WEBVTT
00:00:00.239 --> 00:00:06.080
The data centre industry is moving faster than ever, and behind the scenes, some seriously smart people are making it all happen.
00:00:06.240 --> 00:00:13.039
Welcome to the Inside Data Centre Podcast, where we dive into the stories, insights and innovations driving real change across the industry.
00:00:13.279 --> 00:00:19.920
From hyperscale campuses to game-changing technologies, we're talking to the people shaping the future of digital infrastructure.
00:00:20.079 --> 00:00:25.839
Whether you live and breathe data centers or just want to understand what's powering our connected world, you're in the right place.
00:00:26.079 --> 00:00:28.079
Welcome to the Inside Data Centre Podcast.
00:00:28.239 --> 00:00:34.640
We're back with another edition of our monthly, that is not monthly, Data Centre News and GPUs.
00:00:34.799 --> 00:00:35.759
Afternoon, Ben.
00:00:36.079 --> 00:00:37.200
How are you doing, Andy?
00:00:37.280 --> 00:00:43.039
Um, yeah, this is podcast, monthly podcast number three in a 12-month period, so we're we're doing well at the moment.
00:00:43.280 --> 00:00:43.920
I know.
00:00:44.240 --> 00:00:48.240
I had to look back to when our last episode went out, and it was August 2025.
00:00:48.320 --> 00:00:53.520
So I can only apologise to the two or three people that have been sat there waiting for the next monthly edition.
00:00:53.600 --> 00:00:56.240
But we're back, we are here now, and we will try and do it.
00:00:56.320 --> 00:00:58.880
Um I'm not even going to commit to monthly anymore.
00:00:59.520 --> 00:01:03.119
We will try and do another one before the end of the year.
00:01:03.359 --> 00:01:05.840
Time being uh time allowed, really.
00:01:05.920 --> 00:01:10.079
But been a crazy, crazy nine months when you think about it.
00:01:10.159 --> 00:01:11.840
It's a crazy month, it's a crazy day.
00:01:12.000 --> 00:01:15.840
OpenAI have just cancelled the delayed the UK development.
00:01:15.920 --> 00:01:20.319
So you know the what the world is changing every few hours.
00:01:20.560 --> 00:01:23.439
But let's start with a real simple question.
00:01:23.599 --> 00:01:26.079
What's what's been happening in your world then, Ben?
00:01:26.159 --> 00:01:28.000
Where where are you at right now?
00:01:28.480 --> 00:01:32.879
Obviously, the the biggest thing I think is I mean the biggest thing, pun intended, is scale.
00:01:33.040 --> 00:01:33.680
Now, right?
00:01:33.840 --> 00:01:44.480
So if we cast our minds back to the uh the far-flung past of 12 months ago, there were press releases and pretty big announcements.
00:01:44.640 --> 00:01:49.680
Um, like there was the UK AI summit last April sort of time.
00:01:49.760 --> 00:01:56.239
There was all of the large announcements that came out of France, it looked like Europe was finally starting to do things, things were starting to move in the right direction.
00:01:58.000 --> 00:02:00.239
And very little of that has actually come to pass.
00:02:00.480 --> 00:02:01.359
In Europe at least.
00:02:01.439 --> 00:02:14.080
But then you look at what's happening in the States now, and how the bottom falling out of the H-100 market is now definitely no longer the case, right?
00:02:14.240 --> 00:02:18.319
Prices are going back up, everyone is compute constrained, demand is through the roof.
00:02:18.479 --> 00:02:24.000
I think it really just goes to show how quickly things can change in this market.
00:02:24.080 --> 00:02:27.120
And honestly, I've given up trying to predict what's gonna happen.
00:02:27.280 --> 00:02:35.039
I mean, back in December 24, based on what I was seeing at the time, and there were 150 neo clouds offering the same thing, right?
00:02:35.280 --> 00:02:42.159
Which was H100s or H200s, in what seemed like a commodity market with everyone locked in a race to the bottom.
00:02:42.400 --> 00:02:46.400
I put out a piece in the tech capital saying that I thought that some of them were gonna go bust last year.
00:02:46.479 --> 00:02:47.919
I was roundly wrong.
00:02:48.400 --> 00:02:50.479
I think there were like two that went under.
00:02:50.800 --> 00:02:59.439
Um, and that's really just a market timing standpoint because if they'd been able to hold on for a couple more months, then they probably would have caught this demand wave as well.
00:02:59.520 --> 00:03:01.199
And this is where we find ourselves.
00:03:01.280 --> 00:03:16.800
I mean, the things like OpenClaw, um the improvements in anthropics models with Opus and co-work and dispatch and code and everything now, it's just it's absurd.
00:03:17.199 --> 00:03:20.080
You know, what I don't really know how to describe it any other way.
00:03:20.639 --> 00:03:22.000
No, it is crazy times.
00:03:22.080 --> 00:03:26.400
And I was looking earlier at Synergy had a Synergy Research brought a report out, didn't they?
00:03:26.479 --> 00:03:29.520
I think it was this week or very, very recently, anyway.
00:03:29.759 --> 00:03:36.479
Um, when you look at NeoClouds, because it's still such a mystical area of this industry that not a lot of people know a lot about.
00:03:36.560 --> 00:03:56.479
You kind of you know, myself included, you're trying to learn more about this ever-evolving call it an industry that like we say, changed every 10 minutes, but um revenue exceeded$25 billion for 2025, up 223% year on year in Q4, CAGR of 58%.
00:03:57.199 --> 00:04:00.879
And there's still people out there that say there's no revenue in this.
00:04:00.960 --> 00:04:02.800
You know, where's where's the profit coming from?
00:04:02.960 --> 00:04:04.400
Like, oh the profit's there.
00:04:04.560 --> 00:04:10.800
It's just not there in the scale that you know people are expecting it, and they're not, you know, and maybe these companies are not shouting about.
00:04:10.879 --> 00:04:16.639
And and like you say, you've got companies like Core Weav, Crusoe, Core Scientific, Lambda, Nebius, Nscale.
00:04:16.800 --> 00:04:23.680
It's just like announcement after announcement, whether it's capital raising, new development, you know, it's it's an absolute crazy world.
00:04:23.759 --> 00:04:26.720
I know your your newsletter obviously really focuses on that area.
00:04:26.959 --> 00:04:38.560
Mine's more broad in the data center industry, but it's becoming more frequent that there's multiple news stories and significant news stories from these types of organisations.
00:04:38.639 --> 00:04:39.680
So you're seeing them.
00:04:39.839 --> 00:04:42.079
I think I would say they're definitely mainstream now.
00:04:42.160 --> 00:04:45.759
I would 100% say it's it's a mainstream part of the industry now.
00:04:45.839 --> 00:04:55.439
But like I said, still very unknown, still a bit mystical, still a bit of anti-data center rhetoric around these these types of players.
00:04:55.519 --> 00:05:03.360
Um, you're still seeing Bitcoin miners come into it, which always creates a bit of uh aminosity, animosity, if I can say it.
00:05:03.439 --> 00:05:06.079
So yeah, interesting, but it's crazy, that's for sure.
00:05:06.240 --> 00:05:28.000
Yeah, I mean, I think that the scale of the neo clouds and how they've grown over the course of the past year, because like N scale this time last year, like doing things, but not anything massive at scale, and now they've just bought an American power company that's planning to do an eight gigawatt site, and Cheryl Sandberg's on the board and Nick Clecks.
00:05:28.160 --> 00:05:30.720
Now, I mean I've got opinions about those things.
00:05:32.720 --> 00:05:33.360
Cross over that one.
00:05:33.920 --> 00:05:34.720
Yeah, we'll ignore that.
00:05:35.439 --> 00:05:43.199
Um but ultimately if you're if you're a startup and you're attracting individuals who are operating at that level, right?
00:05:43.279 --> 00:05:52.399
Who've gone from either being on the board of one of the most successful companies of the past 20 years, or I mean both of them like have been in that position, right?
00:05:52.560 --> 00:05:54.560
Coupled with the political influence that comes as well.
00:05:54.639 --> 00:06:00.800
And you're a startup, you can get those people to buy in enough that they want to be a part of whatever it is you're building.
00:06:00.879 --> 00:06:02.480
That certainly says something.
00:06:03.279 --> 00:06:05.920
And it's and that's a year.
00:06:06.160 --> 00:06:06.879
It's absurd.
00:06:07.120 --> 00:06:09.839
We've gone from a two billion dollar raise, like exactly.
00:06:13.519 --> 00:06:18.160
40 megawatt conversations this time last year to if it's not a gigawatt, I'm not interested.
00:06:18.319 --> 00:06:25.120
And that's I mean, like the UK's grid capacity is what, circa 35 gig or something like that, all in.
00:06:25.279 --> 00:06:30.639
So you've brought one 30-5 of the entire capacity of the UK grid.
00:06:31.040 --> 00:06:34.639
And then they just announced an 8-gigawatt campus, it's absurd.
00:06:35.040 --> 00:06:38.319
Like the scale is absolutely enormous.
00:06:39.040 --> 00:06:43.600
I do think we have to touch on the point though, around what's announced and what's delivered, don't we?
00:06:43.759 --> 00:06:44.879
What's executed?
00:06:44.959 --> 00:07:28.639
Like, because I had I did I recorded another podcast yesterday, funny enough, which isn't it probably go out after this one, but um around execution and and delivery and kind of the the thought process of organisations, it was a report by BCS Consultancy, um, which is obviously largely UK centric, but um a lot of a lot of what you're seeing is these huge announcements, so let's pick a gigawatt number, it doesn't really matter what it is, but the detail might say first phase is 20 megawatts, and the second phase is 60 megawatts, and you know, it's it's um I think it's important that I know why these organisations, well I don't know, but I have a have my thinking around why these organizations do it.
00:07:28.720 --> 00:07:36.560
You know, they're trying to attract customers, they're trying to kind of uh attract community engagement, but it's it also can do the opposite.
00:07:36.639 --> 00:07:43.279
So I think there's it's definitely a conversation around what's announced and can you execute it?
00:07:43.439 --> 00:07:50.560
And I think when you look at Neo Cloud as well, the commitments they have are unbelievable.
00:07:50.720 --> 00:07:59.439
So can you, even if you've got everything, all your ducks in a row, you're ready to go, you're gonna build, can you deliver it and execute it in the speed that you have to do it?
00:07:59.519 --> 00:08:03.839
I think there's so many different angles to that conversation.
00:08:04.000 --> 00:08:10.720
Um, I'd love to see a graph of what what's um what's announced and what's delivered over the next two, three years.
00:08:10.879 --> 00:08:13.680
I think it'll be uh it'll be an interesting bar chart.
00:08:13.920 --> 00:08:14.800
Yeah, no, absolutely.
00:08:14.959 --> 00:08:16.240
I'm building something like that actually.
00:08:16.319 --> 00:08:17.759
So once that's live, I'll let you know.
00:08:17.920 --> 00:08:19.120
Um Yeah, over to you.
00:08:19.279 --> 00:08:19.920
Over to you.
00:08:20.319 --> 00:08:26.720
So honestly, I I think the I mean the part of the reason I started the newsletter, right, was because I was tired of waiting through press releases.
00:08:26.959 --> 00:08:32.480
Um because it's very difficult to sell GPUs to someone if they haven't got any money, but it makes it seem like they have got money, right?
00:08:32.559 --> 00:08:35.759
You just end up spending all of your time talking to tire kickers.
00:08:35.919 --> 00:08:38.720
There are plenty of tire kickers in this space.
00:08:38.879 --> 00:08:52.000
But then you read for a press release, and to your point, right, you'll have this 500 words of game changing this and world changing that and all of this fun stuff, and you get to the end, and then there's one sentence which is like, and we expect to close funding for this project in Q126.
00:08:52.080 --> 00:08:55.039
It's like, oh, so you haven't actually got any money.
00:08:56.480 --> 00:09:07.200
So it's it sounds great, and like sometimes it's a cut before the horses thing, whereby it's like, right, if we get the timings of this press release right, investors are prone to FOMO, just like everybody else, right?
00:09:07.279 --> 00:09:20.639
So if we can get this out and we can make enough of the right noise in the right place with the right people, by the time the dust hustles from this announcement going out, we'll actually probably then have enough cash and enough interest to then be able to deliver on what we've said, right?
00:09:20.799 --> 00:09:24.159
And everyone knows that there's some of that stuff going on at the moment, right?
00:09:24.240 --> 00:09:40.000
Given how opaque this market is and the um, shall we say, the issues with governance um that seem to be bubbling to the surface in some ways, like for example, the 50% of data centers which have apparently been cancelled in the US recently, right?
00:09:40.639 --> 00:09:41.919
There's a lot of noise around them.
00:09:42.000 --> 00:09:42.799
Why were they cancelled?
00:09:42.879 --> 00:09:44.480
What were the reasons for those cancellations?
00:09:44.559 --> 00:09:45.840
Is it because they weren't fundable?
00:09:45.919 --> 00:09:47.519
Is it because they weren't viable?
00:09:47.840 --> 00:09:48.960
I don't know, really.
00:09:49.120 --> 00:09:50.240
Who lost money, right?
00:09:50.320 --> 00:09:52.559
And where can we find that information?
00:09:52.720 --> 00:10:02.879
Um there's also an element of were they ever going to happen as well, because yeah, are you basing that on an announcement that actually had no fact no facts behind it?
00:10:02.960 --> 00:10:04.559
It was simply an amp an announcement.
00:10:04.639 --> 00:10:20.240
So yeah, I think um, and obviously you're getting delays and you're getting it challenges at the moment, you know, supply chains getting smashed again, like it's um it's hard for everyone to keep up, but yeah, it's I think you just you just one of the great things about this industry is you never really know.
00:10:20.399 --> 00:10:32.559
Like you can hear different stories from two very well placed people, and you think, well, one of you you can't both be right, but maybe there is a a a middle ground somewhere in that conversation.
00:10:32.720 --> 00:10:45.600
I um we I don't think many people ever really know what's uh what's happening, and now it has got to the scale of some of these, it has got to the government kind of table, and therefore you're never gonna know unless you're in that conversation.
00:10:45.919 --> 00:10:49.200
Yeah, and I mean fair play, right?
00:10:49.279 --> 00:10:54.320
Like everyone building something is hard at the best of times, right?
00:10:54.559 --> 00:11:06.960
Building something at that scale is incredibly hard because once you start operating at that level, the number of stakeholders that you're having to manage to be able to get something to go, that's a full-time job with the team around you in and of itself.
00:11:07.120 --> 00:11:09.600
Never mind actually building and running the business, right?
00:11:09.679 --> 00:11:14.720
You spend so much time working on the business that you're not necessarily able to be down in the detail.
00:11:14.879 --> 00:11:23.679
So if I mean, again, going from nowhere to an eight gigawatt campus over the period of a couple of years.
00:11:24.159 --> 00:11:25.279
Fair play.
00:11:26.240 --> 00:11:26.799
I can do it.
00:11:27.200 --> 00:11:28.960
No, 100% fair play.
00:11:29.039 --> 00:11:31.840
And also think like what do you think about the shifting customers as well?
00:11:32.000 --> 00:11:35.360
Because again, that's become more noisy, hasn't it?
00:11:35.519 --> 00:11:41.600
Around we have a customer, said customers pulled out, we now have a new customer.
00:11:41.759 --> 00:11:50.320
And I think you've seen it quite a lot in in the US, particularly where some of the major sites like Abilene and the like have had, you know, we've had a customer and the customer's changed.
00:11:50.639 --> 00:11:57.919
By the time it's announced in the media that um they've lost a customer, everyone actually knows that they've already got a new customer.
00:11:58.080 --> 00:12:06.159
So it's not even um it it's sort of it's public knowledge that the new customer secured before they're announcing that the the old customers moved on.
00:12:06.240 --> 00:12:10.480
But you're definitely seeing a bit of a shift in, I guess, organisations.
00:12:11.120 --> 00:12:18.799
Um a lot of it's probably geographic around where they do and and where it's best placed for that um capacity at any time.
00:12:18.879 --> 00:12:24.399
But you are seeing some of these major scale facilities, you know, like I've got a 400 megawatt customer, all of a sudden I haven't.
00:12:24.480 --> 00:12:28.000
It's it's it's not a small amount of capacity that you've then got to shift.
00:12:28.080 --> 00:12:41.600
But on your point at the beginning, you know, the demand's so crazy that that's going as quickly as it's um then you end up with the the problem like of like uh uh who is able to sign a contract for 400 megawatts.
00:12:41.840 --> 00:12:47.360
Of that scale, who's got a balance sheet that is going to be able to support that kind of risk?
00:12:47.519 --> 00:12:50.559
Who are you going to trust to be able to honour that kind of a commitment?
00:12:50.639 --> 00:12:56.559
And there's realistically only like six globally who can take on a project of that size.
00:12:56.799 --> 00:12:59.759
And I mean, everyone wants to work with them, right?
00:12:59.840 --> 00:13:06.879
It's it's your Microsoft's, it's your Metas, it's your Oracle's to a degree, um, OpenAI, Anthropic.
00:13:07.679 --> 00:13:11.120
Everyone wants those customers, everyone wants to sign them.
00:13:11.840 --> 00:13:15.759
But there is it it it's a buyer's market in that regard, right?
00:13:15.840 --> 00:13:20.720
And equally with customers on that scale, they're not signing your contract, you're signing their contracts, right?
00:13:21.039 --> 00:13:29.039
There's probably gonna be terminations and convenience clauses in there as well, for example, like you have to sell us the compute at the lowest rate in the market, otherwise we're gonna walk.
00:13:29.120 --> 00:13:30.320
Because what are you gonna do?
00:13:30.559 --> 00:13:32.399
Sue us, right?
00:13:33.039 --> 00:13:41.039
So I think that area of the market's interesting because that's a that's a very, very different game to be playing.
00:13:41.120 --> 00:13:46.240
But I think the other thing that we're starting to see is the this kind of proliferation of basically neolaps in the middle, right?
00:13:46.399 --> 00:13:50.960
So 12 months ago, only games in town were the the big model builders.
00:13:51.279 --> 00:14:14.480
Now we're seeing I mean companies like Base Camp Research in the UK, for example, right, building out foundation models on monumental genomic data sets that they've been building from research trips into the deepest, darkest Amazon for the past 10 years or so, and then all of a sudden they've got the Eden family of models, which are LLMs trained on that genomic data.
00:14:15.120 --> 00:14:15.440
Right?
00:14:15.840 --> 00:14:17.200
And that's super interesting.
00:14:17.279 --> 00:14:19.679
That's a lot more interesting than a chat box.
00:14:20.000 --> 00:14:29.200
Similar with things like Luminary AI, which is physics AI, and all of these new labs that are going after like a specific set of problems, um using a specific set of data.
00:14:29.279 --> 00:14:44.080
And I think we're really going to see more of that going forward, especially considering it looks like NVIDIA are kind of getting involved in the open source model game a little bit at the moment, which also makes it seem a little bit like they're competing with their customers, which is interesting.
00:14:44.159 --> 00:14:47.919
Um, but that also makes sense because both OpenAI and Anthropic are designing their own chips.
00:14:48.080 --> 00:15:04.559
So why would you not compete with your customers in that regard if they're going to muscle in on your business and you can build some open source models which then lower the barrier to entry for an enterprise deployment who can take one of your open source models off the shelf, fine-tune it on their company data, and it means that you can then sell them hardware so they can deploy on-prem.
00:15:04.639 --> 00:15:06.159
It's a win for everybody, right?
00:15:06.240 --> 00:15:08.879
Especially with the conversations around data sovereignty.
00:15:09.519 --> 00:15:12.399
So again, it's just speed.
00:15:12.480 --> 00:15:14.399
None of this existed 12 months ago.
00:15:14.559 --> 00:15:14.879
Right?
00:15:15.120 --> 00:15:16.240
None of it existed now.
00:15:16.320 --> 00:15:17.200
Here we are.
00:15:18.080 --> 00:15:19.039
Yeah, it's crazy.
00:15:19.120 --> 00:15:24.639
And another point as well, it's kind of the the circle of capital, which is kind of what you you touched on as well.
00:15:26.480 --> 00:15:36.960
Yeah, the cost of the cost of capital as well, you know, like it's it's very different times economically to what it was 12 months ago, you know, even three months ago, six months ago.
00:15:37.360 --> 00:15:41.440
Geopolitically, it's a bit of a it's an interesting place at the moment.
00:15:41.519 --> 00:15:48.080
And I mean I think people are forgetting the cost of capital, you know, like these the the the money behind all of this has to make money.
00:15:48.159 --> 00:15:49.759
You know, that's that's how money works.
00:15:49.840 --> 00:15:51.440
It's it's in it for a return.
00:15:51.519 --> 00:15:55.600
And as capital increases in cost, you're obviously going to see that impact.
00:15:56.000 --> 00:16:04.159
Either the cost to build these facilities is going to um significantly increase, or you're gonna see capital withdraw.
00:16:04.240 --> 00:16:16.879
And like, and will it redeploy into areas such as the UA in the Middle East where there's now going to be a big reconstruction, uh, hopefully, if everything settles down there, then you know, will capital redeploy?
00:16:16.960 --> 00:16:23.360
So there's all these different angles to this industry that are always um kind of shaking things up a little bit.
00:16:23.840 --> 00:16:25.759
I mean, loads of different moving parts.
00:16:25.840 --> 00:16:47.120
But I mean, the capital cost conversation is interesting because, on the one hand, with all of the bigger picture geopolitical risks that everyone's having to manage at the moment, coupled with things like the memory supply chain constraints or the HCD or SCD supply chain constraints, which are increasing capex for the clusters for, I don't know, like 40% or something stupid like that.
00:16:47.440 --> 00:16:56.240
On the other hand, you've got a lot more players who are able to fund GPUs and fund the compute clusters that didn't exist previously.
00:16:56.399 --> 00:17:00.639
Um, so that that part of the financing ecosystem is now starting to mature.
00:17:00.720 --> 00:17:04.160
Like you've got the bigger ticket investment banks who are now actually getting involved.
00:17:04.319 --> 00:17:24.400
You've got financialization companies like Awn, for example, who are looking to create forward markets or are creating forward markets, or Compute Index, um, who are another player about Switzerland or the data community are more focused on financializing.
00:17:32.640 --> 00:17:34.799
Again, another bit of just part one to go right.
00:17:35.039 --> 00:17:55.200
So I think on one hand, it seems like the the easy money to throw at just a data center because I'm a landowner who happens to have a decent power connection and I might have a good enough fiber connection to be able to do something.
00:17:55.359 --> 00:17:58.319
I think that easy money is now gone, right?
00:17:58.720 --> 00:18:04.160
Um, it seems like investors are doing a lot more due diligence into whether or not site is actually viable.
00:18:04.319 --> 00:18:08.559
One metric that I've seen or heard being used in the US is no longer time to power.
00:18:08.720 --> 00:18:10.160
Because what does time to power mean?
00:18:10.240 --> 00:18:13.599
Like, does that mean that you're going to have a grid connection?
00:18:13.759 --> 00:18:15.680
Does that mean that you're going to have a transformer?
00:18:15.839 --> 00:18:17.119
Do you gonna have your substation?
00:18:17.200 --> 00:18:17.839
Like, what is that?
00:18:18.160 --> 00:18:20.720
Metric that's being used there more often is time to gas.
00:18:21.279 --> 00:18:25.440
Um, how quickly can you get connected to whatever gas pipeline?
00:18:25.599 --> 00:18:30.160
Um and that seems to be the way that the really big stuff is getting going.
00:18:30.240 --> 00:18:39.039
Like the XAI Memphis clusters, um, for example, with Colossus and Colossus 2, it's all gas turbines, it's all off-grid, it's all behind the wire.
00:18:39.200 --> 00:18:41.440
And that seems to be the way that things are going.
00:18:41.599 --> 00:18:48.400
And it's really only going to accelerate as well, because like if you can reduce your time to market, gives your competitors less of an advantage.
00:18:48.480 --> 00:18:50.319
And then there's also the time value of money as well.
00:18:50.400 --> 00:18:59.519
If you can get your money working quicker because you're using, I don't know, gas instead of a grid connection, that's gonna put less strain on the grid for one hand, and you're gonna start making a return quicker as well.
00:18:59.680 --> 00:19:07.680
And like, again, 12 months ago, a few players were talking about behind the meter, behind the wire pricing or micro grids or whatever, right?
00:19:07.920 --> 00:19:13.599
And it was kind of a bit fringe, and the I mean, you've been in the industry for far longer than I, right?
00:19:13.839 --> 00:19:22.640
Everyone has kind of done things the same way for a very long period of time, and then all of a sudden there's all of these upsarts, which a year ago were like, this is the way this isn't the way that it works.
00:19:22.799 --> 00:19:26.880
Those upsarts have been loudly proven correct, and now everyone is having to scramble.
00:19:28.079 --> 00:19:30.559
So again, this is change.
00:19:31.759 --> 00:19:33.519
Have an upstart to change.
00:19:33.920 --> 00:19:34.559
Yeah, exactly.
00:19:34.640 --> 00:19:36.480
No, I don't, I, I, I, I love it.
00:19:36.559 --> 00:19:36.880
I love it.
00:19:37.039 --> 00:19:38.480
The way the industry's been shaken up.
00:19:38.559 --> 00:19:40.720
It's it was far too stale, far too long.
00:19:40.799 --> 00:19:42.400
And I think everyone knows that's my opinion.
00:19:42.480 --> 00:19:44.960
Some people like it, some people don't, but that is my opinion.
00:19:45.119 --> 00:19:50.720
I think um I was talking to a really cool uh startup in the US.
00:19:50.880 --> 00:19:55.599
Uh I've got to be careful what I say here, but um, they're uh heavily linked to gas-powered sites.
00:19:55.680 --> 00:20:03.920
You know, you're they're talking three to five gigawatt demand from a customer for gas-powered sites that have access to these sites, it's all secured and they're ready to go.
00:20:04.079 --> 00:20:08.720
But that is simply that speed to market, speed to capacity, you know, like whatever you want to call it.
00:20:08.799 --> 00:20:17.920
They um the power's there, the power's ready to go, it's gas, it's the US has very become very gas-centric, I think, for the scale that that we're talking about in this conversation.
00:20:18.079 --> 00:20:26.960
I think the uh the grid and the other markets have uh, you know, still are still there, still needed, but when you're talking scale, it's it's not viable.
00:20:27.119 --> 00:20:32.640
I think um you know, like the microgrid I went to, pure data centres and AVK in um in Dublin.
00:20:32.799 --> 00:20:38.880
Cool stuff like that's happening in Europe, but it's not uh it's nowhere near the scale of what we're talking about on the stage.
00:20:39.279 --> 00:20:41.440
Yeah, yeah, like it's only 60 megawatt, Andy.
00:20:41.519 --> 00:20:43.039
That's a that's a quarter of a haul.
00:20:44.799 --> 00:20:55.920
Well that again, you get so used to talking about these monumental amounts of power, and you get so used to operating in these massive numbers, but then you take a step back and you think, right, like how much power actually is that?
00:20:56.559 --> 00:21:02.640
Hundreds of thousands of homes, and it's just one big shed that's doing math and producing heat.
00:21:04.160 --> 00:21:13.440
Yeah, and I think the you know the industry, and again, when you talk about the way we communicate and a way we work, there's so much good stuff that we could do with this power as well.
00:21:13.519 --> 00:21:19.200
Like, there's so much good that we can do, community, like getting that power back back into the grid.
00:21:19.279 --> 00:21:20.400
There's so much we can do.
00:21:20.559 --> 00:21:47.039
I think um I think the the major scale projects definitely need to do some are doing a really good job at this, but some aren't of communicating the value and the you know what we can return into into the communities as well as be seen as a a take, take take on my earlier point around if you announce a three gigawatt site in a community somewhere, it's probably in Texas or wherever it might be.
00:21:47.200 --> 00:21:50.000
Um you know what what are you doing for that community?
00:21:50.079 --> 00:21:59.759
Like you are bringing in just you're bringing it economic benefit, but equally, like you know, as um John and Sandra are living in their house in in Texas, what's the benefit?
00:22:00.079 --> 00:22:01.759
Benefit to them, you know, what can you do for them?
00:22:01.920 --> 00:22:05.680
Like the way we communicate needs still needs to be significantly improved.
00:22:05.839 --> 00:22:09.200
I know like Applied Digital have done a really good job, happy to name someone.
00:22:09.279 --> 00:22:15.599
They've done a lot of work on the community and inviting them into the site, showing people round, trying to give back.
00:22:15.680 --> 00:22:22.000
I think that's um equally as important as kind of you know what we're trying to do.
00:22:22.319 --> 00:22:35.440
It yes, it values us as a society by having access to whatever it is we've got access to, but equally there's got to be some community benefits to the everyday person on the streets.
00:22:35.759 --> 00:22:36.640
Yeah, absolutely.
00:22:36.720 --> 00:22:47.440
And I saw a survey recently um of can't remember where it wasn't the US, but people would prefer to have an Amazon distribution warehouse close to them than they would a data center.
00:22:47.759 --> 00:22:53.519
And I think that kind of tells you everything that you need to know about how the industry is still being perceived at the moment.
00:22:53.759 --> 00:23:17.200
But I think that because there are so many perceptions and uh everyone has an opinion about what proliferation, wide-scale usage of AI means from an employment standpoint, from a we're now going to have to subscribe to absolutely every everything standpoint, right, through surveillance and all that sort of stuff.
00:23:17.359 --> 00:23:26.640
That that tends to dominate the conversation in terms of like what's possible, what do people understand to be possible, what are we communicating is possible.
00:23:26.720 --> 00:23:30.000
There's always this kind of slightly negative lens to it, right?
00:23:30.240 --> 00:23:33.039
Which can very well maybe true, right?
00:23:33.839 --> 00:23:51.839
But just from a personal standpoint, playing with these tools over the course of the past few months or so, and really switching from Chat GPT to Claude, if I'm honest, it's like going from riding a horse to driving a car or using going from using Bing to using Google, I think it's probably a testimony.
00:23:53.440 --> 00:23:56.000
I saw someone with that exact analogy on LinkedIn.
00:23:56.160 --> 00:23:58.799
I think it was today or yes, or yesterday, it made me laugh.
00:23:59.119 --> 00:24:02.559
It's it's so so so accurate.
00:24:02.799 --> 00:24:08.079
But then you realize that, like, okay, what problems am I trying to solve for myself initially?
00:24:08.400 --> 00:24:22.240
And because of the way that things have generally works forever up until this point, you generally have a pretty good understanding of what the limits of your personal capability are, right?
00:24:22.319 --> 00:24:32.240
And you'll stop short of that limit so you don't commit some sort of social faux pas or make yourself look bad professionally or something along those lines, right?
00:24:32.319 --> 00:24:34.480
And you'll stop at that limit and you won't push any further.
00:24:34.559 --> 00:24:38.000
But with these tools, they are infinitely capable, right?
00:24:38.400 --> 00:24:47.599
So if you're trying to build a solution to a problem that you're facing, you'll come up to that limit, and then you'll suddenly realize it's incredibly porous because you've got a tool that's going to help you navigate it.
00:24:47.920 --> 00:25:03.279
So, I mean, I've found that what I'm able to do with these tools, the stuff I'm able to build, the information I'm able to gather, the understanding I've been able to develop with all of this other stuff I've been able to build as well, just in the past couple of months or so, is incredibly different.
00:25:03.359 --> 00:25:18.400
And I think that if more people understood that the big grey box, which is using a lot of power very, very close to you while kicking out a huge amount of heat, is a massive force multiplier if you spend time playing with these tools.
00:25:18.640 --> 00:25:27.920
That I think we generally have a better time communicating because it it's still an intangible benefit, it's more of an experiential benefit.
00:25:28.000 --> 00:25:34.240
And I think the only way that people are really going to understand that is to take some time to play with it.
00:25:34.319 --> 00:25:40.960
But then it's really difficult sometimes to communicate that depending on what industry you're operating in or what job you're doing.
00:25:41.279 --> 00:25:51.839
If you're playing with these tools, and these tools are actively undermining your ability to be employed or to make any money, it's you're gonna go in with a perception, and that's a completely fair enough perception, right?
00:25:52.480 --> 00:26:09.039
So I think bridging that gap, um, and then also trying to make sure that things are a little bit more fair in that we're not just using these massive American models which are incredibly proficient, but then all of that data is going overseas, all of that tax money is going overseas, all of that benefit is going overseas.
00:26:09.200 --> 00:26:17.440
Um, that's a really, really difficult thing, especially in the UK, because we've got, I mean, what, the um the different AI growth songs?
00:26:17.759 --> 00:26:19.440
Great idea in principle, right?
00:26:19.839 --> 00:26:20.640
Where's the demand?
00:26:20.880 --> 00:26:28.960
The demand exists, but it's all being funneled through claw.a or chat GPT at the moment.
00:26:29.279 --> 00:26:30.880
There isn't anything really sovereign.
00:26:30.960 --> 00:26:38.799
The open like and there seems to be this resistance to using the open source models um in the UK at scale because they're Chinese.
00:26:39.039 --> 00:26:40.319
And we don't like that.
00:26:40.720 --> 00:26:45.279
Even though it's on primary gaps, and you can completely like it's not going to do anything.
00:26:45.359 --> 00:26:49.359
You can take it, you can fine-tune it, and you can get rid of all the stuff that you don't want in there, right?
00:26:49.440 --> 00:26:51.599
People don't want to do that because we don't want to take the risk.
00:26:51.759 --> 00:26:54.960
So it's difficult, it's a difficult situation.
00:26:55.920 --> 00:27:01.200
Yeah, I think with the use of AI, I'm I'm quite uh, I guess, opinionated on this one.
00:27:01.279 --> 00:27:03.680
Like, um, you've got to educate people, haven't you?
00:27:03.920 --> 00:27:04.720
But you can't.
00:27:04.880 --> 00:27:09.599
I think I think there's this balance, and I don't want to get political on it.
00:27:09.680 --> 00:27:12.160
Like, they kind of do and they don't want you to use it.
00:27:12.640 --> 00:27:19.680
So it's like, what's what's the what's the balance between educating individuals how you use it?
00:27:19.759 --> 00:27:30.880
You know, like again, like you've gone off on your own with an understand basic understanding of it, you've you you've used Claude, you've developed things with it, like in simple terms, you know, but you've had to do that all on your own.
00:27:30.960 --> 00:27:33.680
You probably have to go and find how to use it on your own.
00:27:33.759 --> 00:27:39.359
You've had to do it, unless you go on a course or an and that's obviously economic, you know, challenge some people kind of forward, things like that.
00:27:39.519 --> 00:27:40.640
You've got to do it all yourself.
00:27:40.720 --> 00:27:44.079
And I think um that applies all the way down the education spectrum right now.
00:27:44.160 --> 00:27:59.359
It's all like they're you know, in China they're teaching kids in infant school equivalent how to use these things, whereas my children are nine and twelve, and if it wasn't for me saying, Oh, you know, you could do this or you could do that, they wouldn't have a have the foggiest idea.
00:27:59.440 --> 00:28:04.640
Well, it's a very British term, they wouldn't have the clue like what to do or how to do it.
00:28:04.720 --> 00:28:11.839
And then, but then as a as a country, we're like, oh, we're an AI, I know we're not, but we say, Oh, we're an AI, we want to be an AI powerhouse.
00:28:11.920 --> 00:28:14.079
You're never gonna be an AI powerhouse in anything.
00:28:14.319 --> 00:28:22.720
Can't even can't even build a blue data center, but but how do you expect to be that if you don't help people learn and understand it?
00:28:22.799 --> 00:28:25.119
You know, like I I have a bit of a fear.
00:28:25.279 --> 00:28:28.240
It's even myself, like I said to you, I'm not technical at all.
00:28:28.319 --> 00:28:33.839
Like, and I'd love to be able to kind of spend the time like you have to develop things and look at how it can benefit.
00:28:33.920 --> 00:28:43.759
And we use it as a business, and we have people in our business that are this is what they do, but it's you still have to, I still think it's in that period where you've got to learn it all yourself, so you've got to be really into it, haven't you?
00:28:44.079 --> 00:28:47.920
And you have to have a basic technical understanding really to work out how it works.
00:28:48.079 --> 00:28:52.480
And and you've got time, time is the biggest thing as well, right?
00:28:52.640 --> 00:28:57.680
Like it's um I'm fortunate that I I work for myself.
00:28:57.839 --> 00:29:06.480
I'm it's taken many years to get to a point where I have enough time to be able to step away from trying to keep the lights to play, yeah, to be able to play with stuff, right?
00:29:06.559 --> 00:29:14.400
Like I'm no longer having to put out fires on a 24 by 7 basis, it's only like a 16 by 7 basis now, which gives me a little bit of extra time.
00:29:14.640 --> 00:29:23.039
But a lot of people just don't have that opportunity at the moment, um for any number of reasons, right?
00:29:23.200 --> 00:29:27.680
And I think that that's the that's the difficulty.
00:29:28.079 --> 00:29:32.319
Um, how do you find ways to to bridge that gap?
00:29:32.480 --> 00:29:53.920
How do you I mean create enough one time and two cognitive headspace for someone to think, right, I'm gonna sit down and I'm gonna experiment with this stuff and see what I can do with it, as opposed to I don't know, my rent has gone up, or I've got this going on with my job, or I'm not being paid enough, or this bid has come in, there's all of this stuff going on as well.
00:29:54.079 --> 00:29:57.759
And that's the difficulty, I think.
00:29:58.160 --> 00:30:09.119
Um and then you if you're someone who's in that position, right, and you see individuals who are doing all of this stuff with it and you don't have the time to do it, that's only gonna breed resentment, right?
00:30:09.200 --> 00:30:16.319
It's it's never gonna come across in a positive way, and then these massive companies raising huge amounts of money when you're trying to keep food on the table.
00:30:16.400 --> 00:30:18.400
Again, how do you bridge that gap?
00:30:19.119 --> 00:30:19.839
That's difficult.
00:30:19.920 --> 00:30:21.279
Yeah, and I think that's the bigger problem.
00:30:21.440 --> 00:30:24.799
I think it yeah, I think my I guess I was thinking about that when you're talking.
00:30:24.960 --> 00:30:29.680
My view, if you're an employer, and we're gonna we're probably going a bit off topic here, but it's worth it's worth a rant.
00:30:30.000 --> 00:30:35.039
If you're if you're an employer, you've got to allocate time for your employees to learn this stuff.
00:30:35.119 --> 00:30:36.640
It's only going to benefit you as a business.
00:30:36.799 --> 00:30:38.880
Like, and I think um don't see it as a threat.
00:30:39.039 --> 00:30:40.480
100% do not see it as a threat.
00:30:40.559 --> 00:30:48.000
Because if you are seeing it as a threat, you your business and your people are probably uh are are under are in big problems, shall we say.
00:30:49.519 --> 00:30:50.079
Yeah, yeah.
00:30:50.240 --> 00:30:54.559
You you have a bigger problem not doing it than you do doing it, is probably a good way to put it.
00:30:54.640 --> 00:30:57.039
Like this is coming, it's there's no doubt about it.
00:30:57.119 --> 00:31:04.160
And like you say, I think when you get when you get into it and you see real proper business case uses for it, then you start to understand what's coming.
00:31:04.240 --> 00:31:10.960
Like when I went, like I said to you earlier, when I went to that Microsoft AI event in in London, you see what this stuff can do and what people are doing with it.
00:31:11.039 --> 00:31:17.680
And it's a very small amount of people using it at the moment because people don't understand, but it's coming, like there's no, there's no doubt about it.
00:31:17.759 --> 00:31:31.359
It's gonna totally shift the way we work, the way we live, the way we do everything, and having an understanding of it is only going to be advantageous to you across everything, you know, just in your life, really, not just in your in your work.
00:31:31.519 --> 00:31:32.079
But there you go.
00:31:32.240 --> 00:31:33.759
Anyway, data centers, let's go back.
00:31:34.160 --> 00:31:34.240
Yeah.
00:31:35.440 --> 00:31:39.440
There's probably probably quite an interesting segue into how the demand, how the shape of the market's changing, right?
00:31:39.759 --> 00:31:44.640
So again, last year, conversations, massive facilities, there's still massive facilities, right?
00:31:44.720 --> 00:31:47.839
But those massive facilities tend to be what training, right?
00:31:47.920 --> 00:31:52.480
It's not necessarily interesting from an inference standpoint because you've got all kinds of issues around latency.
00:31:52.640 --> 00:32:06.799
Like, don't get me wrong, rural northern Norway is a fantastic place because power's nice and cheap, um, and if the connectivity is good enough, but that's not going to be great for real-time inference use cases, so medical, self-driving cars, that sort of thing, right?
00:32:07.119 --> 00:32:08.799
It's just not going to work.
00:32:09.920 --> 00:32:14.000
I think inference is now very much a massive demand driver.
00:32:14.160 --> 00:32:24.079
And we're seeing that, I mean, NVIDIA's groc acquisition is basically a tacit admission that the GPU is no longer the only viable thing, and we're going to integrate the LPUs as well.
00:32:24.160 --> 00:32:33.200
You've got the likes of Cerebrus, Furiosa, various other novel emerging semiconductors raising huge amounts of money in the UK as well.
00:32:33.839 --> 00:32:36.319
What does that do to the shape of the data center?
00:32:36.960 --> 00:32:47.039
So, one thing that I'm seeing, it'd be interesting to get your opinion here, is that suddenly it seems like there's finally some conversation happening around like smaller, decentralized, modular deployments that are either easy.
00:32:47.200 --> 00:32:47.839
Should we call it edge?
00:32:49.920 --> 00:32:53.759
I mean, can you really call like one or two megawatts an edge site though?
00:32:56.160 --> 00:32:58.079
Edge still makes me laugh to this day.
00:32:58.400 --> 00:33:04.960
I've got podcasts from four or five years ago when people are talking about edge, and I used to ask everybody, what does what does edge mean?
00:33:05.119 --> 00:33:05.519
What does it mean?
00:33:05.759 --> 00:33:07.359
No one even knew, yeah, no one knew what it was.
00:33:07.440 --> 00:33:10.799
But yeah, 100% like inference is yeah, is a driver now.
00:33:10.880 --> 00:33:21.279
Definitely, I think um you are seeing platform investment, I say private equity, let's say you're seeing PE look at this market now, which again tells you where it's coming.
00:33:21.440 --> 00:33:37.440
Like I've had a couple of conversations with quite significant PE houses or whatever you want to call them, um, who are investing, you know, let's say hundreds of millions, maybe into the bit billions into these areas, and and there's a reason for that, and that is demand.
00:33:37.519 --> 00:33:45.119
There's always um there's always a period where people talk about a lot of things, not a lot happens, you can see it coming, but you don't know when.
00:33:45.279 --> 00:33:48.640
Then when the money starts to look at it, you know it's coming.
00:33:48.720 --> 00:33:53.359
And yes, definitely um seeing it in Europe and the US 100%.
00:33:53.680 --> 00:33:59.680
And I think I'm sure I'm not as active in APAC, but I don't doubt it's uh it's no different out there as well.
00:33:59.759 --> 00:34:01.920
But yeah, it's uh it's definitely a market.
00:34:02.079 --> 00:34:11.599
I'm not gonna call it edge, you can call it whatever you want, but it is um like you say, smaller data centers in urban areas.
00:34:11.840 --> 00:34:13.599
Let's call it that capture.
00:34:14.639 --> 00:34:16.320
It really rolls off the toilet, doesn't it?
00:34:16.400 --> 00:34:19.760
Yeah, I mean, the um to your point, right?
00:34:19.840 --> 00:34:33.760
That the money moving in is is an out of dating, and I think that that's gonna be a much more interesting market in Europe and the UK, especially because we don't have big old patches of land with a spare gig lying here and lying there, like in the US.
00:34:33.920 --> 00:34:39.760
Um, it tends to be smaller pockets, which I don't know, could be in some sort of other distressed asset.
00:34:40.000 --> 00:34:50.000
Um the power's locked up, the grid connections there, it's too small for a big training data center, but it then becomes much, much more interesting for something inference focused.
00:34:50.079 --> 00:34:52.800
And if things like three persistence, right?
00:34:52.880 --> 00:35:01.519
Like it's uh 5,000 tokens per second throughput or something ridiculous like that, and it's a 23 kilowatt system versus like a one megawatt rack.
00:35:01.920 --> 00:35:08.320
That then becomes much, much more interesting for those um for those edge inference sites.
00:35:08.480 --> 00:35:10.639
Um and I think don't call them edge.
00:35:12.000 --> 00:35:12.400
Edge.
00:35:12.480 --> 00:35:42.559
Oh but I also think um when you think about the energy problem as well, the energy solution and the solution and and uh and the community impact as well, integrating these um smaller urban data centers into um I'm gonna just call it that for the crack now, into uh into developments um where you're building a microgrid and you're building the data center and you're offsetting your heat into the community, all of a sudden you've got like a all-encompassing development.
00:35:42.639 --> 00:35:45.039
You don't, you're not taking from the grid, you're actually giving back.
00:35:45.280 --> 00:35:47.039
You're not taking heat, you're actually giving back.
00:35:47.119 --> 00:35:55.840
You know, like and I think you can do that at that scale, and economically, um, from the conversations I've had, it's viable to the private equity as well.
00:35:55.920 --> 00:36:05.039
Whereas a at a significant scale, it's very hard to make the power sort of circular economy viable.
00:36:05.119 --> 00:36:06.320
But that's the challenge.
00:36:07.199 --> 00:36:08.480
Yeah, definitely.
00:36:08.719 --> 00:36:12.000
What's the most um interesting project that you've had up recently?
00:36:12.880 --> 00:36:15.599
At the moment, it's uh let's have a think.
00:36:15.840 --> 00:36:22.079
There's a lot of new, I think what's happening though, I'd say in the last month has been more.
00:36:22.159 --> 00:36:26.960
I put a post about it yesterday, more talk about countries like Spain now going.
00:36:27.119 --> 00:36:41.840
Like they've been talked about for ages, but you know, there's development investment, Microsoft in Zagolfa, Amazon's Aragon, but all of a sudden you're seeing um new developments coming, new platforms being created, power secured.
00:36:41.920 --> 00:36:49.599
So I think you're going to see more of the newer regions start to come to the fore now, and that's there's there's multiple reasons for that.
00:36:49.760 --> 00:36:55.840
A lot of it is just um the power pressure in the the areas where they want to be is just impossible now.
00:36:56.000 --> 00:37:02.880
You've seen it with Finland as well, like Finland, a lot, you know, like Nebias and uh and the like starting to move into these regions.
00:37:02.960 --> 00:37:11.760
So the geographic spread's definitely um changing, and I mean at the scale, like there's obviously already been developments in these areas, but now you're talking about the scale.
00:37:11.920 --> 00:37:20.559
Um project-wise, I think a lot of the bigger ones have not really come through yet.
00:37:20.719 --> 00:37:29.840
I think you've got like your end scale and the like that are out there that you know trying to build a data centre in a ski resort, which is always uh always challenging.
00:37:30.000 --> 00:37:46.000
But there's a lot of like there's a lot coming that I don't know if I can talk about that are um significant significant scale in quite exciting areas, which I think will be um they'll definitely be challenging from a people perspective, which is why they're talking to me about them.
00:37:46.159 --> 00:37:53.199
But um, yeah, nothing uh nothing specific springs to mind that's but it's it's everywhere, isn't it?
00:37:53.280 --> 00:38:08.960
Now like I think um when you start looking at I still think the repurposing of buildings is uh is quite a cool play when companies are doing that, redeveloping on um power stations, power grids, that's kind of where the industry needs to go.
00:38:09.039 --> 00:38:30.960
I'm talking a lot of um a lot of new monies looking at the industrial sites now, which you've you've seen probably multiple announcements where they're they've got they've acquired X gigawatt of repo, you know, oh X power sites of some description, whether it's uh coal or whatever it might be, they're definitely the that's definitely a play-out of like a significant play.
00:38:31.039 --> 00:38:37.039
I think you'll see um you'll see much more of that over the next say 12 to 18 months.
00:38:38.159 --> 00:38:42.800
Yeah, I think my site is right, it's new names starting to appear.
00:38:43.039 --> 00:39:01.119
Um well I was thinking earlier when we were to when we were talking, sorry to interrupt, but like we say like you know, like you're NScale and Nebus, and these companies have been around for a while, but they've only really come to the fore, let's say 12 to 18 month period approximately, probably less than that when you actually look.
00:39:01.840 --> 00:39:06.079
Is there going to be any more that come into the scale of them?
00:39:06.400 --> 00:39:15.920
Because it's always you know, they're they're they're there now, they've got an as an amount of the market, they're obviously ahead from a capital perspective.
00:39:16.000 --> 00:39:19.039
So, can somebody else get into that area?
00:39:19.119 --> 00:39:23.440
Look, you used to have these conversations about yeah, the the standard colo data centers.
00:39:23.519 --> 00:39:26.000
Can anyone break into the digital equinex type world?
00:39:26.079 --> 00:39:33.679
And people did, um, but can you get into that world because you've got to do it so quickly with so much money?
00:39:33.760 --> 00:39:40.079
Is is there an opportunity for someone to go jump to the four that we may have heard of, but don't know that well, right?
00:39:40.400 --> 00:39:41.280
100%, 100%.
00:39:41.599 --> 00:39:45.280
I mean, so two come to mind, um, both Australian, right?
00:39:45.360 --> 00:39:47.519
So Fermos is one, just making apps.
00:39:47.840 --> 00:39:52.320
Firmo is a monumental amount of money, and they're building a day center in Tasmania.
00:39:53.119 --> 00:39:55.519
And the other one is um Sharon AI.
00:39:55.599 --> 00:40:07.679
Um, so they went public relatively recently, and from that initial prospectus filing, the initial filings they put in, they went public with 432 GPUs online, like 432 ignore them or hoppers.
00:40:08.000 --> 00:40:10.480
They've just signed Canva as a customer, yeah.
00:40:10.880 --> 00:40:12.480
And that's one big marquee customer.
00:40:12.800 --> 00:40:14.320
Galaxy pretty cool as well.
00:40:14.480 --> 00:40:14.800
Yep.
00:40:15.039 --> 00:40:17.280
They announced their results today in my head.
00:40:17.360 --> 00:40:20.400
That's why, because I read the press release probably about two hours ago.
00:40:20.480 --> 00:40:26.320
But yeah, so yeah, when you when you say it like that, there are and I guess it's what's known and what's unknown, isn't it?
00:40:26.480 --> 00:40:27.920
Yeah, we've we've probably heard of these.
00:40:28.079 --> 00:40:32.480
A lot of people may not have done because geographically they might be in different regions or they're just not eyes on it.
00:40:32.559 --> 00:40:37.280
But yeah, there are uh when you think about all the money that's being based.
00:40:38.559 --> 00:40:39.440
Yeah, exactly.
00:40:39.679 --> 00:40:50.480
But then I think like the the overwhelming majority of enterprises at this stage aren't really doing anything at massive scale just yet, right?
00:40:50.800 --> 00:41:09.920
Like HSBC are working with Mistral, for example, or you've got Canva signing with Sharon AI, and these are a couple of big names, and then there's probably going to be loads of other companies who are doing stuff on-prem because they're just not going to ever put their data in a cloud environment because of the nature of whatever their business is, but like that's a minority, right?
00:41:10.159 --> 00:41:17.760
And there are a lot of incredibly valuable um companies out there who are going to want their own infrastructure, right?
00:41:17.920 --> 00:41:19.920
And this this is really just the beginning.
00:41:20.000 --> 00:41:26.239
And if you think about I mean, OpenClaw is has been a massive, massive demand driver.
00:41:26.480 --> 00:41:34.880
The improvement of open source models has also been a massive, massive demand driver, and then you realise that like today is the worst that this stuff is ever going to be.
00:41:35.599 --> 00:41:38.079
It's only, only going to get better.
00:41:38.400 --> 00:41:47.119
And I mean, think about how many people pay for access to Chat GPT or Cloud versus the free users, right?
00:41:47.199 --> 00:41:56.079
And then think about how many people as an overall percentage of the entire population are using this stuff, and this is the demand problem that we're having at the moment, this is the capacity problem that we're having right now.
00:41:56.320 --> 00:41:57.119
It's insane.
00:41:57.280 --> 00:42:00.000
So I think there's plenty of space for other players to get in there.
00:42:00.320 --> 00:42:10.320
Um, and I think we'll probably see some other pretty big names appear relatively soon because it seems like from the conversations I'm having and from other announcements and relationships, right?
00:42:10.480 --> 00:42:13.760
Kinds of people who are now staff team to get involved, they're different.
00:42:14.000 --> 00:42:25.519
They're a different class of operator who've watched what's happening, waited to see what mistakes people make, so they don't make those same mistakes when they get involved as well.
00:42:25.679 --> 00:42:31.519
And there's a lot of that happening at the moment, um, especially in the UK, um, and especially in Europe.
00:42:31.599 --> 00:42:33.599
So it's gonna be quite an exciting time, I think.
00:42:33.679 --> 00:42:36.079
I mean, again, this is this is always the same sign-off.
00:42:36.159 --> 00:42:38.239
It's a very exciting place to be, and things are moving quickly.
00:42:38.960 --> 00:42:41.280
You mean it's you mean it's not gonna be boring anymore?
00:42:41.440 --> 00:42:42.079
What are you gonna do?
00:42:43.039 --> 00:42:44.400
All this spare time I've got.
00:42:44.800 --> 00:42:48.719
Yeah, no, it's uh it's an amazing place to be, I think.
00:42:48.800 --> 00:42:53.280
Um on your point before we wrap up, it's that risk-view reward, isn't it?
00:42:53.440 --> 00:42:56.480
Like you've got different capital has different level of risk.
00:42:56.639 --> 00:43:02.000
You know, some takes, let's say, five percent risk, some it's in the one percent, some it's under that.
00:43:02.159 --> 00:43:07.280
And when you start getting into the significant capital, there's a very, very low level of risk, quite rightly.
00:43:07.519 --> 00:43:14.079
And and when you start to see that coming into the market, it's um it's a very educated investment, isn't it?
00:43:14.159 --> 00:43:23.760
And I guess that's your point around that there will still be more capital, it'd be different capital, and this capital probably a bit of a less risk averse portfolio.
00:43:23.840 --> 00:43:26.320
So it's uh be interesting to see what happens.
00:43:26.480 --> 00:43:27.679
But anyway, we better wrap it up.
00:43:27.760 --> 00:43:36.480
Otherwise, we talk all day about um ed the education system and the the use of AI in the minor control into politics, though, yeah.
00:43:37.119 --> 00:43:42.000
I thought we did well to actually uh brought it back getting too opinionated on it.
00:43:42.079 --> 00:43:42.639
But no, definitely.
00:43:42.719 --> 00:43:45.920
But if anyone wants to carry on that conversation, please reach out, happy to do so.
00:43:46.000 --> 00:43:49.119
Um maybe not recorded.
00:43:49.440 --> 00:43:50.000
Yeah, exactly.
00:43:50.079 --> 00:43:51.599
Yeah, we're definitely not recording that bit.
00:43:51.760 --> 00:43:53.519
Um yeah, I'll let I'll let you go.
00:43:53.599 --> 00:44:00.400
If obviously, as I always say, Ben's GPU newsletter, so Ben can direct you thpu.ai.
00:44:00.719 --> 00:44:04.960
But if you want to um definitely subscribe, if you want to learn more about his stuff, I'm learning about it.
00:44:05.039 --> 00:44:06.800
That's why Ben's here, so he can teach us all.
00:44:06.880 --> 00:44:10.960
But um, yeah, if you want to know anything more about that world, get in touch with Ben.
00:44:11.119 --> 00:44:16.639
Obviously, if you want to speak to me for whatever reason, you you know I'm quite easy to find these days.
00:44:17.039 --> 00:44:21.440
And you're everywhere, everywhere and anywhere, always somewhere, as I say.
00:44:21.519 --> 00:44:21.920
But yeah, yeah.
00:44:22.079 --> 00:44:22.800
All right, I let it go.
00:44:22.960 --> 00:44:23.519
Good to catch up.
00:44:23.599 --> 00:44:29.599
We'll do this again in um not in a month, but uh in a smaller time period than we did the previous one.
00:44:29.760 --> 00:44:30.079
Yeah.
00:44:30.480 --> 00:44:34.800
And I'm sure something will have happened by then in this ever boring industry.
00:44:35.760 --> 00:44:36.400
Thanks everyone.
00:44:36.639 --> 00:44:39.840
Cheers and that's it for this episode of the Inside Data Centre Podcast.
00:44:40.000 --> 00:44:40.960
Thanks for tuning in.
00:44:41.119 --> 00:44:49.039
This show is brought to you by DataX and Data Exec, the recruitment partners powering global talent across the data centre and digital infrastructure world.
00:44:49.280 --> 00:44:53.119
If you enjoyed today's conversations, don't forget to follow, like, and subscribe.
00:44:53.280 --> 00:44:55.760
It really helps us grow the podcast community.
00:44:56.000 --> 00:45:01.199
Stay tuned for more stories, insights, and conversations of the people shaping the future of our industry.
00:00:00.239 --> 00:00:06.080
The data centre industry is moving faster than ever, and behind the scenes, some seriously smart people are making it all happen.
00:00:06.240 --> 00:00:13.039
Welcome to the Inside Data Centre Podcast, where we dive into the stories, insights and innovations driving real change across the industry.
00:00:13.279 --> 00:00:19.920
From hyperscale campuses to game-changing technologies, we're talking to the people shaping the future of digital infrastructure.
00:00:20.079 --> 00:00:25.839
Whether you live and breathe data centers or just want to understand what's powering our connected world, you're in the right place.
00:00:26.079 --> 00:00:28.079
Welcome to the Inside Data Centre Podcast.
00:00:28.239 --> 00:00:34.640
We're back with another edition of our monthly, that is not monthly, Data Centre News and GPUs.
00:00:34.799 --> 00:00:35.759
Afternoon, Ben.
00:00:36.079 --> 00:00:37.200
How are you doing, Andy?
00:00:37.280 --> 00:00:43.039
Um, yeah, this is podcast, monthly podcast number three in a 12-month period, so we're we're doing well at the moment.
00:00:43.280 --> 00:00:43.920
I know.
00:00:44.240 --> 00:00:48.240
I had to look back to when our last episode went out, and it was August 2025.
00:00:48.320 --> 00:00:53.520
So I can only apologise to the two or three people that have been sat there waiting for the next monthly edition.
00:00:53.600 --> 00:00:56.240
But we're back, we are here now, and we will try and do it.
00:00:56.320 --> 00:00:58.880
Um I'm not even going to commit to monthly anymore.
00:00:59.520 --> 00:01:03.119
We will try and do another one before the end of the year.
00:01:03.359 --> 00:01:05.840
Time being uh time allowed, really.
00:01:05.920 --> 00:01:10.079
But been a crazy, crazy nine months when you think about it.
00:01:10.159 --> 00:01:11.840
It's a crazy month, it's a crazy day.
00:01:12.000 --> 00:01:15.840
OpenAI have just cancelled the delayed the UK development.
00:01:15.920 --> 00:01:20.319
So you know the what the world is changing every few hours.
00:01:20.560 --> 00:01:23.439
But let's start with a real simple question.
00:01:23.599 --> 00:01:26.079
What's what's been happening in your world then, Ben?
00:01:26.159 --> 00:01:28.000
Where where are you at right now?
00:01:28.480 --> 00:01:32.879
Obviously, the the biggest thing I think is I mean the biggest thing, pun intended, is scale.
00:01:33.040 --> 00:01:33.680
Now, right?
00:01:33.840 --> 00:01:44.480
So if we cast our minds back to the uh the far-flung past of 12 months ago, there were press releases and pretty big announcements.
00:01:44.640 --> 00:01:49.680
Um, like there was the UK AI summit last April sort of time.
00:01:49.760 --> 00:01:56.239
There was all of the large announcements that came out of France, it looked like Europe was finally starting to do things, things were starting to move in the right direction.
00:01:58.000 --> 00:02:00.239
And very little of that has actually come to pass.
00:02:00.480 --> 00:02:01.359
In Europe at least.
00:02:01.439 --> 00:02:14.080
But then you look at what's happening in the States now, and how the bottom falling out of the H-100 market is now definitely no longer the case, right?
00:02:14.240 --> 00:02:18.319
Prices are going back up, everyone is compute constrained, demand is through the roof.
00:02:18.479 --> 00:02:24.000
I think it really just goes to show how quickly things can change in this market.
00:02:24.080 --> 00:02:27.120
And honestly, I've given up trying to predict what's gonna happen.
00:02:27.280 --> 00:02:35.039
I mean, back in December 24, based on what I was seeing at the time, and there were 150 neo clouds offering the same thing, right?
00:02:35.280 --> 00:02:42.159
Which was H100s or H200s, in what seemed like a commodity market with everyone locked in a race to the bottom.
00:02:42.400 --> 00:02:46.400
I put out a piece in the tech capital saying that I thought that some of them were gonna go bust last year.
00:02:46.479 --> 00:02:47.919
I was roundly wrong.
00:02:48.400 --> 00:02:50.479
I think there were like two that went under.
00:02:50.800 --> 00:02:59.439
Um, and that's really just a market timing standpoint because if they'd been able to hold on for a couple more months, then they probably would have caught this demand wave as well.
00:02:59.520 --> 00:03:01.199
And this is where we find ourselves.
00:03:01.280 --> 00:03:16.800
I mean, the things like OpenClaw, um the improvements in anthropics models with Opus and co-work and dispatch and code and everything now, it's just it's absurd.
00:03:17.199 --> 00:03:20.080
You know, what I don't really know how to describe it any other way.
00:03:20.639 --> 00:03:22.000
No, it is crazy times.
00:03:22.080 --> 00:03:26.400
And I was looking earlier at Synergy had a Synergy Research brought a report out, didn't they?
00:03:26.479 --> 00:03:29.520
I think it was this week or very, very recently, anyway.
00:03:29.759 --> 00:03:36.479
Um, when you look at NeoClouds, because it's still such a mystical area of this industry that not a lot of people know a lot about.
00:03:36.560 --> 00:03:56.479
You kind of you know, myself included, you're trying to learn more about this ever-evolving call it an industry that like we say, changed every 10 minutes, but um revenue exceeded$25 billion for 2025, up 223% year on year in Q4, CAGR of 58%.
00:03:57.199 --> 00:04:00.879
And there's still people out there that say there's no revenue in this.
00:04:00.960 --> 00:04:02.800
You know, where's where's the profit coming from?
00:04:02.960 --> 00:04:04.400
Like, oh the profit's there.
00:04:04.560 --> 00:04:10.800
It's just not there in the scale that you know people are expecting it, and they're not, you know, and maybe these companies are not shouting about.
00:04:10.879 --> 00:04:16.639
And and like you say, you've got companies like Core Weav, Crusoe, Core Scientific, Lambda, Nebius, Nscale.
00:04:16.800 --> 00:04:23.680
It's just like announcement after announcement, whether it's capital raising, new development, you know, it's it's an absolute crazy world.
00:04:23.759 --> 00:04:26.720
I know your your newsletter obviously really focuses on that area.
00:04:26.959 --> 00:04:38.560
Mine's more broad in the data center industry, but it's becoming more frequent that there's multiple news stories and significant news stories from these types of organisations.
00:04:38.639 --> 00:04:39.680
So you're seeing them.
00:04:39.839 --> 00:04:42.079
I think I would say they're definitely mainstream now.
00:04:42.160 --> 00:04:45.759
I would 100% say it's it's a mainstream part of the industry now.
00:04:45.839 --> 00:04:55.439
But like I said, still very unknown, still a bit mystical, still a bit of anti-data center rhetoric around these these types of players.
00:04:55.519 --> 00:05:03.360
Um, you're still seeing Bitcoin miners come into it, which always creates a bit of uh aminosity, animosity, if I can say it.
00:05:03.439 --> 00:05:06.079
So yeah, interesting, but it's crazy, that's for sure.
00:05:06.240 --> 00:05:28.000
Yeah, I mean, I think that the scale of the neo clouds and how they've grown over the course of the past year, because like N scale this time last year, like doing things, but not anything massive at scale, and now they've just bought an American power company that's planning to do an eight gigawatt site, and Cheryl Sandberg's on the board and Nick Clecks.
00:05:28.160 --> 00:05:30.720
Now, I mean I've got opinions about those things.
00:05:32.720 --> 00:05:33.360
Cross over that one.
00:05:33.920 --> 00:05:34.720
Yeah, we'll ignore that.
00:05:35.439 --> 00:05:43.199
Um but ultimately if you're if you're a startup and you're attracting individuals who are operating at that level, right?
00:05:43.279 --> 00:05:52.399
Who've gone from either being on the board of one of the most successful companies of the past 20 years, or I mean both of them like have been in that position, right?
00:05:52.560 --> 00:05:54.560
Coupled with the political influence that comes as well.
00:05:54.639 --> 00:06:00.800
And you're a startup, you can get those people to buy in enough that they want to be a part of whatever it is you're building.
00:06:00.879 --> 00:06:02.480
That certainly says something.
00:06:03.279 --> 00:06:05.920
And it's and that's a year.
00:06:06.160 --> 00:06:06.879
It's absurd.
00:06:07.120 --> 00:06:09.839
We've gone from a two billion dollar raise, like exactly.
00:06:13.519 --> 00:06:18.160
40 megawatt conversations this time last year to if it's not a gigawatt, I'm not interested.
00:06:18.319 --> 00:06:25.120
And that's I mean, like the UK's grid capacity is what, circa 35 gig or something like that, all in.
00:06:25.279 --> 00:06:30.639
So you've brought one 30-5 of the entire capacity of the UK grid.
00:06:31.040 --> 00:06:34.639
And then they just announced an 8-gigawatt campus, it's absurd.
00:06:35.040 --> 00:06:38.319
Like the scale is absolutely enormous.
00:06:39.040 --> 00:06:43.600
I do think we have to touch on the point though, around what's announced and what's delivered, don't we?
00:06:43.759 --> 00:06:44.879
What's executed?
00:06:44.959 --> 00:07:28.639
Like, because I had I did I recorded another podcast yesterday, funny enough, which isn't it probably go out after this one, but um around execution and and delivery and kind of the the thought process of organisations, it was a report by BCS Consultancy, um, which is obviously largely UK centric, but um a lot of a lot of what you're seeing is these huge announcements, so let's pick a gigawatt number, it doesn't really matter what it is, but the detail might say first phase is 20 megawatts, and the second phase is 60 megawatts, and you know, it's it's um I think it's important that I know why these organisations, well I don't know, but I have a have my thinking around why these organizations do it.
00:07:28.720 --> 00:07:36.560
You know, they're trying to attract customers, they're trying to kind of uh attract community engagement, but it's it also can do the opposite.
00:07:36.639 --> 00:07:43.279
So I think there's it's definitely a conversation around what's announced and can you execute it?
00:07:43.439 --> 00:07:50.560
And I think when you look at Neo Cloud as well, the commitments they have are unbelievable.
00:07:50.720 --> 00:07:59.439
So can you, even if you've got everything, all your ducks in a row, you're ready to go, you're gonna build, can you deliver it and execute it in the speed that you have to do it?
00:07:59.519 --> 00:08:03.839
I think there's so many different angles to that conversation.
00:08:04.000 --> 00:08:10.720
Um, I'd love to see a graph of what what's um what's announced and what's delivered over the next two, three years.
00:08:10.879 --> 00:08:13.680
I think it'll be uh it'll be an interesting bar chart.
00:08:13.920 --> 00:08:14.800
Yeah, no, absolutely.
00:08:14.959 --> 00:08:16.240
I'm building something like that actually.
00:08:16.319 --> 00:08:17.759
So once that's live, I'll let you know.
00:08:17.920 --> 00:08:19.120
Um Yeah, over to you.
00:08:19.279 --> 00:08:19.920
Over to you.
00:08:20.319 --> 00:08:26.720
So honestly, I I think the I mean the part of the reason I started the newsletter, right, was because I was tired of waiting through press releases.
00:08:26.959 --> 00:08:32.480
Um because it's very difficult to sell GPUs to someone if they haven't got any money, but it makes it seem like they have got money, right?
00:08:32.559 --> 00:08:35.759
You just end up spending all of your time talking to tire kickers.
00:08:35.919 --> 00:08:38.720
There are plenty of tire kickers in this space.
00:08:38.879 --> 00:08:52.000
But then you read for a press release, and to your point, right, you'll have this 500 words of game changing this and world changing that and all of this fun stuff, and you get to the end, and then there's one sentence which is like, and we expect to close funding for this project in Q126.
00:08:52.080 --> 00:08:55.039
It's like, oh, so you haven't actually got any money.
00:08:56.480 --> 00:09:07.200
So it's it sounds great, and like sometimes it's a cut before the horses thing, whereby it's like, right, if we get the timings of this press release right, investors are prone to FOMO, just like everybody else, right?
00:09:07.279 --> 00:09:20.639
So if we can get this out and we can make enough of the right noise in the right place with the right people, by the time the dust hustles from this announcement going out, we'll actually probably then have enough cash and enough interest to then be able to deliver on what we've said, right?
00:09:20.799 --> 00:09:24.159
And everyone knows that there's some of that stuff going on at the moment, right?
00:09:24.240 --> 00:09:40.000
Given how opaque this market is and the um, shall we say, the issues with governance um that seem to be bubbling to the surface in some ways, like for example, the 50% of data centers which have apparently been cancelled in the US recently, right?
00:09:40.639 --> 00:09:41.919
There's a lot of noise around them.
00:09:42.000 --> 00:09:42.799
Why were they cancelled?
00:09:42.879 --> 00:09:44.480
What were the reasons for those cancellations?
00:09:44.559 --> 00:09:45.840
Is it because they weren't fundable?
00:09:45.919 --> 00:09:47.519
Is it because they weren't viable?
00:09:47.840 --> 00:09:48.960
I don't know, really.
00:09:49.120 --> 00:09:50.240
Who lost money, right?
00:09:50.320 --> 00:09:52.559
And where can we find that information?
00:09:52.720 --> 00:10:02.879
Um there's also an element of were they ever going to happen as well, because yeah, are you basing that on an announcement that actually had no fact no facts behind it?
00:10:02.960 --> 00:10:04.559
It was simply an amp an announcement.
00:10:04.639 --> 00:10:20.240
So yeah, I think um, and obviously you're getting delays and you're getting it challenges at the moment, you know, supply chains getting smashed again, like it's um it's hard for everyone to keep up, but yeah, it's I think you just you just one of the great things about this industry is you never really know.
00:10:20.399 --> 00:10:32.559
Like you can hear different stories from two very well placed people, and you think, well, one of you you can't both be right, but maybe there is a a a middle ground somewhere in that conversation.
00:10:32.720 --> 00:10:45.600
I um we I don't think many people ever really know what's uh what's happening, and now it has got to the scale of some of these, it has got to the government kind of table, and therefore you're never gonna know unless you're in that conversation.
00:10:45.919 --> 00:10:49.200
Yeah, and I mean fair play, right?
00:10:49.279 --> 00:10:54.320
Like everyone building something is hard at the best of times, right?
00:10:54.559 --> 00:11:06.960
Building something at that scale is incredibly hard because once you start operating at that level, the number of stakeholders that you're having to manage to be able to get something to go, that's a full-time job with the team around you in and of itself.
00:11:07.120 --> 00:11:09.600
Never mind actually building and running the business, right?
00:11:09.679 --> 00:11:14.720
You spend so much time working on the business that you're not necessarily able to be down in the detail.
00:11:14.879 --> 00:11:23.679
So if I mean, again, going from nowhere to an eight gigawatt campus over the period of a couple of years.
00:11:24.159 --> 00:11:25.279
Fair play.
00:11:26.240 --> 00:11:26.799
I can do it.
00:11:27.200 --> 00:11:28.960
No, 100% fair play.
00:11:29.039 --> 00:11:31.840
And also think like what do you think about the shifting customers as well?
00:11:32.000 --> 00:11:35.360
Because again, that's become more noisy, hasn't it?
00:11:35.519 --> 00:11:41.600
Around we have a customer, said customers pulled out, we now have a new customer.
00:11:41.759 --> 00:11:50.320
And I think you've seen it quite a lot in in the US, particularly where some of the major sites like Abilene and the like have had, you know, we've had a customer and the customer's changed.
00:11:50.639 --> 00:11:57.919
By the time it's announced in the media that um they've lost a customer, everyone actually knows that they've already got a new customer.
00:11:58.080 --> 00:12:06.159
So it's not even um it it's sort of it's public knowledge that the new customer secured before they're announcing that the the old customers moved on.
00:12:06.240 --> 00:12:10.480
But you're definitely seeing a bit of a shift in, I guess, organisations.
00:12:11.120 --> 00:12:18.799
Um a lot of it's probably geographic around where they do and and where it's best placed for that um capacity at any time.
00:12:18.879 --> 00:12:24.399
But you are seeing some of these major scale facilities, you know, like I've got a 400 megawatt customer, all of a sudden I haven't.
00:12:24.480 --> 00:12:28.000
It's it's it's not a small amount of capacity that you've then got to shift.
00:12:28.080 --> 00:12:41.600
But on your point at the beginning, you know, the demand's so crazy that that's going as quickly as it's um then you end up with the the problem like of like uh uh who is able to sign a contract for 400 megawatts.
00:12:41.840 --> 00:12:47.360
Of that scale, who's got a balance sheet that is going to be able to support that kind of risk?
00:12:47.519 --> 00:12:50.559
Who are you going to trust to be able to honour that kind of a commitment?
00:12:50.639 --> 00:12:56.559
And there's realistically only like six globally who can take on a project of that size.
00:12:56.799 --> 00:12:59.759
And I mean, everyone wants to work with them, right?
00:12:59.840 --> 00:13:06.879
It's it's your Microsoft's, it's your Metas, it's your Oracle's to a degree, um, OpenAI, Anthropic.
00:13:07.679 --> 00:13:11.120
Everyone wants those customers, everyone wants to sign them.
00:13:11.840 --> 00:13:15.759
But there is it it it's a buyer's market in that regard, right?
00:13:15.840 --> 00:13:20.720
And equally with customers on that scale, they're not signing your contract, you're signing their contracts, right?
00:13:21.039 --> 00:13:29.039
There's probably gonna be terminations and convenience clauses in there as well, for example, like you have to sell us the compute at the lowest rate in the market, otherwise we're gonna walk.
00:13:29.120 --> 00:13:30.320
Because what are you gonna do?
00:13:30.559 --> 00:13:32.399
Sue us, right?
00:13:33.039 --> 00:13:41.039
So I think that area of the market's interesting because that's a that's a very, very different game to be playing.
00:13:41.120 --> 00:13:46.240
But I think the other thing that we're starting to see is the this kind of proliferation of basically neolaps in the middle, right?
00:13:46.399 --> 00:13:50.960
So 12 months ago, only games in town were the the big model builders.
00:13:51.279 --> 00:14:14.480
Now we're seeing I mean companies like Base Camp Research in the UK, for example, right, building out foundation models on monumental genomic data sets that they've been building from research trips into the deepest, darkest Amazon for the past 10 years or so, and then all of a sudden they've got the Eden family of models, which are LLMs trained on that genomic data.
00:14:15.120 --> 00:14:15.440
Right?
00:14:15.840 --> 00:14:17.200
And that's super interesting.
00:14:17.279 --> 00:14:19.679
That's a lot more interesting than a chat box.
00:14:20.000 --> 00:14:29.200
Similar with things like Luminary AI, which is physics AI, and all of these new labs that are going after like a specific set of problems, um using a specific set of data.
00:14:29.279 --> 00:14:44.080
And I think we're really going to see more of that going forward, especially considering it looks like NVIDIA are kind of getting involved in the open source model game a little bit at the moment, which also makes it seem a little bit like they're competing with their customers, which is interesting.
00:14:44.159 --> 00:14:47.919
Um, but that also makes sense because both OpenAI and Anthropic are designing their own chips.
00:14:48.080 --> 00:15:04.559
So why would you not compete with your customers in that regard if they're going to muscle in on your business and you can build some open source models which then lower the barrier to entry for an enterprise deployment who can take one of your open source models off the shelf, fine-tune it on their company data, and it means that you can then sell them hardware so they can deploy on-prem.
00:15:04.639 --> 00:15:06.159
It's a win for everybody, right?
00:15:06.240 --> 00:15:08.879
Especially with the conversations around data sovereignty.
00:15:09.519 --> 00:15:12.399
So again, it's just speed.
00:15:12.480 --> 00:15:14.399
None of this existed 12 months ago.
00:15:14.559 --> 00:15:14.879
Right?
00:15:15.120 --> 00:15:16.240
None of it existed now.
00:15:16.320 --> 00:15:17.200
Here we are.
00:15:18.080 --> 00:15:19.039
Yeah, it's crazy.
00:15:19.120 --> 00:15:24.639
And another point as well, it's kind of the the circle of capital, which is kind of what you you touched on as well.
00:15:26.480 --> 00:15:36.960
Yeah, the cost of the cost of capital as well, you know, like it's it's very different times economically to what it was 12 months ago, you know, even three months ago, six months ago.
00:15:37.360 --> 00:15:41.440
Geopolitically, it's a bit of a it's an interesting place at the moment.
00:15:41.519 --> 00:15:48.080
And I mean I think people are forgetting the cost of capital, you know, like these the the the money behind all of this has to make money.
00:15:48.159 --> 00:15:49.759
You know, that's that's how money works.
00:15:49.840 --> 00:15:51.440
It's it's in it for a return.
00:15:51.519 --> 00:15:55.600
And as capital increases in cost, you're obviously going to see that impact.
00:15:56.000 --> 00:16:04.159
Either the cost to build these facilities is going to um significantly increase, or you're gonna see capital withdraw.
00:16:04.240 --> 00:16:16.879
And like, and will it redeploy into areas such as the UA in the Middle East where there's now going to be a big reconstruction, uh, hopefully, if everything settles down there, then you know, will capital redeploy?
00:16:16.960 --> 00:16:23.360
So there's all these different angles to this industry that are always um kind of shaking things up a little bit.
00:16:23.840 --> 00:16:25.759
I mean, loads of different moving parts.
00:16:25.840 --> 00:16:47.120
But I mean, the capital cost conversation is interesting because, on the one hand, with all of the bigger picture geopolitical risks that everyone's having to manage at the moment, coupled with things like the memory supply chain constraints or the HCD or SCD supply chain constraints, which are increasing capex for the clusters for, I don't know, like 40% or something stupid like that.
00:16:47.440 --> 00:16:56.240
On the other hand, you've got a lot more players who are able to fund GPUs and fund the compute clusters that didn't exist previously.
00:16:56.399 --> 00:17:00.639
Um, so that that part of the financing ecosystem is now starting to mature.
00:17:00.720 --> 00:17:04.160
Like you've got the bigger ticket investment banks who are now actually getting involved.
00:17:04.319 --> 00:17:24.400
You've got financialization companies like Awn, for example, who are looking to create forward markets or are creating forward markets, or Compute Index, um, who are another player about Switzerland or the data community are more focused on financializing.
00:17:32.640 --> 00:17:34.799
Again, another bit of just part one to go right.
00:17:35.039 --> 00:17:55.200
So I think on one hand, it seems like the the easy money to throw at just a data center because I'm a landowner who happens to have a decent power connection and I might have a good enough fiber connection to be able to do something.
00:17:55.359 --> 00:17:58.319
I think that easy money is now gone, right?
00:17:58.720 --> 00:18:04.160
Um, it seems like investors are doing a lot more due diligence into whether or not site is actually viable.
00:18:04.319 --> 00:18:08.559
One metric that I've seen or heard being used in the US is no longer time to power.
00:18:08.720 --> 00:18:10.160
Because what does time to power mean?
00:18:10.240 --> 00:18:13.599
Like, does that mean that you're going to have a grid connection?
00:18:13.759 --> 00:18:15.680
Does that mean that you're going to have a transformer?
00:18:15.839 --> 00:18:17.119
Do you gonna have your substation?
00:18:17.200 --> 00:18:17.839
Like, what is that?
00:18:18.160 --> 00:18:20.720
Metric that's being used there more often is time to gas.
00:18:21.279 --> 00:18:25.440
Um, how quickly can you get connected to whatever gas pipeline?
00:18:25.599 --> 00:18:30.160
Um and that seems to be the way that the really big stuff is getting going.
00:18:30.240 --> 00:18:39.039
Like the XAI Memphis clusters, um, for example, with Colossus and Colossus 2, it's all gas turbines, it's all off-grid, it's all behind the wire.
00:18:39.200 --> 00:18:41.440
And that seems to be the way that things are going.
00:18:41.599 --> 00:18:48.400
And it's really only going to accelerate as well, because like if you can reduce your time to market, gives your competitors less of an advantage.
00:18:48.480 --> 00:18:50.319
And then there's also the time value of money as well.
00:18:50.400 --> 00:18:59.519
If you can get your money working quicker because you're using, I don't know, gas instead of a grid connection, that's gonna put less strain on the grid for one hand, and you're gonna start making a return quicker as well.
00:18:59.680 --> 00:19:07.680
And like, again, 12 months ago, a few players were talking about behind the meter, behind the wire pricing or micro grids or whatever, right?
00:19:07.920 --> 00:19:13.599
And it was kind of a bit fringe, and the I mean, you've been in the industry for far longer than I, right?
00:19:13.839 --> 00:19:22.640
Everyone has kind of done things the same way for a very long period of time, and then all of a sudden there's all of these upsarts, which a year ago were like, this is the way this isn't the way that it works.
00:19:22.799 --> 00:19:26.880
Those upsarts have been loudly proven correct, and now everyone is having to scramble.
00:19:28.079 --> 00:19:30.559
So again, this is change.
00:19:31.759 --> 00:19:33.519
Have an upstart to change.
00:19:33.920 --> 00:19:34.559
Yeah, exactly.
00:19:34.640 --> 00:19:36.480
No, I don't, I, I, I, I love it.
00:19:36.559 --> 00:19:36.880
I love it.
00:19:37.039 --> 00:19:38.480
The way the industry's been shaken up.
00:19:38.559 --> 00:19:40.720
It's it was far too stale, far too long.
00:19:40.799 --> 00:19:42.400
And I think everyone knows that's my opinion.
00:19:42.480 --> 00:19:44.960
Some people like it, some people don't, but that is my opinion.
00:19:45.119 --> 00:19:50.720
I think um I was talking to a really cool uh startup in the US.
00:19:50.880 --> 00:19:55.599
Uh I've got to be careful what I say here, but um, they're uh heavily linked to gas-powered sites.
00:19:55.680 --> 00:20:03.920
You know, you're they're talking three to five gigawatt demand from a customer for gas-powered sites that have access to these sites, it's all secured and they're ready to go.
00:20:04.079 --> 00:20:08.720
But that is simply that speed to market, speed to capacity, you know, like whatever you want to call it.
00:20:08.799 --> 00:20:17.920
They um the power's there, the power's ready to go, it's gas, it's the US has very become very gas-centric, I think, for the scale that that we're talking about in this conversation.
00:20:18.079 --> 00:20:26.960
I think the uh the grid and the other markets have uh, you know, still are still there, still needed, but when you're talking scale, it's it's not viable.
00:20:27.119 --> 00:20:32.640
I think um you know, like the microgrid I went to, pure data centres and AVK in um in Dublin.
00:20:32.799 --> 00:20:38.880
Cool stuff like that's happening in Europe, but it's not uh it's nowhere near the scale of what we're talking about on the stage.
00:20:39.279 --> 00:20:41.440
Yeah, yeah, like it's only 60 megawatt, Andy.
00:20:41.519 --> 00:20:43.039
That's a that's a quarter of a haul.
00:20:44.799 --> 00:20:55.920
Well that again, you get so used to talking about these monumental amounts of power, and you get so used to operating in these massive numbers, but then you take a step back and you think, right, like how much power actually is that?
00:20:56.559 --> 00:21:02.640
Hundreds of thousands of homes, and it's just one big shed that's doing math and producing heat.
00:21:04.160 --> 00:21:13.440
Yeah, and I think the you know the industry, and again, when you talk about the way we communicate and a way we work, there's so much good stuff that we could do with this power as well.
00:21:13.519 --> 00:21:19.200
Like, there's so much good that we can do, community, like getting that power back back into the grid.
00:21:19.279 --> 00:21:20.400
There's so much we can do.
00:21:20.559 --> 00:21:47.039
I think um I think the the major scale projects definitely need to do some are doing a really good job at this, but some aren't of communicating the value and the you know what we can return into into the communities as well as be seen as a a take, take take on my earlier point around if you announce a three gigawatt site in a community somewhere, it's probably in Texas or wherever it might be.
00:21:47.200 --> 00:21:50.000
Um you know what what are you doing for that community?
00:21:50.079 --> 00:21:59.759
Like you are bringing in just you're bringing it economic benefit, but equally, like you know, as um John and Sandra are living in their house in in Texas, what's the benefit?
00:22:00.079 --> 00:22:01.759
Benefit to them, you know, what can you do for them?
00:22:01.920 --> 00:22:05.680
Like the way we communicate needs still needs to be significantly improved.
00:22:05.839 --> 00:22:09.200
I know like Applied Digital have done a really good job, happy to name someone.
00:22:09.279 --> 00:22:15.599
They've done a lot of work on the community and inviting them into the site, showing people round, trying to give back.
00:22:15.680 --> 00:22:22.000
I think that's um equally as important as kind of you know what we're trying to do.
00:22:22.319 --> 00:22:35.440
It yes, it values us as a society by having access to whatever it is we've got access to, but equally there's got to be some community benefits to the everyday person on the streets.
00:22:35.759 --> 00:22:36.640
Yeah, absolutely.
00:22:36.720 --> 00:22:47.440
And I saw a survey recently um of can't remember where it wasn't the US, but people would prefer to have an Amazon distribution warehouse close to them than they would a data center.
00:22:47.759 --> 00:22:53.519
And I think that kind of tells you everything that you need to know about how the industry is still being perceived at the moment.
00:22:53.759 --> 00:23:17.200
But I think that because there are so many perceptions and uh everyone has an opinion about what proliferation, wide-scale usage of AI means from an employment standpoint, from a we're now going to have to subscribe to absolutely every everything standpoint, right, through surveillance and all that sort of stuff.
00:23:17.359 --> 00:23:26.640
That that tends to dominate the conversation in terms of like what's possible, what do people understand to be possible, what are we communicating is possible.
00:23:26.720 --> 00:23:30.000
There's always this kind of slightly negative lens to it, right?
00:23:30.240 --> 00:23:33.039
Which can very well maybe true, right?
00:23:33.839 --> 00:23:51.839
But just from a personal standpoint, playing with these tools over the course of the past few months or so, and really switching from Chat GPT to Claude, if I'm honest, it's like going from riding a horse to driving a car or using going from using Bing to using Google, I think it's probably a testimony.
00:23:53.440 --> 00:23:56.000
I saw someone with that exact analogy on LinkedIn.
00:23:56.160 --> 00:23:58.799
I think it was today or yes, or yesterday, it made me laugh.
00:23:59.119 --> 00:24:02.559
It's it's so so so accurate.
00:24:02.799 --> 00:24:08.079
But then you realize that, like, okay, what problems am I trying to solve for myself initially?
00:24:08.400 --> 00:24:22.240
And because of the way that things have generally works forever up until this point, you generally have a pretty good understanding of what the limits of your personal capability are, right?
00:24:22.319 --> 00:24:32.240
And you'll stop short of that limit so you don't commit some sort of social faux pas or make yourself look bad professionally or something along those lines, right?
00:24:32.319 --> 00:24:34.480
And you'll stop at that limit and you won't push any further.
00:24:34.559 --> 00:24:38.000
But with these tools, they are infinitely capable, right?
00:24:38.400 --> 00:24:47.599
So if you're trying to build a solution to a problem that you're facing, you'll come up to that limit, and then you'll suddenly realize it's incredibly porous because you've got a tool that's going to help you navigate it.
00:24:47.920 --> 00:25:03.279
So, I mean, I've found that what I'm able to do with these tools, the stuff I'm able to build, the information I'm able to gather, the understanding I've been able to develop with all of this other stuff I've been able to build as well, just in the past couple of months or so, is incredibly different.
00:25:03.359 --> 00:25:18.400
And I think that if more people understood that the big grey box, which is using a lot of power very, very close to you while kicking out a huge amount of heat, is a massive force multiplier if you spend time playing with these tools.
00:25:18.640 --> 00:25:27.920
That I think we generally have a better time communicating because it it's still an intangible benefit, it's more of an experiential benefit.
00:25:28.000 --> 00:25:34.240
And I think the only way that people are really going to understand that is to take some time to play with it.
00:25:34.319 --> 00:25:40.960
But then it's really difficult sometimes to communicate that depending on what industry you're operating in or what job you're doing.
00:25:41.279 --> 00:25:51.839
If you're playing with these tools, and these tools are actively undermining your ability to be employed or to make any money, it's you're gonna go in with a perception, and that's a completely fair enough perception, right?
00:25:52.480 --> 00:26:09.039
So I think bridging that gap, um, and then also trying to make sure that things are a little bit more fair in that we're not just using these massive American models which are incredibly proficient, but then all of that data is going overseas, all of that tax money is going overseas, all of that benefit is going overseas.
00:26:09.200 --> 00:26:17.440
Um, that's a really, really difficult thing, especially in the UK, because we've got, I mean, what, the um the different AI growth songs?
00:26:17.759 --> 00:26:19.440
Great idea in principle, right?
00:26:19.839 --> 00:26:20.640
Where's the demand?
00:26:20.880 --> 00:26:28.960
The demand exists, but it's all being funneled through claw.a or chat GPT at the moment.
00:26:29.279 --> 00:26:30.880
There isn't anything really sovereign.
00:26:30.960 --> 00:26:38.799
The open like and there seems to be this resistance to using the open source models um in the UK at scale because they're Chinese.
00:26:39.039 --> 00:26:40.319
And we don't like that.
00:26:40.720 --> 00:26:45.279
Even though it's on primary gaps, and you can completely like it's not going to do anything.
00:26:45.359 --> 00:26:49.359
You can take it, you can fine-tune it, and you can get rid of all the stuff that you don't want in there, right?
00:26:49.440 --> 00:26:51.599
People don't want to do that because we don't want to take the risk.
00:26:51.759 --> 00:26:54.960
So it's difficult, it's a difficult situation.
00:26:55.920 --> 00:27:01.200
Yeah, I think with the use of AI, I'm I'm quite uh, I guess, opinionated on this one.
00:27:01.279 --> 00:27:03.680
Like, um, you've got to educate people, haven't you?
00:27:03.920 --> 00:27:04.720
But you can't.
00:27:04.880 --> 00:27:09.599
I think I think there's this balance, and I don't want to get political on it.
00:27:09.680 --> 00:27:12.160
Like, they kind of do and they don't want you to use it.
00:27:12.640 --> 00:27:19.680
So it's like, what's what's the what's the balance between educating individuals how you use it?
00:27:19.759 --> 00:27:30.880
You know, like again, like you've gone off on your own with an understand basic understanding of it, you've you you've used Claude, you've developed things with it, like in simple terms, you know, but you've had to do that all on your own.
00:27:30.960 --> 00:27:33.680
You probably have to go and find how to use it on your own.
00:27:33.759 --> 00:27:39.359
You've had to do it, unless you go on a course or an and that's obviously economic, you know, challenge some people kind of forward, things like that.
00:27:39.519 --> 00:27:40.640
You've got to do it all yourself.
00:27:40.720 --> 00:27:44.079
And I think um that applies all the way down the education spectrum right now.
00:27:44.160 --> 00:27:59.359
It's all like they're you know, in China they're teaching kids in infant school equivalent how to use these things, whereas my children are nine and twelve, and if it wasn't for me saying, Oh, you know, you could do this or you could do that, they wouldn't have a have the foggiest idea.
00:27:59.440 --> 00:28:04.640
Well, it's a very British term, they wouldn't have the clue like what to do or how to do it.
00:28:04.720 --> 00:28:11.839
And then, but then as a as a country, we're like, oh, we're an AI, I know we're not, but we say, Oh, we're an AI, we want to be an AI powerhouse.
00:28:11.920 --> 00:28:14.079
You're never gonna be an AI powerhouse in anything.
00:28:14.319 --> 00:28:22.720
Can't even can't even build a blue data center, but but how do you expect to be that if you don't help people learn and understand it?
00:28:22.799 --> 00:28:25.119
You know, like I I have a bit of a fear.
00:28:25.279 --> 00:28:28.240
It's even myself, like I said to you, I'm not technical at all.
00:28:28.319 --> 00:28:33.839
Like, and I'd love to be able to kind of spend the time like you have to develop things and look at how it can benefit.
00:28:33.920 --> 00:28:43.759
And we use it as a business, and we have people in our business that are this is what they do, but it's you still have to, I still think it's in that period where you've got to learn it all yourself, so you've got to be really into it, haven't you?
00:28:44.079 --> 00:28:47.920
And you have to have a basic technical understanding really to work out how it works.
00:28:48.079 --> 00:28:52.480
And and you've got time, time is the biggest thing as well, right?
00:28:52.640 --> 00:28:57.680
Like it's um I'm fortunate that I I work for myself.
00:28:57.839 --> 00:29:06.480
I'm it's taken many years to get to a point where I have enough time to be able to step away from trying to keep the lights to play, yeah, to be able to play with stuff, right?
00:29:06.559 --> 00:29:14.400
Like I'm no longer having to put out fires on a 24 by 7 basis, it's only like a 16 by 7 basis now, which gives me a little bit of extra time.
00:29:14.640 --> 00:29:23.039
But a lot of people just don't have that opportunity at the moment, um for any number of reasons, right?
00:29:23.200 --> 00:29:27.680
And I think that that's the that's the difficulty.
00:29:28.079 --> 00:29:32.319
Um, how do you find ways to to bridge that gap?
00:29:32.480 --> 00:29:53.920
How do you I mean create enough one time and two cognitive headspace for someone to think, right, I'm gonna sit down and I'm gonna experiment with this stuff and see what I can do with it, as opposed to I don't know, my rent has gone up, or I've got this going on with my job, or I'm not being paid enough, or this bid has come in, there's all of this stuff going on as well.
00:29:54.079 --> 00:29:57.759
And that's the difficulty, I think.
00:29:58.160 --> 00:30:09.119
Um and then you if you're someone who's in that position, right, and you see individuals who are doing all of this stuff with it and you don't have the time to do it, that's only gonna breed resentment, right?
00:30:09.200 --> 00:30:16.319
It's it's never gonna come across in a positive way, and then these massive companies raising huge amounts of money when you're trying to keep food on the table.
00:30:16.400 --> 00:30:18.400
Again, how do you bridge that gap?
00:30:19.119 --> 00:30:19.839
That's difficult.
00:30:19.920 --> 00:30:21.279
Yeah, and I think that's the bigger problem.
00:30:21.440 --> 00:30:24.799
I think it yeah, I think my I guess I was thinking about that when you're talking.
00:30:24.960 --> 00:30:29.680
My view, if you're an employer, and we're gonna we're probably going a bit off topic here, but it's worth it's worth a rant.
00:30:30.000 --> 00:30:35.039
If you're if you're an employer, you've got to allocate time for your employees to learn this stuff.
00:30:35.119 --> 00:30:36.640
It's only going to benefit you as a business.
00:30:36.799 --> 00:30:38.880
Like, and I think um don't see it as a threat.
00:30:39.039 --> 00:30:40.480
100% do not see it as a threat.
00:30:40.559 --> 00:30:48.000
Because if you are seeing it as a threat, you your business and your people are probably uh are are under are in big problems, shall we say.
00:30:49.519 --> 00:30:50.079
Yeah, yeah.
00:30:50.240 --> 00:30:54.559
You you have a bigger problem not doing it than you do doing it, is probably a good way to put it.
00:30:54.640 --> 00:30:57.039
Like this is coming, it's there's no doubt about it.
00:30:57.119 --> 00:31:04.160
And like you say, I think when you get when you get into it and you see real proper business case uses for it, then you start to understand what's coming.
00:31:04.240 --> 00:31:10.960
Like when I went, like I said to you earlier, when I went to that Microsoft AI event in in London, you see what this stuff can do and what people are doing with it.
00:31:11.039 --> 00:31:17.680
And it's a very small amount of people using it at the moment because people don't understand, but it's coming, like there's no, there's no doubt about it.
00:31:17.759 --> 00:31:31.359
It's gonna totally shift the way we work, the way we live, the way we do everything, and having an understanding of it is only going to be advantageous to you across everything, you know, just in your life, really, not just in your in your work.
00:31:31.519 --> 00:31:32.079
But there you go.
00:31:32.240 --> 00:31:33.759
Anyway, data centers, let's go back.
00:31:34.160 --> 00:31:34.240
Yeah.
00:31:35.440 --> 00:31:39.440
There's probably probably quite an interesting segue into how the demand, how the shape of the market's changing, right?
00:31:39.759 --> 00:31:44.640
So again, last year, conversations, massive facilities, there's still massive facilities, right?
00:31:44.720 --> 00:31:47.839
But those massive facilities tend to be what training, right?
00:31:47.920 --> 00:31:52.480
It's not necessarily interesting from an inference standpoint because you've got all kinds of issues around latency.
00:31:52.640 --> 00:32:06.799
Like, don't get me wrong, rural northern Norway is a fantastic place because power's nice and cheap, um, and if the connectivity is good enough, but that's not going to be great for real-time inference use cases, so medical, self-driving cars, that sort of thing, right?
00:32:07.119 --> 00:32:08.799
It's just not going to work.
00:32:09.920 --> 00:32:14.000
I think inference is now very much a massive demand driver.
00:32:14.160 --> 00:32:24.079
And we're seeing that, I mean, NVIDIA's groc acquisition is basically a tacit admission that the GPU is no longer the only viable thing, and we're going to integrate the LPUs as well.
00:32:24.160 --> 00:32:33.200
You've got the likes of Cerebrus, Furiosa, various other novel emerging semiconductors raising huge amounts of money in the UK as well.
00:32:33.839 --> 00:32:36.319
What does that do to the shape of the data center?
00:32:36.960 --> 00:32:47.039
So, one thing that I'm seeing, it'd be interesting to get your opinion here, is that suddenly it seems like there's finally some conversation happening around like smaller, decentralized, modular deployments that are either easy.
00:32:47.200 --> 00:32:47.839
Should we call it edge?
00:32:49.920 --> 00:32:53.759
I mean, can you really call like one or two megawatts an edge site though?
00:32:56.160 --> 00:32:58.079
Edge still makes me laugh to this day.
00:32:58.400 --> 00:33:04.960
I've got podcasts from four or five years ago when people are talking about edge, and I used to ask everybody, what does what does edge mean?
00:33:05.119 --> 00:33:05.519
What does it mean?
00:33:05.759 --> 00:33:07.359
No one even knew, yeah, no one knew what it was.
00:33:07.440 --> 00:33:10.799
But yeah, 100% like inference is yeah, is a driver now.
00:33:10.880 --> 00:33:21.279
Definitely, I think um you are seeing platform investment, I say private equity, let's say you're seeing PE look at this market now, which again tells you where it's coming.
00:33:21.440 --> 00:33:37.440
Like I've had a couple of conversations with quite significant PE houses or whatever you want to call them, um, who are investing, you know, let's say hundreds of millions, maybe into the bit billions into these areas, and and there's a reason for that, and that is demand.
00:33:37.519 --> 00:33:45.119
There's always um there's always a period where people talk about a lot of things, not a lot happens, you can see it coming, but you don't know when.
00:33:45.279 --> 00:33:48.640
Then when the money starts to look at it, you know it's coming.
00:33:48.720 --> 00:33:53.359
And yes, definitely um seeing it in Europe and the US 100%.
00:33:53.680 --> 00:33:59.680
And I think I'm sure I'm not as active in APAC, but I don't doubt it's uh it's no different out there as well.
00:33:59.759 --> 00:34:01.920
But yeah, it's uh it's definitely a market.
00:34:02.079 --> 00:34:11.599
I'm not gonna call it edge, you can call it whatever you want, but it is um like you say, smaller data centers in urban areas.
00:34:11.840 --> 00:34:13.599
Let's call it that capture.
00:34:14.639 --> 00:34:16.320
It really rolls off the toilet, doesn't it?
00:34:16.400 --> 00:34:19.760
Yeah, I mean, the um to your point, right?
00:34:19.840 --> 00:34:33.760
That the money moving in is is an out of dating, and I think that that's gonna be a much more interesting market in Europe and the UK, especially because we don't have big old patches of land with a spare gig lying here and lying there, like in the US.
00:34:33.920 --> 00:34:39.760
Um, it tends to be smaller pockets, which I don't know, could be in some sort of other distressed asset.
00:34:40.000 --> 00:34:50.000
Um the power's locked up, the grid connections there, it's too small for a big training data center, but it then becomes much, much more interesting for something inference focused.
00:34:50.079 --> 00:34:52.800
And if things like three persistence, right?
00:34:52.880 --> 00:35:01.519
Like it's uh 5,000 tokens per second throughput or something ridiculous like that, and it's a 23 kilowatt system versus like a one megawatt rack.
00:35:01.920 --> 00:35:08.320
That then becomes much, much more interesting for those um for those edge inference sites.
00:35:08.480 --> 00:35:10.639
Um and I think don't call them edge.
00:35:12.000 --> 00:35:12.400
Edge.
00:35:12.480 --> 00:35:42.559
Oh but I also think um when you think about the energy problem as well, the energy solution and the solution and and uh and the community impact as well, integrating these um smaller urban data centers into um I'm gonna just call it that for the crack now, into uh into developments um where you're building a microgrid and you're building the data center and you're offsetting your heat into the community, all of a sudden you've got like a all-encompassing development.
00:35:42.639 --> 00:35:45.039
You don't, you're not taking from the grid, you're actually giving back.
00:35:45.280 --> 00:35:47.039
You're not taking heat, you're actually giving back.
00:35:47.119 --> 00:35:55.840
You know, like and I think you can do that at that scale, and economically, um, from the conversations I've had, it's viable to the private equity as well.
00:35:55.920 --> 00:36:05.039
Whereas a at a significant scale, it's very hard to make the power sort of circular economy viable.
00:36:05.119 --> 00:36:06.320
But that's the challenge.
00:36:07.199 --> 00:36:08.480
Yeah, definitely.
00:36:08.719 --> 00:36:12.000
What's the most um interesting project that you've had up recently?
00:36:12.880 --> 00:36:15.599
At the moment, it's uh let's have a think.
00:36:15.840 --> 00:36:22.079
There's a lot of new, I think what's happening though, I'd say in the last month has been more.
00:36:22.159 --> 00:36:26.960
I put a post about it yesterday, more talk about countries like Spain now going.
00:36:27.119 --> 00:36:41.840
Like they've been talked about for ages, but you know, there's development investment, Microsoft in Zagolfa, Amazon's Aragon, but all of a sudden you're seeing um new developments coming, new platforms being created, power secured.
00:36:41.920 --> 00:36:49.599
So I think you're going to see more of the newer regions start to come to the fore now, and that's there's there's multiple reasons for that.
00:36:49.760 --> 00:36:55.840
A lot of it is just um the power pressure in the the areas where they want to be is just impossible now.
00:36:56.000 --> 00:37:02.880
You've seen it with Finland as well, like Finland, a lot, you know, like Nebias and uh and the like starting to move into these regions.
00:37:02.960 --> 00:37:11.760
So the geographic spread's definitely um changing, and I mean at the scale, like there's obviously already been developments in these areas, but now you're talking about the scale.
00:37:11.920 --> 00:37:20.559
Um project-wise, I think a lot of the bigger ones have not really come through yet.
00:37:20.719 --> 00:37:29.840
I think you've got like your end scale and the like that are out there that you know trying to build a data centre in a ski resort, which is always uh always challenging.
00:37:30.000 --> 00:37:46.000
But there's a lot of like there's a lot coming that I don't know if I can talk about that are um significant significant scale in quite exciting areas, which I think will be um they'll definitely be challenging from a people perspective, which is why they're talking to me about them.
00:37:46.159 --> 00:37:53.199
But um, yeah, nothing uh nothing specific springs to mind that's but it's it's everywhere, isn't it?
00:37:53.280 --> 00:38:08.960
Now like I think um when you start looking at I still think the repurposing of buildings is uh is quite a cool play when companies are doing that, redeveloping on um power stations, power grids, that's kind of where the industry needs to go.
00:38:09.039 --> 00:38:30.960
I'm talking a lot of um a lot of new monies looking at the industrial sites now, which you've you've seen probably multiple announcements where they're they've got they've acquired X gigawatt of repo, you know, oh X power sites of some description, whether it's uh coal or whatever it might be, they're definitely the that's definitely a play-out of like a significant play.
00:38:31.039 --> 00:38:37.039
I think you'll see um you'll see much more of that over the next say 12 to 18 months.
00:38:38.159 --> 00:38:42.800
Yeah, I think my site is right, it's new names starting to appear.
00:38:43.039 --> 00:39:01.119
Um well I was thinking earlier when we were to when we were talking, sorry to interrupt, but like we say like you know, like you're NScale and Nebus, and these companies have been around for a while, but they've only really come to the fore, let's say 12 to 18 month period approximately, probably less than that when you actually look.
00:39:01.840 --> 00:39:06.079
Is there going to be any more that come into the scale of them?
00:39:06.400 --> 00:39:15.920
Because it's always you know, they're they're they're there now, they've got an as an amount of the market, they're obviously ahead from a capital perspective.
00:39:16.000 --> 00:39:19.039
So, can somebody else get into that area?
00:39:19.119 --> 00:39:23.440
Look, you used to have these conversations about yeah, the the standard colo data centers.
00:39:23.519 --> 00:39:26.000
Can anyone break into the digital equinex type world?
00:39:26.079 --> 00:39:33.679
And people did, um, but can you get into that world because you've got to do it so quickly with so much money?
00:39:33.760 --> 00:39:40.079
Is is there an opportunity for someone to go jump to the four that we may have heard of, but don't know that well, right?
00:39:40.400 --> 00:39:41.280
100%, 100%.
00:39:41.599 --> 00:39:45.280
I mean, so two come to mind, um, both Australian, right?
00:39:45.360 --> 00:39:47.519
So Fermos is one, just making apps.
00:39:47.840 --> 00:39:52.320
Firmo is a monumental amount of money, and they're building a day center in Tasmania.
00:39:53.119 --> 00:39:55.519
And the other one is um Sharon AI.
00:39:55.599 --> 00:40:07.679
Um, so they went public relatively recently, and from that initial prospectus filing, the initial filings they put in, they went public with 432 GPUs online, like 432 ignore them or hoppers.
00:40:08.000 --> 00:40:10.480
They've just signed Canva as a customer, yeah.
00:40:10.880 --> 00:40:12.480
And that's one big marquee customer.
00:40:12.800 --> 00:40:14.320
Galaxy pretty cool as well.
00:40:14.480 --> 00:40:14.800
Yep.
00:40:15.039 --> 00:40:17.280
They announced their results today in my head.
00:40:17.360 --> 00:40:20.400
That's why, because I read the press release probably about two hours ago.
00:40:20.480 --> 00:40:26.320
But yeah, so yeah, when you when you say it like that, there are and I guess it's what's known and what's unknown, isn't it?
00:40:26.480 --> 00:40:27.920
Yeah, we've we've probably heard of these.
00:40:28.079 --> 00:40:32.480
A lot of people may not have done because geographically they might be in different regions or they're just not eyes on it.
00:40:32.559 --> 00:40:37.280
But yeah, there are uh when you think about all the money that's being based.
00:40:38.559 --> 00:40:39.440
Yeah, exactly.
00:40:39.679 --> 00:40:50.480
But then I think like the the overwhelming majority of enterprises at this stage aren't really doing anything at massive scale just yet, right?
00:40:50.800 --> 00:41:09.920
Like HSBC are working with Mistral, for example, or you've got Canva signing with Sharon AI, and these are a couple of big names, and then there's probably going to be loads of other companies who are doing stuff on-prem because they're just not going to ever put their data in a cloud environment because of the nature of whatever their business is, but like that's a minority, right?
00:41:10.159 --> 00:41:17.760
And there are a lot of incredibly valuable um companies out there who are going to want their own infrastructure, right?
00:41:17.920 --> 00:41:19.920
And this this is really just the beginning.
00:41:20.000 --> 00:41:26.239
And if you think about I mean, OpenClaw is has been a massive, massive demand driver.
00:41:26.480 --> 00:41:34.880
The improvement of open source models has also been a massive, massive demand driver, and then you realise that like today is the worst that this stuff is ever going to be.
00:41:35.599 --> 00:41:38.079
It's only, only going to get better.
00:41:38.400 --> 00:41:47.119
And I mean, think about how many people pay for access to Chat GPT or Cloud versus the free users, right?
00:41:47.199 --> 00:41:56.079
And then think about how many people as an overall percentage of the entire population are using this stuff, and this is the demand problem that we're having at the moment, this is the capacity problem that we're having right now.
00:41:56.320 --> 00:41:57.119
It's insane.
00:41:57.280 --> 00:42:00.000
So I think there's plenty of space for other players to get in there.
00:42:00.320 --> 00:42:10.320
Um, and I think we'll probably see some other pretty big names appear relatively soon because it seems like from the conversations I'm having and from other announcements and relationships, right?
00:42:10.480 --> 00:42:13.760
Kinds of people who are now staff team to get involved, they're different.
00:42:14.000 --> 00:42:25.519
They're a different class of operator who've watched what's happening, waited to see what mistakes people make, so they don't make those same mistakes when they get involved as well.
00:42:25.679 --> 00:42:31.519
And there's a lot of that happening at the moment, um, especially in the UK, um, and especially in Europe.
00:42:31.599 --> 00:42:33.599
So it's gonna be quite an exciting time, I think.
00:42:33.679 --> 00:42:36.079
I mean, again, this is this is always the same sign-off.
00:42:36.159 --> 00:42:38.239
It's a very exciting place to be, and things are moving quickly.
00:42:38.960 --> 00:42:41.280
You mean it's you mean it's not gonna be boring anymore?
00:42:41.440 --> 00:42:42.079
What are you gonna do?
00:42:43.039 --> 00:42:44.400
All this spare time I've got.
00:42:44.800 --> 00:42:48.719
Yeah, no, it's uh it's an amazing place to be, I think.
00:42:48.800 --> 00:42:53.280
Um on your point before we wrap up, it's that risk-view reward, isn't it?
00:42:53.440 --> 00:42:56.480
Like you've got different capital has different level of risk.
00:42:56.639 --> 00:43:02.000
You know, some takes, let's say, five percent risk, some it's in the one percent, some it's under that.
00:43:02.159 --> 00:43:07.280
And when you start getting into the significant capital, there's a very, very low level of risk, quite rightly.
00:43:07.519 --> 00:43:14.079
And and when you start to see that coming into the market, it's um it's a very educated investment, isn't it?
00:43:14.159 --> 00:43:23.760
And I guess that's your point around that there will still be more capital, it'd be different capital, and this capital probably a bit of a less risk averse portfolio.
00:43:23.840 --> 00:43:26.320
So it's uh be interesting to see what happens.
00:43:26.480 --> 00:43:27.679
But anyway, we better wrap it up.
00:43:27.760 --> 00:43:36.480
Otherwise, we talk all day about um ed the education system and the the use of AI in the minor control into politics, though, yeah.
00:43:37.119 --> 00:43:42.000
I thought we did well to actually uh brought it back getting too opinionated on it.
00:43:42.079 --> 00:43:42.639
But no, definitely.
00:43:42.719 --> 00:43:45.920
But if anyone wants to carry on that conversation, please reach out, happy to do so.
00:43:46.000 --> 00:43:49.119
Um maybe not recorded.
00:43:49.440 --> 00:43:50.000
Yeah, exactly.
00:43:50.079 --> 00:43:51.599
Yeah, we're definitely not recording that bit.
00:43:51.760 --> 00:43:53.519
Um yeah, I'll let I'll let you go.
00:43:53.599 --> 00:44:00.400
If obviously, as I always say, Ben's GPU newsletter, so Ben can direct you thpu.ai.
00:44:00.719 --> 00:44:04.960
But if you want to um definitely subscribe, if you want to learn more about his stuff, I'm learning about it.
00:44:05.039 --> 00:44:06.800
That's why Ben's here, so he can teach us all.
00:44:06.880 --> 00:44:10.960
But um, yeah, if you want to know anything more about that world, get in touch with Ben.
00:44:11.119 --> 00:44:16.639
Obviously, if you want to speak to me for whatever reason, you you know I'm quite easy to find these days.
00:44:17.039 --> 00:44:21.440
And you're everywhere, everywhere and anywhere, always somewhere, as I say.
00:44:21.519 --> 00:44:21.920
But yeah, yeah.
00:44:22.079 --> 00:44:22.800
All right, I let it go.
00:44:22.960 --> 00:44:23.519
Good to catch up.
00:44:23.599 --> 00:44:29.599
We'll do this again in um not in a month, but uh in a smaller time period than we did the previous one.
00:44:29.760 --> 00:44:30.079
Yeah.
00:44:30.480 --> 00:44:34.800
And I'm sure something will have happened by then in this ever boring industry.
00:44:35.760 --> 00:44:36.400
Thanks everyone.
00:44:36.639 --> 00:44:39.840
Cheers and that's it for this episode of the Inside Data Centre Podcast.
00:44:40.000 --> 00:44:40.960
Thanks for tuning in.
00:44:41.119 --> 00:44:49.039
This show is brought to you by DataX and Data Exec, the recruitment partners powering global talent across the data centre and digital infrastructure world.
00:44:49.280 --> 00:44:53.119
If you enjoyed today's conversations, don't forget to follow, like, and subscribe.
00:44:53.280 --> 00:44:55.760
It really helps us grow the podcast community.
00:44:56.000 --> 00:45:01.199
Stay tuned for more stories, insights, and conversations of the people shaping the future of our industry.
