:43.039 --> 00:29:45.839
So power availability, for example, right?
00:29:45.920 --> 00:29:50.079
That's now a very much mainstream conversation, especially in the UK.
00:29:50.400 --> 00:29:56.880
Um, but then on the other hand, it does make it more difficult to try and keep up with what's actually going on, right?
00:29:57.039 --> 00:30:14.000
As opposed to having to filter through secondhand, third hand sources of information to try and figure out, right, what's the announcement, what was the announcement of the announcement, what act what's actually important, what isn't important, what's some really clever PR that doesn't actually mean anything.
00:30:14.079 --> 00:30:16.240
Um, and that's difficult at the best of times.
00:30:16.319 --> 00:30:19.599
Now it's feels more challenging than it was.
00:30:20.240 --> 00:30:21.039
Yeah, no, totally.
00:30:22.799 --> 00:30:24.559
Trying to write a weekly newsletter.
00:30:24.720 --> 00:30:25.119
Yeah, right.
00:30:25.839 --> 00:30:27.039
What do you what do you put in it?
00:30:27.200 --> 00:30:28.240
What do you choose?
00:30:28.559 --> 00:30:29.039
Exactly.
00:30:29.200 --> 00:30:40.960
It's very enjoyable, and you know, I'm not complaining by any stretch of um both of us talking for Ben, but I'm sure he's as amazed by the success of his as I am of mine and love the fact that it adds so much value.
00:30:41.119 --> 00:30:57.279
But yeah, it gets diff more difficult every every say few months to kind of collate all that information because there's so much noise now and news is repeated, regurgitated, you know, people will release a press release three or four times in a year, which I'm sure a lot of people don't even see, but it happens quite a lot.
00:30:57.440 --> 00:31:00.559
And I have to go back and check have I have I said this before?
00:31:01.119 --> 00:31:05.359
Is this a deja vu moment, or or have I actually written this before?
00:31:05.440 --> 00:31:08.480
And a lot of the time you have written it before, so yeah, it's uh difficult.
00:31:08.559 --> 00:31:15.279
So if any of the PR companies out there could um plot try and keep it aligned, you know, do it do a better job, we'd be uh much appreciated.
00:31:16.640 --> 00:31:17.920
Make our lives a little bit easier.
00:31:18.160 --> 00:31:18.720
Yeah, exactly.
00:31:19.039 --> 00:31:23.039
Feel free to send us any press releases because it means we don't have to uh don't have to dig around for it.
00:31:23.440 --> 00:31:24.720
Yeah, legit.
00:31:25.519 --> 00:31:26.880
No, yeah, I'm definitely serious.
00:31:27.839 --> 00:31:28.319
Yeah, yeah.
00:31:29.039 --> 00:31:31.279
I'd both file email addresses to the show notes.
00:31:31.359 --> 00:31:42.720
Um the other thing I wanted to shout out was um shout out to Stephen Beard and C Lau, who are of night Frank, I don't know if everyone would have seen this, but they've just run 600 kilometres in 10 days.
00:31:42.960 --> 00:31:46.400
Seven UAU states um for autism awareness.
00:31:46.480 --> 00:31:56.319
Um Stephen's child um you know has has autism, and Steve's trying to raise a million dollars for an app to you know help parents and and children that are in a similar situation.
00:31:56.480 --> 00:31:59.440
So I want to shout out like ridiculous achievement.
00:31:59.599 --> 00:32:02.559
I think um sometimes these things go under the radar.
00:32:02.640 --> 00:32:04.079
Um so I just wanted to give it a shout out.
00:32:04.160 --> 00:32:07.680
I'll put a GoFundMe in the show notes of this episode.
00:32:07.839 --> 00:32:10.000
So if you can support, please support.
00:32:10.079 --> 00:32:18.000
And if you can't, you know, just um give it a repost and share a bit of awareness for for what they've done because it's ridiculous to be honest.
00:32:18.079 --> 00:32:22.880
You know, I couldn't even run 600 kilometres in 10 years, they've done it in 10 days.
00:32:24.079 --> 00:32:29.039
Yeah, I mean, ultramarathons are a different kind of experience.
00:32:29.599 --> 00:32:31.759
Um, yeah, I did see that actually.
00:32:31.920 --> 00:32:36.640
Um what an achievement, especially in that part of the world as well.
00:32:36.720 --> 00:32:47.200
Like from a weather standpoint, it's not necessarily conducive to a pleasant experience, and running that kind of distance isn't pleasant.
00:32:47.440 --> 00:32:56.640
Um, I've done what I've done before, I've done 80 kilometers or something over the course of two days, uh, and that sucked.
00:32:56.960 --> 00:33:00.079
So 600 over the course of what, 10?
00:33:00.559 --> 00:33:02.480
Fair play, fair play to them.
00:33:02.799 --> 00:33:03.440
Yeah, exactly.
00:33:04.160 --> 00:33:06.559
And again, like always give causes a shout-out.
00:33:06.640 --> 00:33:18.240
I know Data Centre Cycling Club have got the event coming up where they cycle to CAN every year, which I think, or I feel like they're doing it before or after this year, but you'll find the website Data Centre Cycling Club, go and have a look, six days of cycling.
00:33:18.319 --> 00:33:28.240
But there's a lot of good causes out there in the industry, a lot of great people doing great things for for causes that are outside of the industry, not related to the industry, but they're by people in the industry.
00:33:28.319 --> 00:33:34.720
So, you know, if anyone's if anyone's seeing anything or doing anything, I'll tell you what, if anyone's doing anything, come on this with me and Ben.
00:33:34.799 --> 00:33:39.440
We'll give you, we'll give you the uh we'll give you five minutes to have a little little chat about what you're doing.
00:33:39.519 --> 00:33:42.799
So I um I'd love to give a shout out to anything like that.
00:33:43.519 --> 00:33:44.559
Anything else, Ben?
00:33:44.640 --> 00:33:50.000
I know we think we've just covered six weeks of news in the covered quite a bit, haven't we?
00:33:50.319 --> 00:33:50.720
I think so.
00:33:50.799 --> 00:33:53.279
I think what'll be interesting, I think, to hear from your side.
00:33:53.599 --> 00:33:59.839
Obviously, for me, with things like the 600 kilowatt racks that are on the horizon with the the Rubin Ultras, right?
00:34:02.559 --> 00:34:04.640
Two things, well, two other things that stood out.
00:34:04.799 --> 00:34:08.960
One, Jensen's comments in the keynote around Hopper being dead, right?
00:34:09.119 --> 00:34:14.079
Don't buy Hopper, um, and then branding himself as the chief revenue destroyer.
00:34:14.559 --> 00:34:26.079
Um that doesn't sound like a particularly easy place to be if you've just placed a very large order for what is now effectively being termed as dead kit, that doesn't make life particularly easy.
00:34:26.960 --> 00:34:38.320
What was the reaction that you saw from 600 kilowatts plus racks being legitimately positioned as what's coming next?
00:34:38.559 --> 00:34:42.639
Like what's the what's the reaction being from outside of the rack?
00:34:42.960 --> 00:34:45.920
Um how does one cope with it?
00:34:46.800 --> 00:34:56.880
I think people have been talking about it for a while, not specifically obviously the 600 kilowatt announcement, but the density and the challenge that that creates.
00:34:57.039 --> 00:35:09.280
I think um the the people I always try and talk about, the CTOs of this world who are you know involved in the the technical and the design and you know the future proofing of these facilities are in a very, very difficult spot.
00:35:09.440 --> 00:35:18.239
And I think it's I would say there's there's not I didn't have any specific like calls around it going, oh you know, Andy, have you heard the things untrue?
00:35:18.559 --> 00:35:21.920
Because I think everybody kind of knew that it's going in that direction.
00:35:22.000 --> 00:35:27.280
It's just a case of how how high and you know, where does it end, I guess, is the challenge.
00:35:27.440 --> 00:35:35.280
And ensuring that your developments and your data centers are designed for a point in the future and not for today.
00:35:35.440 --> 00:35:41.360
It's absolutely pointless designing a data center for whatever kit you have today.
00:35:41.440 --> 00:35:42.800
I think that's the biggest challenge.
00:35:42.880 --> 00:35:48.639
And and I think that's why we're seeing that, like I said earlier, that uh analysis of CapEx.
00:35:48.960 --> 00:35:52.559
I don't think it's like we don't need capacity anymore.
00:35:53.039 --> 00:35:58.159
I think it's more what on earth do we need our facility to look like?
00:35:58.559 --> 00:36:00.159
Where do we need capacity?
00:36:00.320 --> 00:36:03.039
What you know, who is going to be the customer?
00:36:03.280 --> 00:36:05.039
I think they're all the questions they're asking.
00:36:05.199 --> 00:36:24.000
There'll be some there'll be some casualties, I would say, of data centers that have been recently launched with a certain infrastructure slash kit inside them, and how long is that actually fit for purpose?
00:36:24.159 --> 00:36:33.119
I think that'll I'm talking reading between lines, no one said anything to me, but I think that will cause some people some problems, would be my my view.
00:36:33.280 --> 00:36:37.760
But yeah, the the density, scale, everything has been the conversation for ages.
00:36:37.840 --> 00:36:42.559
Yeah, most of the podcasts I talk to, when I talk to anybody that's technical, that's where you end up.
00:36:42.800 --> 00:36:44.400
But no one knows the answer, do they?
00:36:44.639 --> 00:36:48.400
Jensen's obviously going to say whatever he's doing is the right thing to do, right?
00:36:48.480 --> 00:36:51.280
Yeah, and I get that as well, and I think a lot of people see that.
00:36:51.360 --> 00:36:58.159
It's um that's why I said earlier, Jensen's one off some, but um, yeah, it's it's fascinating where it's going.
00:36:58.239 --> 00:37:02.320
I I personally don't think anybody really knows.
00:37:02.480 --> 00:37:09.039
And I I I think I did a post on LinkedIn not long ago basically saying that you know anyone who gives you the answer probably doesn't know the answer.
00:37:09.199 --> 00:37:10.800
Because I I just don't think there is one.
00:37:10.880 --> 00:37:15.199
You can have assumptions and you can have direction, you can have education and guesses and viewpoints.
00:37:15.519 --> 00:37:20.079
If there's a definitive answer, we'd all be lumping our money somewhere, wouldn't we?
00:37:20.159 --> 00:37:21.920
And you've only got to look at the financial markets.
00:37:22.000 --> 00:37:29.039
If like there's a lot of uncertainty with regards to which stock's gonna go, which one's gonna win, who's gonna do best out of this.
00:37:29.119 --> 00:37:34.800
And you know, capital normally gives you the um the greatest barometer of direction, I think.
00:37:35.039 --> 00:37:37.280
Follow the follow the money, as they say.
00:37:38.000 --> 00:37:42.400
Yeah, except when it's deep seek, because then apparently the market gets it wrong.
00:37:42.480 --> 00:37:44.880
Right to quote Jensen, that's not me, that's not my perspective.
00:37:45.039 --> 00:37:47.519
To quote Jensen, then the market gets it wrong.
00:37:47.760 --> 00:37:51.519
And that's when you get into the geopolitical market, though, isn't it?
00:37:51.599 --> 00:37:52.880
It's like again, we don't know.
00:37:52.960 --> 00:37:53.599
We just don't know.
00:37:53.679 --> 00:37:57.599
You have to just look at it and go make a deck, well, I'll make an educated decision myself.
00:37:57.840 --> 00:38:02.400
But I understand that politically they're gonna say something.
00:38:02.880 --> 00:38:03.360
Yeah.
00:38:03.599 --> 00:38:05.280
That's all I can say on that.
00:38:05.599 --> 00:38:06.159
Yeah.
00:38:06.480 --> 00:38:08.159
Okay, so follow-on question then.
00:38:08.320 --> 00:38:19.920
If the the power is going in one direction, the rack density is going in one direction, occasionally you'll see like an announcement, new data center facility, 10 to 15 megawatts of power, 45,000 square foot facility, right?
00:38:20.079 --> 00:38:32.239
If you've got 600 kilowatt racks that they're rated for and they'll have probably 50% more power draw in the burn-in phase, that is 10 racks in a 45,000 square foot facility.
00:38:32.480 --> 00:38:34.400
What do you do with the rest of that space?
00:38:34.559 --> 00:38:41.440
Because that to me as an outsider who's trying to come in and understand how things are working, that just doesn't seem like the way to build things anymore.
00:38:41.519 --> 00:38:43.280
It just doesn't seem to make sense.
00:38:43.599 --> 00:38:44.400
No, I agreed.
00:38:44.480 --> 00:38:51.119
And again, I had a conversation with someone, a genuine conversation with someone not long ago about five-side football pictures and go-kart tracks.
00:38:51.519 --> 00:38:51.840
Right.
00:38:52.000 --> 00:38:54.880
No, and I and it wasn't, you know, it wasn't a joke.
00:38:54.960 --> 00:39:11.119
It was a genuine, like, you know, what there's gonna be so much empty space, I think, um, and obviously a limitation of power, which means you can't just say, okay, we'll just put more racks in it, because you haven't got the power to switch them on, for one of putting it in a more simple term.
00:39:11.360 --> 00:39:19.920
So I'd I I'm with you on this, and I think that is part of the challenges with where to put your money and how to build.
00:39:20.079 --> 00:39:38.000
I think it's um you can obviously identify those sites that have significant power availability and kind of build your first phase, knowing that you've got, say, another three, four hundred meg available to you when you need it, but you might not need the building, you might just need the power.
00:39:38.159 --> 00:39:51.280
But for the legacy data centers, which is probably the ones that you're more we're more talking about, the ones that are there and they're already, you know, 100,000 square metres or whatever they might be, I don't know.
00:39:51.440 --> 00:39:56.639
It will they reduce the size of it and offset part of the land?
00:39:56.719 --> 00:39:58.239
I I genuinely have no idea.
00:39:58.320 --> 00:40:00.239
Like, I'm serious with the five-side football pictures.
00:40:00.320 --> 00:40:03.440
I think it's uh, you know, it's a definite possibility.
00:40:03.519 --> 00:40:05.440
I think there is one somewhere with a go-kart track in it.
00:40:05.519 --> 00:40:06.159
I might have made that up.
00:40:06.239 --> 00:40:07.599
I'm pretty sure there is a data center somewhere.
00:40:07.840 --> 00:40:08.400
That's fantastic.
00:40:08.480 --> 00:40:10.480
If there, if there isn't one, that sounds like an opportunity.
00:40:10.719 --> 00:40:12.159
Yeah, but it I don't know.
00:40:12.320 --> 00:40:14.639
And I and it is a conversation that happens.
00:40:14.880 --> 00:40:19.679
I you know, I talk to people about, and again, like, is that I think that's a period we're in.
00:40:19.760 --> 00:40:20.400
I genuinely do.
00:40:20.480 --> 00:40:25.519
Yeah, when Jensen says an inflection point, I think the industry's at an inflection point as well.
00:40:25.599 --> 00:40:28.239
It's not just AI that's at an inflection point.
00:40:28.320 --> 00:40:30.559
It's like, what do we do?
00:40:30.960 --> 00:40:46.159
I wish I knew the answer, but yeah, again, yeah, there's there's probably someone listening that's got a better educated guess of where it'll go than I have, but I I do believe that nobody has a degree of certainty in their in their viewpoint.
00:40:46.559 --> 00:41:14.639
Yeah, yeah, and I think like the the reduction in footprint size does kind of rest on the assumption that ultimately NVIDIA's way of doing things right now is the right way to do things, and the only way to do things right, which is more power, more rack density, um, more powerful hardware off the back of it, and with some of the emerging semiconductor manufacturers, they are showing that there is potentially a different way of doing things.
00:41:14.800 --> 00:41:28.079
Now, how much success they end up having, um, that kind of remains to be seen because with the the Kudo ecosystem, I mean NVIDIA's got a pretty compelling competitive advantage there, right?
00:41:28.239 --> 00:41:38.480
Um, coupled with the fact that they've open sourced a load of the new software off the back of GTC2, like Dynamo, for example, for orchestrating large-scale clusters for inference workloads.
00:41:38.800 --> 00:41:47.519
But then if you look at what's also happened in the semiconductor space recently, I think the most interesting one is Furiosa AI.
00:41:47.760 --> 00:42:01.039
So if anyone's not familiar, they're a South Korean semiconductor manufacturer focused specifically on inference hardware, and they rejected, or apparently rejected, an$800 million bid from Meta to buy them out recently.
00:42:01.280 --> 00:42:07.679
Um that's quite interesting because I mean that's backing yourself to a pretty significant degree.
00:42:07.760 --> 00:42:11.360
I think saying no to 800 mil fair play, if it was said.
00:42:11.760 --> 00:42:16.559
Um but if that's the right business decision, what do they see in the future?
00:42:16.639 --> 00:42:28.559
And their cards are interesting because they're 150 watt cards, and a server gets potentially double the token throughput for say Lava 3.1, um, than the AMD hardware, again, at a significantly lower price point, right?
00:42:28.639 --> 00:42:35.840
So for power-constrained markets or older facilities, for example, um, that's potentially an interesting solution.
00:42:36.079 --> 00:42:45.920
Now, the emerging semiconductors, I mean, even AMD to a certain extent, they're kind of struggling to gain market share at the moment.
00:42:46.000 --> 00:42:52.400
But then when we do see larger scale adoption, I think the market's going to be very, very large, right?
00:42:52.559 --> 00:42:54.960
And there's it's a new market at the end of the day.
00:42:55.039 --> 00:43:00.320
Uh, and there's always space for new players and new solutions in new markets.
00:43:00.480 --> 00:43:07.360
And at the end, like the massive amount of floor space and the power constraints is a problem to be solved, right?
00:43:07.599 --> 00:43:12.559
So whoever can solve that particular problem, that's potentially something quite interesting.
00:43:12.639 --> 00:43:18.480
So I think uh if you do have a 45,000 square foot facility and you're worried about only having 10 racks in that, I wouldn't worry too much.
00:43:18.559 --> 00:43:23.840
There's probably something else that we just don't know about yet that's gonna appear out of nowhere and provide you with a solution, hopefully.
00:43:24.239 --> 00:43:26.480
Yeah, something will change the game, it always does.
00:43:26.559 --> 00:43:28.800
And yeah, gives us something to talk about anyway, doesn't it?
00:43:29.119 --> 00:43:31.199
If it wasn't for that, we wouldn't be talking now.
00:43:31.360 --> 00:43:33.360
So appreciate all this uncertainty.
00:43:33.519 --> 00:43:35.280
It gives us something to chat about.
00:43:35.599 --> 00:43:36.719
I think that's all.
00:43:36.880 --> 00:43:40.880
Um, all I was gonna say is obviously this is the second episode we're done.
00:43:40.960 --> 00:43:41.679
We are gonna do it.
00:43:41.760 --> 00:43:43.760
Try and get to once a month, hopefully.
00:43:43.840 --> 00:43:45.920
But if not, we'll stick at six weeks.
00:43:46.320 --> 00:43:47.119
We're doing our best.
00:43:47.280 --> 00:43:56.159
Um, if anyone wants to come on that's listening, and you've got if any news breaks in the next month or six weeks, depending on when we record and you want to come on and talk about it, please do.
00:43:56.239 --> 00:44:00.880
I'm I'm I'm all up for getting some other opinions um of what's happening in the market.
00:44:00.960 --> 00:44:02.960
We'll always give you our opinion and what we're seeing.
00:44:03.039 --> 00:44:05.199
But if anybody else wants to join us, you know, please do.
00:44:05.280 --> 00:44:20.559
And like I said earlier, if anyone's doing anything um a bit different, you know, in the industry, charitable or events or whatever it might be, you know, happy to happy to give you a plug and if it if it helps raise some money for somebody, all the better.
00:44:20.960 --> 00:44:21.920
We better get on.
00:44:22.480 --> 00:44:23.920
We've got newsletters to write.
00:44:24.400 --> 00:44:26.000
Got newsletters to write, yeah, right.
00:44:26.239 --> 00:44:27.679
It's only Monday when we're recording it.
00:44:27.840 --> 00:44:29.199
Probably release this on Friday.
00:44:29.280 --> 00:44:33.920
So hopefully our uh our newsletters won't be far away by the time by the time you hear this.
00:44:34.000 --> 00:44:36.960
But I'll put the GoFundMe in the show notes.
00:44:37.039 --> 00:44:40.719
Um, obviously, if you need myself or Ben, we're pretty easy to find.
00:44:40.800 --> 00:44:41.920
Just come and find us.
00:44:42.000 --> 00:44:42.960
But thanks for your time.
00:44:43.119 --> 00:44:44.320
Ben, good to catch up again.
00:44:44.400 --> 00:44:45.760
Now you're back in the UK.
00:44:46.079 --> 00:44:46.960
Likewise, Andy.
00:44:47.039 --> 00:44:47.280
Yep.
00:44:47.360 --> 00:44:48.559
One final question for you.
00:44:48.639 --> 00:44:51.599
Um, hopefully for in a month's time, predictions.
00:44:51.920 --> 00:45:00.719
Based on what you're seeing right now and where there's a little bit of interest, um, what do you reckon we might end up potentially be talking about next month?
00:45:01.119 --> 00:45:02.159
Prediction time.
00:45:02.239 --> 00:45:04.239
Uh, you could have given me a heads up on this one.
00:45:04.320 --> 00:45:24.719
I think we're gonna see a bit more of this uncertainty with CapEx, would be my I think we might see a few more, whether it's analysts or news articles that break about projects being, let's say, delayed or re-evaluated, I would guess in April.
00:45:24.800 --> 00:45:26.079
It's April Fool's Day tomorrow.
00:45:26.239 --> 00:45:29.599
So from April, I would say my that'll be my prediction.
00:45:29.760 --> 00:45:30.559
What about you?
00:45:30.880 --> 00:45:31.280
Cool.
00:45:31.760 --> 00:45:35.039
I think more discussion around sovereign AI.
00:45:35.199 --> 00:45:42.320
I think that's going to become more of a talking point that's gonna slowly work its way into the mainstream a little bit more.
00:45:42.719 --> 00:45:45.599
Um there you go.
00:45:45.679 --> 00:45:48.800
We'll record the next episode when both of those predictions come true.
00:45:48.960 --> 00:45:53.039
So it might be in four weeks, it might see it again in two years.
00:45:53.119 --> 00:45:54.719
But anyway, thanks for listening, everyone.
00:45:54.800 --> 00:45:57.280
We'll do it again in uh in the next month or so.
00:00:00.560 --> 00:00:03.839
Next, it's the installed data center podcast.
00:00:23.199 --> 00:00:25.679
It is your host, Andy David.
00:00:28.079 --> 00:00:30.399
Welcome to the Inside Data Center Podcast.
00:00:30.559 --> 00:00:37.920
We are back again, joined by Ben Vallieri for the second edition of Data Center News and GPUs.
00:00:38.159 --> 00:00:39.119
Afternoon, Ben.
00:00:40.159 --> 00:00:41.280
Good afternoon, Andy.
00:00:41.359 --> 00:00:44.000
Yeah, we've uh it's been a little while since the last one, isn't it?
00:00:44.240 --> 00:00:45.920
What's happened in the past couple of months?
00:00:46.240 --> 00:00:46.640
Exactly.
00:00:46.719 --> 00:00:47.039
I know.
00:00:47.119 --> 00:00:48.159
We said we do one a month.
00:00:48.240 --> 00:00:54.079
We're on about one every six weeks at the moment, which is uh make making it difficult to catch up with the news, but there you go.
00:00:54.159 --> 00:00:57.679
But you know, you've been out at um GTC and obviously we've been busy travelling around.
00:00:57.759 --> 00:00:59.840
So but still, good to get it in.
00:00:59.920 --> 00:01:02.719
We're still stuck with the name, Data Center News and GPUs.
00:01:02.880 --> 00:01:07.200
Haven't had any suggestions for better ones, so that's what we're uh what we're going with now.
00:01:07.359 --> 00:01:18.640
We're gonna give you another whistle stop run through of some of the biggest news or the most traction news that we've had, um whether it's GPU or or data center related.
00:01:18.959 --> 00:01:20.400
So now we've started this, Ben.
00:01:20.480 --> 00:01:21.040
Do you want to start?
00:01:21.120 --> 00:01:24.319
Ben's fresh off a plane from the US, so there you go.
00:01:24.959 --> 00:01:27.280
Yeah, the uh the joys of a Heathrow fire.
00:01:27.359 --> 00:01:31.680
Um I think the the biggest one has got to be GTC, um, to be honest.
00:01:31.920 --> 00:01:34.959
The GTC was super interesting.
00:01:35.040 --> 00:01:36.959
Um, I was out there for the week.
00:01:37.280 --> 00:01:53.439
The keynote, I mean, aside from feeling like we're in a bit of a cult, to be honest, because when you're queuing at half past seven in the morning for a 20,000-person stadium and you're a long way at the back of the queue, and then it's sold out and people are being turned away to see a tech CEO speak.
00:01:53.599 --> 00:02:00.719
I mean, it's it's a little bit culty, but on the other hand, super, super interesting based on the content.
00:02:00.799 --> 00:02:04.400
And two things really stood out to me from it.
00:02:04.719 --> 00:02:12.560
One being there was no mention of training anywhere in the keynote, no mention of model training at all.
00:02:12.639 --> 00:02:50.479
Um, the focus was almost entirely on inference, um, and this arbitrary measure that Jensen seems to have introduced for the AI factory concept, which is tokens per second per megawatt, which I think is quite an interesting um nod as to where NVIDIA seems to be trying to guide the market, um, which from where I'm sat and from the massive amount of financial modelling work that have done over the course of the past six months, months or so, makes perfect sense because training, you're constrained really, in terms of how much revenue you can make.
00:02:50.560 --> 00:02:51.759
It's a race to the bottom.
00:02:51.919 --> 00:03:09.439
Someone is always going to be able to undercut you on GPU per hour charges, but inference, where you've got the Xerox model, where it's copies basically, you deploy the infrastructure, you get your customers to come in, um, and their charges are coming out of OpEx, the cash flow profiles are fundamentally different.
00:03:09.599 --> 00:03:14.000
So, from a hardware owner standpoint, really interesting that that's where things are going.
00:03:14.639 --> 00:03:27.919
From a business standpoint as well, and then actually creating enterprise value with all of this wonderful training that has been done and these models that are now reaching these lofty heights by various benchmarks.
00:03:28.960 --> 00:03:31.199
Inference is where the value is going to be created.
00:03:31.280 --> 00:03:36.719
Now, what that value creation looks like, I think that's still the question that needs to be asked.
00:03:37.039 --> 00:03:48.319
But it seems like that's where the industry is going to be going probably pretty quickly over the course of the past over the past next 12 months or so.
00:03:48.400 --> 00:03:51.759
Um, which, as you know, how this industry moves, right?
00:03:51.840 --> 00:03:54.479
It's like dog years, 12 months is going to be seven years.
00:03:54.560 --> 00:03:56.319
Um, so we'll see where we end up.
00:03:56.800 --> 00:03:57.520
Yeah, yeah.
00:03:57.599 --> 00:03:58.800
I watched the Jensen keynote.
00:03:58.879 --> 00:04:08.960
Obviously, I wasn't, I wasn't there, I wasn't in the queue, but yeah, he he kept talking about it being an inflection point, didn't he, for AI and trillion dollar data center industry, et cetera, et cetera.
00:04:09.039 --> 00:04:18.560
And then since then you've seen quite a lot of shifts in um the hyperscalers kind of strategy, I guess, is for want of putting it a better way.
00:04:18.639 --> 00:04:26.079
So there's definitely a a period of um thought going on at the moment as to what's next, what do we do next?
00:04:26.160 --> 00:04:32.800
You know, what I think someone said the other day, you don't want to build a data center that is totally useless because that's kind of where they're getting to now.
00:04:32.959 --> 00:04:35.279
So I think inflection point is a good way to look at it.
00:04:35.360 --> 00:04:43.839
And like you say, moving away from training and moving into the actual can we make money out of AI, I suppose, is is where most people are looking at it now.
00:04:43.920 --> 00:04:44.720
Like, where's the money?
00:04:44.800 --> 00:04:45.439
Where's the cash?
00:04:45.519 --> 00:04:46.720
Where's the return of investment?
00:04:46.800 --> 00:04:48.240
Because it's a big investment.
00:04:48.720 --> 00:04:49.600
Yeah, absolutely.
00:04:49.680 --> 00:04:53.360
And I think that Deep Seek has been really interesting in that regard.
00:04:53.439 --> 00:05:00.240
Um, because obviously the market reaction was everything's fallen apart, um, that's it, it's the end of days.
00:05:00.399 --> 00:05:25.360
And then from what was said in the keynote, because Deep Seek and the reasoning models that have proliferated off the back of it due to the fact that DeepSeek open source an awful lot of their work, reasoning models chew through a huge amount of tokens, um, which again, from a hardware owner's perspective, is super interesting.
00:05:25.600 --> 00:05:39.920
Um, I think the statistic that Jensen used was it requires like a hundred times more compute than expected, um, which is great from both the data center standpoint and also from a compute deployment standpoint, because we're gonna need a hell of a lot more capacity.
00:05:40.160 --> 00:05:45.519
But again, that then feeds into it's all well and good needing the capacity, but what is that capacity going to be used for?
00:05:45.759 --> 00:05:54.800
So I think use cases um are gonna be the uh the most important thing going forward, especially the second part of the keynote.
00:05:55.040 --> 00:05:59.920
Um if you are having to deploy 600 kilowatt racks or above, right?
00:06:00.160 --> 00:06:03.839
Which from a data center standpoint sounds like a bit of a headache.
00:06:03.920 --> 00:06:06.879
Um, I'm sure I'm sure you'll have some perspective there.
00:06:07.680 --> 00:06:08.879
Yeah, I think it's crazy.
00:06:09.519 --> 00:06:15.120
The the rack side of it, from a non-technical, it's just crazy where it's going and where it can get to.
00:06:15.199 --> 00:06:18.720
And I think it's one of the biggest challenges, it's what density, where does it end?
00:06:18.800 --> 00:06:20.879
You know, what is the density we're we're building for?
00:06:20.959 --> 00:06:27.120
And I think that is why we've seen a lot of this, as I say, like fluctuation in the markets and a bit of uncertainty.
00:06:27.199 --> 00:06:37.279
And you can totally understand why if you're investing tens and tens of billions of dollars into this industry in a period of such uncertainty, it's it is a big gamble.
00:06:37.360 --> 00:06:42.720
And at some point, the money behind it does have to say, hang on a minute, like you know, are we doing the right thing here?
00:06:42.879 --> 00:06:47.279
They might pause and go again, but they need to re-evaluate and ensure they're doing the right thing.
00:06:47.360 --> 00:06:53.360
And you you've kind of got to re-evaluate every every couple of weeks at the moment, and that's again that that is a challenge.
00:06:53.439 --> 00:06:56.959
And you know, NVIDIA are one of others as well, aren't they?
00:06:57.040 --> 00:07:01.439
So there's there's all this research going on into the background, and yeah, it's a it's a crazy time.
00:07:01.519 --> 00:07:05.040
And I I liked the bit on the the humanoid, the the robots as well.
00:07:05.120 --> 00:07:06.079
Like, you know, that was cool.
00:07:06.160 --> 00:07:13.120
And I think when you start talking about data and capacity and demands in that in that area, you you're in another world.
00:07:13.199 --> 00:07:16.480
And I think, yeah, I know that's not here now, but that's coming.
00:07:16.560 --> 00:07:19.680
And every year he does a keynote as a better robot.
00:07:19.759 --> 00:07:23.680
So I don't think it'll be long until those uh those robots are walking among us.
00:07:23.759 --> 00:07:25.279
And if they're not already, you know, who knows?
00:07:25.360 --> 00:07:25.920
They may be already.
00:07:26.160 --> 00:07:26.800
Yeah, exactly.
00:07:26.879 --> 00:07:30.399
I mean, I think the other interesting piece there to that point too.
00:07:30.560 --> 00:07:34.959
One robotics is the the I mean, no one knows what that's going to look like.
00:07:35.040 --> 00:07:39.519
But I mean, if you've spent enough time reading sci-fi, you've got probably got a pretty good understanding of what's going on.
00:07:39.600 --> 00:07:43.519
Didn't he say it'll be yeah, didn't he say it'll be the biggest industry ever?
00:07:43.680 --> 00:07:44.879
Something along those lines.
00:07:45.199 --> 00:07:46.000
Yeah, right.
00:07:46.480 --> 00:07:57.680
Like everything that is everything that moves could be potentially automated, and everything that moves could therefore, by extension, potentially require some sort of thinking per se, right?
00:07:57.759 --> 00:08:06.160
Which means that it might be a model, it might be whatever the next iteration of AI is, and that's probably going to need some compute of some sort.
00:08:06.319 --> 00:08:13.680
Um, I think the other interesting thing though was that there was no real mention of the hyperscalers in like a partnership context, right?
00:08:13.759 --> 00:08:23.199
Because if you look back at previous keynotes, there was generally a nod towards like partnerships with Meta or partnership with AWS or Azure or something along those lines.
00:08:23.360 --> 00:08:25.360
But those guys are making their own chips now, right?
00:08:25.759 --> 00:08:28.160
That's eating into NVIDIA's market share.
00:08:28.399 --> 00:08:39.840
Um Lepton AI was a purchase that NVIDIA is apparently considering that cropped up recently, and Lepton offers GPUs per hour, right?
00:08:40.000 --> 00:08:41.679
Which is kind of cloud service.
00:08:41.919 --> 00:08:55.360
So wouldn't be surprised based on what was not said and the relative degree of complexity that is going to be, I mean, that is required to deploy a 600 kilowatt rack, right?
00:08:55.440 --> 00:09:00.080
But wouldn't be surprised if we start seeing NVIDIA going a little bit more into the cloud side as well.
00:09:00.159 --> 00:09:01.279
Because I mean it makes sense, right?
00:09:01.360 --> 00:09:05.360
If your competitors are building chips, why not compete with them at their game too?
00:09:05.840 --> 00:09:06.720
Yeah, exactly.
00:09:06.879 --> 00:09:10.320
It is a it is a battle for for business at the end of the day, I think.
00:09:10.399 --> 00:09:11.440
You have to remember that.
00:09:11.679 --> 00:09:31.440
And it's similar on the on my side, the biggest news was was the TD Cowan and Microsoft's news of cancelling leases, which ties in again with the moving away from partnerships because it was all linked to the allegedly it was linked to the Open AI partnership, which now we believe OpenAI are going to look to build, develop their own infrastructure, which again makes sense.
00:09:31.519 --> 00:09:34.559
You know, if you want to own the infrastructure, it's probably the right way to do it.
00:09:34.720 --> 00:09:37.279
Whether that's true or not, we'll have to we'll have to wait and see.
00:09:37.360 --> 00:09:40.879
But yeah, Microsoft cancelling um a number of leases.
00:09:41.039 --> 00:09:44.399
I think this last week it was two gigawatt across the US and Europe.
00:09:45.279 --> 00:09:53.919
Previously, we saw projects start re-evaluation of strategy, and we know there's a strategy review happening here, but again, it ties into that.
00:09:54.240 --> 00:09:55.279
What is the strategy?
00:09:55.360 --> 00:09:59.919
And I don't mean with Microsoft, I mean generally, like what do we do going forward here?
00:10:00.080 --> 00:10:02.320
How how do we capitalise on this?
00:10:02.480 --> 00:10:04.240
Are we moving away from training?
00:10:04.320 --> 00:10:07.759
You know, are these data centers being built for the right purposes?
00:10:07.840 --> 00:10:09.919
Are they in the right places for for what they require?
00:10:10.080 --> 00:10:18.480
I'm sure there's a big review going on somewhere with a lot a lot more intelligent people than myself looking into it and trying to work out where best to spend this capex.
00:10:18.720 --> 00:10:28.799
I never see it as a as an industry challenge, I just see it as a another re-evaluation point where everyone's kind of like, oh no, you know, but what do we do?
00:10:28.960 --> 00:10:32.480
We've just announced$100 billion of CapEx for 2025.
00:10:32.639 --> 00:10:35.279
Are we spending$100 billion in the right place?
00:10:35.440 --> 00:10:37.919
And I think we will see that probably across this year.
00:10:38.000 --> 00:10:40.720
I think we'll continue to see it with um some of the other hyperscalers as well.
00:10:40.960 --> 00:10:46.240
Won't just be uh Microsoft that are that are reviewing where and when to spend.
00:10:46.799 --> 00:10:47.600
Oh yeah, definitely.
00:10:47.679 --> 00:11:00.320
And I I think to the um the Microsoft point with OpenAI, um OpenAI have also just um announced a partnership with or a very large deal with Courtweave, right?
00:11:00.399 --> 00:11:02.639
Who've also just very recently gone public too.
00:11:02.720 --> 00:11:08.159
And I think that amounted to like$12 billion or something, which is absolutely astronomical.
00:11:08.639 --> 00:11:24.960
Um that is kind of from my perspective and from what I'm seeing with other people, contingent on OpenAI being able to successfully close their next funding round, which apparently they're very close to doing with SoftBank it being like a 40 billion round or something absurd like that.
00:11:25.039 --> 00:11:31.120
But then that's also contingent on OpenAI being able to transition to a fully full profit model as well, right?
00:11:31.360 --> 00:11:39.440
And whether or not they're going to be able to do that from a legal challenge standpoint kind of remains to be seen.
00:11:39.679 --> 00:11:50.480
Um, and then also the two gigawatts of cancellations I've read this morning that while Microsoft has pulled back on those, apparently those same sites were scooped up by Meta and Google very, very quickly.
00:11:50.799 --> 00:11:57.759
So I think this kind of taps into what we were speaking about in the the first one that we did of all of those six weeks ago, right?
00:11:57.919 --> 00:12:10.480
In that I think there's a little bit of brinkmanship going on, in that I think things probably started a little bit before people really wanted them to, but you can't not do it in case your competitors have something that you want, right?
00:12:10.559 --> 00:12:15.120
It's better to have something and give it up and pay a little bit of money for it, right?
00:12:15.279 --> 00:12:19.679
As opposed to need something that you can't get because your competition's already got hold of it.
00:12:20.000 --> 00:12:21.440
Yeah, totally, definitely seeing it.
00:12:21.519 --> 00:12:31.759
And yeah, I've had I've had similar as well, which again is why I don't ever see the I I I laugh when I post uh these kind of announcements because you get the panic stations from the from everybody.
00:12:32.399 --> 00:12:40.720
You know, don't read too much into it, is what I always say to people, you know, like do your research, understand the market, talk to people, and you can actually gauge what is actually really happening.
00:12:40.799 --> 00:12:47.840
You'll never see what's really happening in the public, but you know, pick the phone up, have conversations, and you'll and you'll quickly you'll quickly see what's out there.
00:12:48.799 --> 00:12:51.679
Ask people questions and they tend to tell you the answers, right?
00:12:51.919 --> 00:12:52.320
That's it.
00:12:52.480 --> 00:12:53.440
Especially in this industry.
00:12:53.600 --> 00:12:56.240
Can't you can't tell anyone anything, but you can tell anyone everything.
00:12:56.320 --> 00:12:59.120
It's uh look at it.
00:12:59.279 --> 00:13:04.480
Um I also I quickly scanned through some of the projects that broke as well in uh in the last few weeks.
00:13:04.639 --> 00:13:15.200
And again, you know, when you look at investment, you're continuing to see this wave of hundred-acre sites, multi-billion dollar investments, new funds raising.
00:13:15.279 --> 00:13:17.440
It's still happening regularly, regularly.
00:13:17.519 --> 00:13:26.320
So um we had a£1 billion data center and technology part in Anglesey and Wales, which I think I raised a lot of eyebrows with regards to who's the customer for that one.
00:13:26.399 --> 00:13:28.399
You know, it's a it's an interesting location.
00:13:28.639 --> 00:13:38.000
UAE obviously continued to invest significantly, 30 to 50 billion euro AI data center in France, big, big range there.
00:13:38.080 --> 00:13:47.360
Um Stone Shield, 500 megawatt in in Spain, Azora, 2 billion development in Zaragoza against Spain, Evrock, 96 megawatt in France.
00:13:47.600 --> 00:13:55.440
QTS obviously got approval for the big site in the north northeast of England, uh, which is approximately 10 billion dollar investment.
00:13:55.600 --> 00:13:56.799
And it goes on and on.
00:13:56.879 --> 00:14:00.480
So I think you're continuing to see the scale of investments.
00:14:00.559 --> 00:14:20.320
Um, one thing I'm also starting to see now, and or have more conversations with, is actually in the um whether you call it edge or colo or metro edge, whatever you want to call it, but there's certainly a lot of investment now starting to look back at the smaller edge data centers, so five to twenty megawatt city-based.
00:14:20.480 --> 00:14:28.080
I think an area that's been forgotten because you're seeing all these big numbers, but actually is a massive, massive market.
00:14:28.159 --> 00:14:39.120
And I think somebody, some company at some point will take huge advantage of that and you'll see um you'll see somebody do very well out of it while everyone else has got their eyes on the half a gigawatt sites wherever they may be.
00:14:39.360 --> 00:14:40.000
Yeah, right.
00:14:40.240 --> 00:14:46.799
And also like the the absurdity of calling a five to twenty megawatt data center small is kind of funny as well, right?
00:14:47.600 --> 00:14:49.440
I I don't even know what edge is anymore.
00:14:49.519 --> 00:14:57.440
Like I know I used to I remember recording podcasts like three years ago, and I'd ask anyone that said they were at edge what edge means to them, and you've got a different definition every single time.
00:14:57.519 --> 00:14:58.799
And I'm still at that point.
00:14:58.960 --> 00:15:03.919
I don't know what small is, I don't know what edge is, I don't even know what big is anymore.
00:15:04.000 --> 00:15:06.960
It's like what what actually how do we scale any of this?
00:15:07.039 --> 00:15:09.200
It's it's like when people say, What does hyperscale mean?
00:15:09.600 --> 00:15:13.440
Hyperscale is a definition, means like continuing to grow or something along those lines.
00:15:13.519 --> 00:15:19.679
But but actually, we don't ever have any sort of market for these words that we use all the time.
00:15:19.759 --> 00:15:21.039
It's just but there you go, yeah.
00:15:21.120 --> 00:15:33.279
The smaller data centers than the bigger ones, there's definitely a huge uh a huge market for that as you start to join the dots of these major campuses across whatever countries they're in and continents as well.
00:15:33.759 --> 00:15:42.240
Yeah, yeah, and I I think also like the France is an interesting, very, very interesting area at the moment, also from a neo-cloud GPU as a service standpoint.
00:15:42.399 --> 00:15:47.120
Um, one other announcement um to add to your list as well.
00:15:47.360 --> 00:15:51.600
E and I, G42, um, one gigawatt in Italy as well.
00:15:51.840 --> 00:16:01.440
So that I mean the fact that it's like, right, we're gonna have a gigawatt here, we're gonna have a gigawatt there, as opposed to an 100 megawatt or a 200 megawatt, I think that's probably a pretty good indication of the direction of travel.
00:16:01.519 --> 00:16:08.240
But I think France is interesting in that regard specifically, because you've got the the G42 and the AMD deal with Data One, I think it was.
00:16:08.399 --> 00:16:12.559
Yeah, they are um, which is massive in and of itself.
00:16:12.720 --> 00:16:26.399
But then at the same time, same week, this is off the back of the um the AI summit in Paris, Fluid Stack announcing a massive, massive partnership with um Mistral to develop a huge amount of infrastructure in France as well.
00:16:26.559 --> 00:16:29.519
Same with Cesurs as well, another gigawatt facility.
00:16:29.759 --> 00:16:41.360
Um and then ENI, not ENI, sorry, EDF, opening up, I think, two gigawatts of land parcels uh close to various nuclear power generation facilities around the country as well.
00:16:41.519 --> 00:16:58.080
Um France is interesting because if you look at the the energy mix of France relative to every single other European country, aside from the Nordics, they're really the only ones who can support the kinds of developments at the speed that they need to happen right now.
00:16:58.240 --> 00:16:58.879
Why is that?
00:16:59.039 --> 00:17:02.240
They've got a huge amount of nuclear power generation capacity.
00:17:02.399 --> 00:17:16.960
Um, and I think we'll probably see increasing discussion around the need to develop more of that fast in non-traditional markets for nuclear, to be honest, because I mean France has shown that it can work, right?
00:17:17.039 --> 00:17:19.359
They wouldn't be getting these deals if it wasn't viable.
00:17:19.519 --> 00:17:24.559
Um, I wouldn't be surprised if we see some projects rushed through well, relatively speaking, rushed for nuclear.
00:17:24.640 --> 00:17:27.279
So we're probably still, what, five, ten years out, to be honest.
00:17:27.359 --> 00:17:29.440
But I think that's where things are gonna go.
00:17:29.680 --> 00:17:31.519
Yeah, telly house ended last week as well.
00:17:31.599 --> 00:17:41.279
I think a billion euro investment in France, include and it I think it was a uh 90, I might be wrong on this, but around 96 megawatt in Marseille was part of that.
00:17:41.440 --> 00:17:43.200
So yeah, exactly.
00:17:44.480 --> 00:17:47.920
One of the smaller edge facilities in Marseille.
00:17:48.000 --> 00:17:50.079
But Marseille's an interesting again, France is interesting.
00:17:50.240 --> 00:17:51.920
Marseille, a lot of development coming up there.
00:17:52.000 --> 00:17:54.480
Um, it's uh you know, connected very well.
00:17:54.640 --> 00:17:57.440
Obviously, Paris is a big, big challenge for a number of reasons.
00:17:57.680 --> 00:18:00.960
A talent is one of the worst places to to recruit for.
00:18:01.039 --> 00:18:04.799
Um, I can vouch for that with with definite certainty.
00:18:04.880 --> 00:18:06.720
Um, we always try our best, but it's difficult.
00:18:06.799 --> 00:18:10.720
Um, but yeah, southern France, yeah, good, good areas, a lot of a lot of development.
00:18:10.880 --> 00:18:18.480
The other um campus or development that I find quite interesting is old Kevin O'Leary's Wonder Valley in Alberta in Canada.
00:18:18.559 --> 00:18:20.079
I don't know if anyone's seen the video.
00:18:20.240 --> 00:18:22.480
It reminds me of a Centre Parks video.
00:18:22.640 --> 00:18:26.480
Um, every time I watch it, I I see the deers running around in the background.
00:18:26.559 --> 00:18:34.559
I think it's it's like me back at Centre Parks, similar price to a week in centre parks as well, 70 billion dollars to anyone, anyone that's ever been in the UK.
00:18:34.640 --> 00:18:36.000
But it's an interesting development.
00:18:36.160 --> 00:18:44.400
I think he um I think there was a video probably a few weeks ago of him needs to raise the first 10 billion or something for that development.
00:18:44.480 --> 00:18:56.480
But he's um interesting character, he's quite um you know, like for those that that don't know who he is, he's he's kind of similar to Dragon's Den type type character, um, very public, big investor.
00:18:56.640 --> 00:19:00.559
Um, whether it'll happen or not, I think is uh is up for debate, but it's it's worth a look.
00:19:00.640 --> 00:19:01.519
Wonder Valley, it's cool.
00:19:01.599 --> 00:19:03.599
If anyone hasn't seen it, yeah, go and go and watch the video.
00:19:03.680 --> 00:19:06.079
And if you've been centre parts, you'll know exactly what I mean.
00:19:06.160 --> 00:19:10.160
But if he does do it, I'm definitely going because it's uh it looks cool.
00:19:10.400 --> 00:19:11.920
Oh, yeah, it looks super interesting.
00:19:12.000 --> 00:19:16.960
And what's the total power capacities for seven and a half gigs?
00:19:17.200 --> 00:19:18.559
I was gonna say what's stuck in my head.
00:19:18.799 --> 00:19:21.440
Yeah, seven gigs in my ridiculous.
00:19:21.839 --> 00:19:25.680
Um, but then like to that point, it's in Canada, right?
00:19:25.759 --> 00:19:44.079
And they're gonna be targeting US customers, and that's potentially a difficult proposition right now, given uh the ongoing geopolitical struggles, I think would be the uh the most delicate way to put uh a description of the relationship between the US and Canada right now.
00:19:44.319 --> 00:19:47.200
Um I think it's super interesting.
00:19:47.359 --> 00:19:53.759
I think Canada's a really, really interesting proposition at the moment because of the amount of power and the amount of energy that they have available.
00:19:54.319 --> 00:19:59.599
Um it's just if things kind of carry on as they are, who's the off take?
00:19:59.759 --> 00:20:01.359
Because that's a hyperscale facility.
00:20:01.440 --> 00:20:03.279
That has to be a hyperscale facility.
00:20:03.440 --> 00:20:06.000
Um, but all of the hyperscalers are American.
00:20:07.359 --> 00:20:08.880
How do you square that circle, right?
00:20:09.039 --> 00:20:09.440
Exactly.
00:20:09.519 --> 00:20:15.759
And you are seeing you know, geopolitical uh is coming into strategy now, 100% um definitely seeing it.
00:20:15.920 --> 00:20:27.359
Another point I thought of while you were talking is actually, which again, you'll be quite aligned, haven't written it down, but the move away from the likes of Crusoe and Applied and that moving away from the Bitcoin into the into AI infrastructure.
00:20:27.440 --> 00:20:30.799
I think you've seen um obviously it's been happening for a while.
00:20:30.880 --> 00:20:44.319
You know, I've been working with funds that have been doing this for a good few years, but um you've seen a few more public announcements now of we are no longer, you know, in the Bitcoin world, we are now an AI infrastructure developer.
00:20:44.480 --> 00:20:52.880
So that's quite an interesting shift as you start to, especially with the AI, and if we are moving away from learning, etc., where does all that fit in?
00:20:52.960 --> 00:20:57.920
Because a lot of these are huge scale facilities, but in fairly remote locations.
00:20:58.000 --> 00:20:59.920
So it's going to be interesting to see where that market goes.
00:21:00.000 --> 00:21:01.440
But in the US, you're seeing it now.
00:21:01.519 --> 00:21:08.960
There's a very, very clear shift from we're a bit above to we're definitely sitting on the AI side of the fence now.
00:21:09.680 --> 00:21:21.119
Yeah, and I mean, I think that it's an interesting move that makes sense, really, because if you look at what Bitcoin mining is, it's massive amounts of compute deployment incredibly quickly.
00:21:22.000 --> 00:21:27.039
Pretty good experience if you're looking at what's going on in the AI space now, what's required.
00:21:27.200 --> 00:21:32.720
Those facilities tend to have massive power capacity available.
00:21:33.039 --> 00:21:48.960
Um depending on where the where the mining is taking place, that does tend to influence how socially, geopolitically, morally, whatever you want to call it acceptable Bitcoin mining is seen to be.
00:21:49.200 --> 00:21:59.759
Um, I think that perceptions will change because there's some really interesting work going on in the Bitcoin mining space around deployment in support of.
00:22:00.240 --> 00:22:09.839
Renewable energy, for example, like microgrids, for example, and Daniel Batten of C4 Capital does some really, really interesting work in that space.
00:22:10.079 --> 00:22:21.279
Um, but I think the shift away from Bitcoin mining into specific AI compute, GPU compute, AI infrastructure makes sense for a couple of reasons.
00:22:21.440 --> 00:22:28.160
One, as I said before, these companies have the experience in doing it, like um core weave, for example, core scientific.
00:22:29.200 --> 00:22:31.680
It's that's what they were doing first.
00:22:31.759 --> 00:22:34.480
Now they've gone public with Core Weav, right?
00:22:34.799 --> 00:22:42.079
Um Bitdigital has also spanned off a new entity recently, white fiber, again for the same kind of reason.
00:22:42.160 --> 00:22:47.519
And there's also probably a pretty compelling argument that the reason that's being done is from an exit multiple standpoint.
00:22:47.680 --> 00:22:52.880
If you've got any Bitcoin mining on your balance sheet, you're going to be probably capped at maybe a 5x EBITDA multiple.
00:22:53.119 --> 00:22:57.519
If you are AI infrastructure, that's significantly higher, like 15x plus, right?
00:22:57.599 --> 00:23:03.119
So it's significantly more valuable as a business, even though it's basically just servers doing math, right?
00:23:03.759 --> 00:23:08.400
What's the fundamental difference in terms of the activity that is being done?
00:23:08.480 --> 00:23:10.160
It's not really that different.
00:23:10.240 --> 00:23:18.960
It's hardware using huge amounts of power in a big shed somewhere with a huge amount of power available to it that's going to be difficult to cool depending on how much is in there.
00:23:19.119 --> 00:23:21.039
The output is slightly different.
00:23:21.279 --> 00:23:29.920
Um, depending on who you ask, you could probably make a pretty compelling argument that Bitcoin mining is creating more value at the moment than all of the training.
00:23:30.079 --> 00:23:44.240
Um, but I think that's probably a trend that we'll see accelerate because I mean, even if the companies don't shift into hosting and deploying the infrastructure themselves, they've got the PPAs, they've got the facilities, they've got the power available to them.
00:23:44.400 --> 00:23:48.799
And that is the one thing that AI as a whole is going to need going forward.
00:23:48.880 --> 00:23:54.079
So going back to the keynote, first thing that Jensen said, we're a power-constrained industry.
00:23:54.240 --> 00:23:56.559
It's not compute anymore, it's power availability.
00:23:56.640 --> 00:24:06.799
And if you've got 600 kilowatt racks coming in 2027, realistically, we could probably see a meg not long after that, maybe two meg, not too long after that.
00:24:07.279 --> 00:24:08.559
That's a lot of power.
00:24:08.640 --> 00:24:10.000
Um, where's it going to come from?
00:24:10.079 --> 00:24:11.359
Who's already got access to it?
00:24:11.599 --> 00:24:13.759
The miners have got a lot of it right now in the short term.
00:24:14.079 --> 00:24:19.920
Yeah, and I and I regularly get asked, what's the what's the current price for powered land?
00:24:20.000 --> 00:24:36.799
You know, like what is the what is the value of, let's say, or whatever the whatever the power is, but you know, powered land is is certainly going to increase in value, I would say, over the over the coming years, and they'll probably come a point where, yeah, is it better to develop it or flip it?
00:24:36.960 --> 00:24:41.440
And and I guess there'll be the conversations that the the people in those positions will have.
00:24:42.400 --> 00:24:49.200
The other thing I wanted to talk about was um just events, but data centre world London.
00:24:49.359 --> 00:24:51.119
I don't know, did you go or not?
00:24:51.440 --> 00:24:52.640
Yeah, we were on the stand.
00:24:52.880 --> 00:24:57.519
Um, so Pete and I, um Panchayah Limited, we were exhibiting with PMY.
00:24:57.680 --> 00:24:58.720
Um so yeah, we were there.
00:24:58.799 --> 00:25:00.079
It was uh interesting event.
00:25:00.240 --> 00:25:02.079
I just wanted slightly different.
00:25:02.400 --> 00:25:03.839
I just wonder how busy it was.
00:25:04.000 --> 00:25:04.960
That was all I like.
00:25:05.279 --> 00:25:08.240
I just couldn't believe the first day was mad.
00:25:08.319 --> 00:25:10.319
Like absolutely manic.
00:25:11.119 --> 00:25:13.440
Um it was non-stop.
00:25:13.519 --> 00:25:20.880
Like we got onto the sand shortly before nine o'clock, started talking at about half past nine, and then didn't finish talking until about half past four.
00:25:20.960 --> 00:25:23.519
So really, really solid footfall.
00:25:23.839 --> 00:25:37.599
Um, in terms of demographics and the types of conversations that were being had, last year felt a little bit more like people were there and they were ready to deploy, right?
00:25:37.759 --> 00:25:49.039
So they were instead of tire kicking, it was more of right, we're we've got a project, we've got a pilot project, we want to understand this pilot project, maybe we want to deploy some hardware, maybe we want to continue the conversation afterwards.
00:25:49.200 --> 00:25:53.599
Um, and that was also more general, so it wasn't just AI last year.
00:25:53.759 --> 00:25:57.359
Um, there was a bit around visualization, there was a bit around render.
00:25:57.519 --> 00:26:17.759
Um, this year, pretty much only AI, but also a little bit more hesitation around what's going on, how best to do things, um which makes sense given current market conditions and the speed with which everything is moving, but otherwise, really, really good event.
00:26:18.079 --> 00:26:23.680
Um I was just thinking when I was there of yeah, exactly.
00:26:23.839 --> 00:26:32.480
I was just thinking when I was there of I don't know how many years I've been going to that event, and that's why I use it as a comparator because I've been going for let's say 10 years.
00:26:32.720 --> 00:26:43.440
I've no idea how many years I've been going, but every year it just gets busier and busier and busier, and I think it it's a good way to just understand what's happened to the industry, isn't it?
00:26:43.599 --> 00:26:48.799
And I don't mean how many stands are there, I just mean how many people like in general are actually there.
00:26:48.880 --> 00:26:57.920
I'm sure the event organizers will give us stats around you know, there's more stands or more football or whatever, or more registrations, but it's just the the amount of the mass of people.
00:26:58.160 --> 00:27:09.359
I was you'd kind of walk to walk into the XL and you'd get collared 25 times within about a hundred metres of people that you know, and then you go into the event and it just gets getting more and more and more.
00:27:09.440 --> 00:27:12.960
And I think this industry is is mainstream now, isn't it?
00:27:13.039 --> 00:27:15.279
There's no there's no hiding away anymore.
00:27:15.519 --> 00:27:16.400
Yeah, right.
00:27:16.640 --> 00:27:24.720
Yeah, it's yeah, I mean, seeing data centers being I say that as someone as if I'm like someone who understands how this industry works, right?
00:27:24.799 --> 00:27:26.960
I've been in here for about a year, to be honest.
00:27:27.279 --> 00:27:35.359
Um, but public conversation around it is something that does seem to have picked up, even in just the past 12 months, right?
00:27:35.440 --> 00:27:44.799
So I'm sure you've seen the the long tail of the growth and how we've kind of ended up in this position, but it does seem like things have really accelerated over the course of the past year.
00:27:44.960 --> 00:27:54.559
Now, coming from the crypto industry, the fact that everyone is talking about it now does make me a little bit worried because that tends to imply that there's a top signal and everything's about to fall apart.
00:27:54.799 --> 00:27:57.279
But hey, maybe this time is different.
00:27:57.519 --> 00:27:58.319
Yeah, exactly.
00:27:58.480 --> 00:27:59.119
You just don't know.
00:27:59.200 --> 00:28:05.359
But I do I just found it yeah, really I was a couple of colleagues there and I was just saying, you know, how how busy it is now.
00:28:05.440 --> 00:28:10.720
And we held a data center marketing club event the night before the actual thing.
00:28:10.799 --> 00:28:17.839
And even that, you know, we had I don't know how many people like just did it off the cuff, like 60 to 100 people turn up and talk about marketing.
00:28:17.920 --> 00:28:30.559
And it's just crazy now you you can get all these people in a room from just little silos within the industry all all trying to understand where we're at and what's happening because the change is so quick, and I think that's um that's what it is.
00:28:30.640 --> 00:28:41.200
It it changes before your eyes, but you don't really see it, you don't notice it until you're at these points where you're like, I was here last year and it wasn't this busy, or I wasn't having this many conversations last year.
00:28:41.279 --> 00:28:45.039
And I I just find it find it fascinating when you go to the major events.
00:28:45.119 --> 00:28:49.599
And I know I've I know people that have been obviously I went to Kickstar this year, it was the busiest year they've ever had.
00:28:49.759 --> 00:28:51.279
I'm sure MIPIM was crazy.
00:28:51.359 --> 00:28:52.960
I didn't go, but I'm sure it was crazy.
00:28:53.039 --> 00:29:02.799
Obviously, Europe, DTC, you've got Data Center Nation in Milan, I think, this week, which I'm sure will be a busy event because Milan is off the scale with regards to investments.
00:29:02.960 --> 00:29:09.359
You've got all the platform events that are coming up, you've got Data Cloud Can, who I know are planning for a record event.
00:29:09.519 --> 00:29:10.799
It just goes on and on and on.
00:29:11.200 --> 00:29:12.720
And then you add in the US as well.
00:29:12.799 --> 00:29:15.920
You know, I haven't been out to the US for a while, but you add in all the events out there.
00:29:16.160 --> 00:29:20.640
DCD New York was last week, spoke to a couple of people who just said it was the busiest it's ever been.
00:29:20.720 --> 00:29:23.839
So it's just um just getting busier and busier and busier.
00:29:24.079 --> 00:29:31.519
And I think it's um it's important that we we you know remain in the industry and focused on what we're doing because there is so much noise now, so much happening outside.
00:29:32.000 --> 00:29:33.359
Yeah, yeah, massively.
00:29:33.440 --> 00:29:39.119
So and I think the increased public attention is good on the one hand, right?
00:29:39.279 --> 00:29:42.880
Because it means that people are talking about it and more people are aware of the problems, right?
00:29:43.039 --> 00:29:45.839
So power availability, for example, right?
00:29:45.920 --> 00:29:50.079
That's now a very much mainstream conversation, especially in the UK.
00:29:50.400 --> 00:29:56.880
Um, but then on the other hand, it does make it more difficult to try and keep up with what's actually going on, right?
00:29:57.039 --> 00:30:14.000
As opposed to having to filter through secondhand, third hand sources of information to try and figure out, right, what's the announcement, what was the announcement of the announcement, what act what's actually important, what isn't important, what's some really clever PR that doesn't actually mean anything.
00:30:14.079 --> 00:30:16.240
Um, and that's difficult at the best of times.
00:30:16.319 --> 00:30:19.599
Now it's feels more challenging than it was.
00:30:20.240 --> 00:30:21.039
Yeah, no, totally.
00:30:22.799 --> 00:30:24.559
Trying to write a weekly newsletter.
00:30:24.720 --> 00:30:25.119
Yeah, right.
00:30:25.839 --> 00:30:27.039
What do you what do you put in it?
00:30:27.200 --> 00:30:28.240
What do you choose?
00:30:28.559 --> 00:30:29.039
Exactly.
00:30:29.200 --> 00:30:40.960
It's very enjoyable, and you know, I'm not complaining by any stretch of um both of us talking for Ben, but I'm sure he's as amazed by the success of his as I am of mine and love the fact that it adds so much value.
00:30:41.119 --> 00:30:57.279
But yeah, it gets diff more difficult every every say few months to kind of collate all that information because there's so much noise now and news is repeated, regurgitated, you know, people will release a press release three or four times in a year, which I'm sure a lot of people don't even see, but it happens quite a lot.
00:30:57.440 --> 00:31:00.559
And I have to go back and check have I have I said this before?
00:31:01.119 --> 00:31:05.359
Is this a deja vu moment, or or have I actually written this before?
00:31:05.440 --> 00:31:08.480
And a lot of the time you have written it before, so yeah, it's uh difficult.
00:31:08.559 --> 00:31:15.279
So if any of the PR companies out there could um plot try and keep it aligned, you know, do it do a better job, we'd be uh much appreciated.
00:31:16.640 --> 00:31:17.920
Make our lives a little bit easier.
00:31:18.160 --> 00:31:18.720
Yeah, exactly.
00:31:19.039 --> 00:31:23.039
Feel free to send us any press releases because it means we don't have to uh don't have to dig around for it.
00:31:23.440 --> 00:31:24.720
Yeah, legit.
00:31:25.519 --> 00:31:26.880
No, yeah, I'm definitely serious.
00:31:27.839 --> 00:31:28.319
Yeah, yeah.
00:31:29.039 --> 00:31:31.279
I'd both file email addresses to the show notes.
00:31:31.359 --> 00:31:42.720
Um the other thing I wanted to shout out was um shout out to Stephen Beard and C Lau, who are of night Frank, I don't know if everyone would have seen this, but they've just run 600 kilometres in 10 days.
00:31:42.960 --> 00:31:46.400
Seven UAU states um for autism awareness.
00:31:46.480 --> 00:31:56.319
Um Stephen's child um you know has has autism, and Steve's trying to raise a million dollars for an app to you know help parents and and children that are in a similar situation.
00:31:56.480 --> 00:31:59.440
So I want to shout out like ridiculous achievement.
00:31:59.599 --> 00:32:02.559
I think um sometimes these things go under the radar.
00:32:02.640 --> 00:32:04.079
Um so I just wanted to give it a shout out.
00:32:04.160 --> 00:32:07.680
I'll put a GoFundMe in the show notes of this episode.
00:32:07.839 --> 00:32:10.000
So if you can support, please support.
00:32:10.079 --> 00:32:18.000
And if you can't, you know, just um give it a repost and share a bit of awareness for for what they've done because it's ridiculous to be honest.
00:32:18.079 --> 00:32:22.880
You know, I couldn't even run 600 kilometres in 10 years, they've done it in 10 days.
00:32:24.079 --> 00:32:29.039
Yeah, I mean, ultramarathons are a different kind of experience.
00:32:29.599 --> 00:32:31.759
Um, yeah, I did see that actually.
00:32:31.920 --> 00:32:36.640
Um what an achievement, especially in that part of the world as well.
00:32:36.720 --> 00:32:47.200
Like from a weather standpoint, it's not necessarily conducive to a pleasant experience, and running that kind of distance isn't pleasant.
00:32:47.440 --> 00:32:56.640
Um, I've done what I've done before, I've done 80 kilometers or something over the course of two days, uh, and that sucked.
00:32:56.960 --> 00:33:00.079
So 600 over the course of what, 10?
00:33:00.559 --> 00:33:02.480
Fair play, fair play to them.
00:33:02.799 --> 00:33:03.440
Yeah, exactly.
00:33:04.160 --> 00:33:06.559
And again, like always give causes a shout-out.
00:33:06.640 --> 00:33:18.240
I know Data Centre Cycling Club have got the event coming up where they cycle to CAN every year, which I think, or I feel like they're doing it before or after this year, but you'll find the website Data Centre Cycling Club, go and have a look, six days of cycling.
00:33:18.319 --> 00:33:28.240
But there's a lot of good causes out there in the industry, a lot of great people doing great things for for causes that are outside of the industry, not related to the industry, but they're by people in the industry.
00:33:28.319 --> 00:33:34.720
So, you know, if anyone's if anyone's seeing anything or doing anything, I'll tell you what, if anyone's doing anything, come on this with me and Ben.
00:33:34.799 --> 00:33:39.440
We'll give you, we'll give you the uh we'll give you five minutes to have a little little chat about what you're doing.
00:33:39.519 --> 00:33:42.799
So I um I'd love to give a shout out to anything like that.
00:33:43.519 --> 00:33:44.559
Anything else, Ben?
00:33:44.640 --> 00:33:50.000
I know we think we've just covered six weeks of news in the covered quite a bit, haven't we?
00:33:50.319 --> 00:33:50.720
I think so.
00:33:50.799 --> 00:33:53.279
I think what'll be interesting, I think, to hear from your side.
00:33:53.599 --> 00:33:59.839
Obviously, for me, with things like the 600 kilowatt racks that are on the horizon with the the Rubin Ultras, right?
00:34:02.559 --> 00:34:04.640
Two things, well, two other things that stood out.
00:34:04.799 --> 00:34:08.960
One, Jensen's comments in the keynote around Hopper being dead, right?
00:34:09.119 --> 00:34:14.079
Don't buy Hopper, um, and then branding himself as the chief revenue destroyer.
00:34:14.559 --> 00:34:26.079
Um that doesn't sound like a particularly easy place to be if you've just placed a very large order for what is now effectively being termed as dead kit, that doesn't make life particularly easy.
00:34:26.960 --> 00:34:38.320
What was the reaction that you saw from 600 kilowatts plus racks being legitimately positioned as what's coming next?
00:34:38.559 --> 00:34:42.639
Like what's the what's the reaction being from outside of the rack?
00:34:42.960 --> 00:34:45.920
Um how does one cope with it?
00:34:46.800 --> 00:34:56.880
I think people have been talking about it for a while, not specifically obviously the 600 kilowatt announcement, but the density and the challenge that that creates.
00:34:57.039 --> 00:35:09.280
I think um the the people I always try and talk about, the CTOs of this world who are you know involved in the the technical and the design and you know the future proofing of these facilities are in a very, very difficult spot.
00:35:09.440 --> 00:35:18.239
And I think it's I would say there's there's not I didn't have any specific like calls around it going, oh you know, Andy, have you heard the things untrue?
00:35:18.559 --> 00:35:21.920
Because I think everybody kind of knew that it's going in that direction.
00:35:22.000 --> 00:35:27.280
It's just a case of how how high and you know, where does it end, I guess, is the challenge.
00:35:27.440 --> 00:35:35.280
And ensuring that your developments and your data centers are designed for a point in the future and not for today.
00:35:35.440 --> 00:35:41.360
It's absolutely pointless designing a data center for whatever kit you have today.
00:35:41.440 --> 00:35:42.800
I think that's the biggest challenge.
00:35:42.880 --> 00:35:48.639
And and I think that's why we're seeing that, like I said earlier, that uh analysis of CapEx.
00:35:48.960 --> 00:35:52.559
I don't think it's like we don't need capacity anymore.
00:35:53.039 --> 00:35:58.159
I think it's more what on earth do we need our facility to look like?
00:35:58.559 --> 00:36:00.159
Where do we need capacity?
00:36:00.320 --> 00:36:03.039
What you know, who is going to be the customer?
00:36:03.280 --> 00:36:05.039
I think they're all the questions they're asking.
00:36:05.199 --> 00:36:24.000
There'll be some there'll be some casualties, I would say, of data centers that have been recently launched with a certain infrastructure slash kit inside them, and how long is that actually fit for purpose?
00:36:24.159 --> 00:36:33.119
I think that'll I'm talking reading between lines, no one said anything to me, but I think that will cause some people some problems, would be my my view.
00:36:33.280 --> 00:36:37.760
But yeah, the the density, scale, everything has been the conversation for ages.
00:36:37.840 --> 00:36:42.559
Yeah, most of the podcasts I talk to, when I talk to anybody that's technical, that's where you end up.
00:36:42.800 --> 00:36:44.400
But no one knows the answer, do they?
00:36:44.639 --> 00:36:48.400
Jensen's obviously going to say whatever he's doing is the right thing to do, right?
00:36:48.480 --> 00:36:51.280
Yeah, and I get that as well, and I think a lot of people see that.
00:36:51.360 --> 00:36:58.159
It's um that's why I said earlier, Jensen's one off some, but um, yeah, it's it's fascinating where it's going.
00:36:58.239 --> 00:37:02.320
I I personally don't think anybody really knows.
00:37:02.480 --> 00:37:09.039
And I I I think I did a post on LinkedIn not long ago basically saying that you know anyone who gives you the answer probably doesn't know the answer.
00:37:09.199 --> 00:37:10.800
Because I I just don't think there is one.
00:37:10.880 --> 00:37:15.199
You can have assumptions and you can have direction, you can have education and guesses and viewpoints.
00:37:15.519 --> 00:37:20.079
If there's a definitive answer, we'd all be lumping our money somewhere, wouldn't we?
00:37:20.159 --> 00:37:21.920
And you've only got to look at the financial markets.
00:37:22.000 --> 00:37:29.039
If like there's a lot of uncertainty with regards to which stock's gonna go, which one's gonna win, who's gonna do best out of this.
00:37:29.119 --> 00:37:34.800
And you know, capital normally gives you the um the greatest barometer of direction, I think.
00:37:35.039 --> 00:37:37.280
Follow the follow the money, as they say.
00:37:38.000 --> 00:37:42.400
Yeah, except when it's deep seek, because then apparently the market gets it wrong.
00:37:42.480 --> 00:37:44.880
Right to quote Jensen, that's not me, that's not my perspective.
00:37:45.039 --> 00:37:47.519
To quote Jensen, then the market gets it wrong.
00:37:47.760 --> 00:37:51.519
And that's when you get into the geopolitical market, though, isn't it?
00:37:51.599 --> 00:37:52.880
It's like again, we don't know.
00:37:52.960 --> 00:37:53.599
We just don't know.
00:37:53.679 --> 00:37:57.599
You have to just look at it and go make a deck, well, I'll make an educated decision myself.
00:37:57.840 --> 00:38:02.400
But I understand that politically they're gonna say something.
00:38:02.880 --> 00:38:03.360
Yeah.
00:38:03.599 --> 00:38:05.280
That's all I can say on that.
00:38:05.599 --> 00:38:06.159
Yeah.
00:38:06.480 --> 00:38:08.159
Okay, so follow-on question then.
00:38:08.320 --> 00:38:19.920
If the the power is going in one direction, the rack density is going in one direction, occasionally you'll see like an announcement, new data center facility, 10 to 15 megawatts of power, 45,000 square foot facility, right?
00:38:20.079 --> 00:38:32.239
If you've got 600 kilowatt racks that they're rated for and they'll have probably 50% more power draw in the burn-in phase, that is 10 racks in a 45,000 square foot facility.
00:38:32.480 --> 00:38:34.400
What do you do with the rest of that space?
00:38:34.559 --> 00:38:41.440
Because that to me as an outsider who's trying to come in and understand how things are working, that just doesn't seem like the way to build things anymore.
00:38:41.519 --> 00:38:43.280
It just doesn't seem to make sense.
00:38:43.599 --> 00:38:44.400
No, I agreed.
00:38:44.480 --> 00:38:51.119
And again, I had a conversation with someone, a genuine conversation with someone not long ago about five-side football pictures and go-kart tracks.
00:38:51.519 --> 00:38:51.840
Right.
00:38:52.000 --> 00:38:54.880
No, and I and it wasn't, you know, it wasn't a joke.
00:38:54.960 --> 00:39:11.119
It was a genuine, like, you know, what there's gonna be so much empty space, I think, um, and obviously a limitation of power, which means you can't just say, okay, we'll just put more racks in it, because you haven't got the power to switch them on, for one of putting it in a more simple term.
00:39:11.360 --> 00:39:19.920
So I'd I I'm with you on this, and I think that is part of the challenges with where to put your money and how to build.
00:39:20.079 --> 00:39:38.000
I think it's um you can obviously identify those sites that have significant power availability and kind of build your first phase, knowing that you've got, say, another three, four hundred meg available to you when you need it, but you might not need the building, you might just need the power.
00:39:38.159 --> 00:39:51.280
But for the legacy data centers, which is probably the ones that you're more we're more talking about, the ones that are there and they're already, you know, 100,000 square metres or whatever they might be, I don't know.
00:39:51.440 --> 00:39:56.639
It will they reduce the size of it and offset part of the land?
00:39:56.719 --> 00:39:58.239
I I genuinely have no idea.
00:39:58.320 --> 00:40:00.239
Like, I'm serious with the five-side football pictures.
00:40:00.320 --> 00:40:03.440
I think it's uh, you know, it's a definite possibility.
00:40:03.519 --> 00:40:05.440
I think there is one somewhere with a go-kart track in it.
00:40:05.519 --> 00:40:06.159
I might have made that up.
00:40:06.239 --> 00:40:07.599
I'm pretty sure there is a data center somewhere.
00:40:07.840 --> 00:40:08.400
That's fantastic.
00:40:08.480 --> 00:40:10.480
If there, if there isn't one, that sounds like an opportunity.
00:40:10.719 --> 00:40:12.159
Yeah, but it I don't know.
00:40:12.320 --> 00:40:14.639
And I and it is a conversation that happens.
00:40:14.880 --> 00:40:19.679
I you know, I talk to people about, and again, like, is that I think that's a period we're in.
00:40:19.760 --> 00:40:20.400
I genuinely do.
00:40:20.480 --> 00:40:25.519
Yeah, when Jensen says an inflection point, I think the industry's at an inflection point as well.
00:40:25.599 --> 00:40:28.239
It's not just AI that's at an inflection point.
00:40:28.320 --> 00:40:30.559
It's like, what do we do?
00:40:30.960 --> 00:40:46.159
I wish I knew the answer, but yeah, again, yeah, there's there's probably someone listening that's got a better educated guess of where it'll go than I have, but I I do believe that nobody has a degree of certainty in their in their viewpoint.
00:40:46.559 --> 00:41:14.639
Yeah, yeah, and I think like the the reduction in footprint size does kind of rest on the assumption that ultimately NVIDIA's way of doing things right now is the right way to do things, and the only way to do things right, which is more power, more rack density, um, more powerful hardware off the back of it, and with some of the emerging semiconductor manufacturers, they are showing that there is potentially a different way of doing things.
00:41:14.800 --> 00:41:28.079
Now, how much success they end up having, um, that kind of remains to be seen because with the the Kudo ecosystem, I mean NVIDIA's got a pretty compelling competitive advantage there, right?
00:41:28.239 --> 00:41:38.480
Um, coupled with the fact that they've open sourced a load of the new software off the back of GTC2, like Dynamo, for example, for orchestrating large-scale clusters for inference workloads.
00:41:38.800 --> 00:41:47.519
But then if you look at what's also happened in the semiconductor space recently, I think the most interesting one is Furiosa AI.
00:41:47.760 --> 00:42:01.039
So if anyone's not familiar, they're a South Korean semiconductor manufacturer focused specifically on inference hardware, and they rejected, or apparently rejected, an$800 million bid from Meta to buy them out recently.
00:42:01.280 --> 00:42:07.679
Um that's quite interesting because I mean that's backing yourself to a pretty significant degree.
00:42:07.760 --> 00:42:11.360
I think saying no to 800 mil fair play, if it was said.
00:42:11.760 --> 00:42:16.559
Um but if that's the right business decision, what do they see in the future?
00:42:16.639 --> 00:42:28.559
And their cards are interesting because they're 150 watt cards, and a server gets potentially double the token throughput for say Lava 3.1, um, than the AMD hardware, again, at a significantly lower price point, right?
00:42:28.639 --> 00:42:35.840
So for power-constrained markets or older facilities, for example, um, that's potentially an interesting solution.
00:42:36.079 --> 00:42:45.920
Now, the emerging semiconductors, I mean, even AMD to a certain extent, they're kind of struggling to gain market share at the moment.
00:42:46.000 --> 00:42:52.400
But then when we do see larger scale adoption, I think the market's going to be very, very large, right?
00:42:52.559 --> 00:42:54.960
And there's it's a new market at the end of the day.
00:42:55.039 --> 00:43:00.320
Uh, and there's always space for new players and new solutions in new markets.
00:43:00.480 --> 00:43:07.360
And at the end, like the massive amount of floor space and the power constraints is a problem to be solved, right?
00:43:07.599 --> 00:43:12.559
So whoever can solve that particular problem, that's potentially something quite interesting.
00:43:12.639 --> 00:43:18.480
So I think uh if you do have a 45,000 square foot facility and you're worried about only having 10 racks in that, I wouldn't worry too much.
00:43:18.559 --> 00:43:23.840
There's probably something else that we just don't know about yet that's gonna appear out of nowhere and provide you with a solution, hopefully.
00:43:24.239 --> 00:43:26.480
Yeah, something will change the game, it always does.
00:43:26.559 --> 00:43:28.800
And yeah, gives us something to talk about anyway, doesn't it?
00:43:29.119 --> 00:43:31.199
If it wasn't for that, we wouldn't be talking now.
00:43:31.360 --> 00:43:33.360
So appreciate all this uncertainty.
00:43:33.519 --> 00:43:35.280
It gives us something to chat about.
00:43:35.599 --> 00:43:36.719
I think that's all.
00:43:36.880 --> 00:43:40.880
Um, all I was gonna say is obviously this is the second episode we're done.
00:43:40.960 --> 00:43:41.679
We are gonna do it.
00:43:41.760 --> 00:43:43.760
Try and get to once a month, hopefully.
00:43:43.840 --> 00:43:45.920
But if not, we'll stick at six weeks.
00:43:46.320 --> 00:43:47.119
We're doing our best.
00:43:47.280 --> 00:43:56.159
Um, if anyone wants to come on that's listening, and you've got if any news breaks in the next month or six weeks, depending on when we record and you want to come on and talk about it, please do.
00:43:56.239 --> 00:44:00.880
I'm I'm I'm all up for getting some other opinions um of what's happening in the market.
00:44:00.960 --> 00:44:02.960
We'll always give you our opinion and what we're seeing.
00:44:03.039 --> 00:44:05.199
But if anybody else wants to join us, you know, please do.
00:44:05.280 --> 00:44:20.559
And like I said earlier, if anyone's doing anything um a bit different, you know, in the industry, charitable or events or whatever it might be, you know, happy to happy to give you a plug and if it if it helps raise some money for somebody, all the better.
00:44:20.960 --> 00:44:21.920
We better get on.
00:44:22.480 --> 00:44:23.920
We've got newsletters to write.
00:44:24.400 --> 00:44:26.000
Got newsletters to write, yeah, right.
00:44:26.239 --> 00:44:27.679
It's only Monday when we're recording it.
00:44:27.840 --> 00:44:29.199
Probably release this on Friday.
00:44:29.280 --> 00:44:33.920
So hopefully our uh our newsletters won't be far away by the time by the time you hear this.
00:44:34.000 --> 00:44:36.960
But I'll put the GoFundMe in the show notes.
00:44:37.039 --> 00:44:40.719
Um, obviously, if you need myself or Ben, we're pretty easy to find.
00:44:40.800 --> 00:44:41.920
Just come and find us.
00:44:42.000 --> 00:44:42.960
But thanks for your time.
00:44:43.119 --> 00:44:44.320
Ben, good to catch up again.
00:44:44.400 --> 00:44:45.760
Now you're back in the UK.
00:44:46.079 --> 00:44:46.960
Likewise, Andy.
00:44:47.039 --> 00:44:47.280
Yep.
00:44:47.360 --> 00:44:48.559
One final question for you.
00:44:48.639 --> 00:44:51.599
Um, hopefully for in a month's time, predictions.
00:44:51.920 --> 00:45:00.719
Based on what you're seeing right now and where there's a little bit of interest, um, what do you reckon we might end up potentially be talking about next month?
00:45:01.119 --> 00:45:02.159
Prediction time.
00:45:02.239 --> 00:45:04.239
Uh, you could have given me a heads up on this one.
00:45:04.320 --> 00:45:24.719
I think we're gonna see a bit more of this uncertainty with CapEx, would be my I think we might see a few more, whether it's analysts or news articles that break about projects being, let's say, delayed or re-evaluated, I would guess in April.
00:45:24.800 --> 00:45:26.079
It's April Fool's Day tomorrow.
00:45:26.239 --> 00:45:29.599
So from April, I would say my that'll be my prediction.
00:45:29.760 --> 00:45:30.559
What about you?
00:45:30.880 --> 00:45:31.280
Cool.
00:45:31.760 --> 00:45:35.039
I think more discussion around sovereign AI.
00:45:35.199 --> 00:45:42.320
I think that's going to become more of a talking point that's gonna slowly work its way into the mainstream a little bit more.
00:45:42.719 --> 00:45:45.599
Um there you go.
00:45:45.679 --> 00:45:48.800
We'll record the next episode when both of those predictions come true.
00:45:48.960 --> 00:45:53.039
So it might be in four weeks, it might see it again in two years.
00:45:53.119 --> 00:45:54.719
But anyway, thanks for listening, everyone.
00:45:54.800 --> 00:45:57.280
We'll do it again in uh in the next month or so.
