Data Centre News & GPU's - Episode 3

0:00SPEAKER_00:This is the Inside Data Center Podcast. We talk to the people in our data center center to give you inside info on everything happening in Inside Podcast.

0:27SPEAKER_02:Welcome to the Inside Data Center podcast. We are back with another episode of Data Center News and GPUs, joined by my usual uh guest or sidekick or co-host, whatever we want to call it ourselves, Ben Ballieri. Good afternoon, Ben.

0:43SPEAKER_01:Afternoon, Andy. It's been a minute or a one-month cadence as uh it's been a bit lackluster.

0:49SPEAKER_02:I know. I I had to look earlier. It was April we released, so what are we? Three months ago. But to be fair, we've been quite busy and and the industry's been pretty busy as well. So we've got plenty to talk about and quite try and do it within a relatively short time period. So this is going to be a pretty manic uh manic few months, and we've been away, we've been travelling, we've had, you know, some of some of us have had personal events and things, so it's been a pretty, pretty crazy period.

1:15SPEAKER_02:But I thought we'd start last episode, April. You decided to ask each of us for our predictions, um, totally off the cuff. So I thought we'd better start there, really, to prove that maybe we do or we don't know what we're talking about. Um we're like Yeah, exactly. Your prediction was more discussion around sovereign AI. So going but this was the I think we released it around April Fool's Day. So we're going back to the end of March, start of April for that prediction.

1:44SPEAKER_02:So what's happened?

1:46SPEAKER_01:A lot. Um, I think is probably the best way to discuss it. So there's three big things that have happened in the past couple of weeks in that regard. Um, we're now seeing some clarity in regard to overall approaches to AI governance, um, overall strategy from a national standpoint. So last week the Trump administration, the US, put out their AI action plan um to much fanfare with the um who was it? I think it was the all-in podcast um in the US, big, big event.

2:18SPEAKER_01:Um, and then this week we've seen the UK's AI Security Institute team up with their Canadian counterparts, AWS Anthropic, and then a couple of other parties as well. And then this is also at the same time that the Chinese government have come out um with their vision for how the world of AI looks with the proposed creation of an international AI governance organization, um, headquartered in Shanghai, naturally, um, which I assume would probably serve a similar kind of function to the International Atomic Energy Authority, um, in that it makes sure that everyone is on the straight and narrow, as it were, making sure that there's no bad actors operating outside of the rules.

3:04SPEAKER_01:Um, obviously, there are three very different ideologies underpinning these things, um, which then also probably means that there's some different definitions of sovereignty as well. Um, what does sovereignty mean in this instance? Um, and I think Europe and the UK have got one very clear definition, which is that our data from our citizens stays in region and it doesn't leave. Um and in order to support those customers and to process that data, that capacity has to be built in country.

3:40SPEAKER_01:Um with a view to then ultimately being able to build up a bit of domestic capacity as well. And but then when you look at the UK and the EU, there's a bit of a bifurcation there in terms of how far sovereignty actually goes. So from a UK standpoint, it seems to be building up of domestic capacity in the same way, but ultimately that domestic capacity could still then be sold on to a US hyperscaler, which is it really sovereign capacity? I don't know.

4:08SPEAKER_01:Maybe, maybe not, depends on the definition that you're operating on. Europe is going one step further. Um I think it was the the head of legal for Microsoft, I think, was asked under oath um in some court case recently whether or not Microsoft could guarantee that European sovereign data would remain in country if the US government requested that data and this particular individual said no, could not guarantee that. So is the UK sovereign in name only with the approach that they're taking?

4:45SPEAKER_01:Potentially. Um I think this is where we're gonna start to see where the differences really lie. Um, and also is the UK going to want to put the Cloud Act to the test in quite the same way that the the French have, because I mean, naturally, it was the French who were gonna ask that question, right? Um and fair play to them because it's the question that needs asking. Um, so yeah, I think it's it's interesting um how all of these things seem to be moving, because I mean, just three short months ago, um the need for sovereign capacity wasn't really clear in quite the same way.

5:24SPEAKER_01:But I think a lot of places have now started to wake up to the fact that if this is going to become infrastructure, as in AI is going to become infrastructure in and of itself, that infrastructure's gotta be domestic. Um, otherwise you just end up renting from someone else, which is arguably what happened in cloud, right? Which has been great for the US, not necessarily quite so good for everyone else from uh from an economic standpoint, from a tax standpoint.

5:52SPEAKER_01:Um, and I think that trend is only going to continue. And I think the I'm gonna say my prediction for what happens next. Now that we have clarity, I say clarity, now that we have three competing offers out in the market from a governance perspective, um, I think we'll start to see alignment with some countries with those differing ideologies. Um and I mean that's gonna have a massive impact on which markets do well and where. Um China's leading into the open source approach.

6:30SPEAKER_01:Um it's kind of the infrastructure investment model, which has probably got some similarities with the Belt and Road Initiative, um, which is big investments in big infrastructure projects, and then everyone throws their arms up saying, Oh, but it's basically debt trap diplomacy, and if someone defaults, then it ends up being that they take control of a port or something like that. But if you get into the nitty-gritty, the terms of these agreements aren't really that different from the IMF, it's just that we're not doing it, so we don't like it.

6:57SPEAKER_01:I guess maybe I'm a little bit cynical there, but hey-ho. Um the US is has obviously got the the frontier models, um, although they're being challenged by the open source alternatives from China at the moment from a benchmarking and from a capabilities standpoint. Um, and then there's Europe, which is taking the regulatory first approach, which on the one hand may limit innovation, but on the other hand, there is the Brussels effect. Um, whereby if Brussels does something and it works, then people do tend to get on size because as broken as Europe seems to be from a capital market standpoint at the moment, collectively it is a massive, massive market with a huge amount of collective GDP.

7:45SPEAKER_01:Problem is at the moment it's really fragmented. Um, if there's ways to fix that fragmentation, then maybe things get that little bit more interesting. Um and I think over the course of the next few months we'll start to see a bit of delineation between right, who's aligning with what and why. Um, and then that will dictate what happens after that.

8:05SPEAKER_02:And if you look at the development of the project, some of the projects that I wrote down before we started doing this podcast around what's been announced in the last few months, sovereignty is is a big part of that. Where the major developments are happening in regions where they, you know, they're looking for that data. Basically, they want to hold the data, they want they want the development in country, and they're looking at that as a as a s as a strategy now.

8:27SPEAKER_02:And it, you know, Stargate announced in Norway today, like as a as an example, you know, and we'll go through some of that shortly. My prediction was um the last sweet we recorded was hot off the Microsoft pause. Who remembers those crazy days of when like when Microsoft announced a pause? Exactly. When Microsoft announced a pause and everyone messaged me saying has the bubble burst, the obsession with a bubble will always uh amuse me. Um and I said there'll be more, there'll be more announcements, more delays, more um re-evaluations, I guess, of and we did see that.

9:03SPEAKER_02:You know, I think I was I was reading earlier that Wells Fargo at the end of April it was about the AWS reallocation of or delays in leasing conversations. Microsoft was still in the announcement around slowing down. Yeah, you were seeing all the big hyperscalers in the analyst notes, pausing, ceasing, you know, stock dropped heavily around those periods as well. If people keep their eyes on the market, um, but then you've you fast forward to now and and yesterday, I think it was Meta that have increased CapEx, Google have increased CapEx, Microsoft have increased CapEx, Oracle have increased CapEx, and I and I looked, the stat is hyperscalers expect to spend 345 billion uh on capex in 2025, which is 52% year-on-year growth.

9:52SPEAKER_02:So when you when you have these periods where an analyst somewhere gets a rumour and decides to write a note and it impacts the whole market, I always say to people, you know, have conversations, talk. You can find out what's really happening. But and again, a lot of that re-evaluation ties into your point around sovereignty. It's it's where's the right area or region or development to invest this CapEx. If we're gonna invest 345 billion collectively, we want to get the right return on that.

10:21SPEAKER_02:So, where do we spend it? And I think that's become more of the conversation, whether that is it's still in the same region, but it's a different site, or whether it's we're gonna invest that into an existing site, or whether it is actually that region doesn't make sense anymore, so we're gonna reinvest it somewhere else. Yeah, that happens all the time in this market. That's that's the way this industry moves, and because it's moving so quickly right now, that can cause, I'd say ripples, but it's probably more rather large waves across the industry, which gets people thinking.

10:53SPEAKER_02:So yeah, a bit fascinating when I look back and thought that has all happened within the last three months. You know, we've gone from everyone panicking that was the uneducated panicking, I guess is a better way of putting it, um, to now people looking at it going, oh, CapEx is increasing again and again and again. And they've all said as well that CapEx will increase further in 2026. They're all saying that in their earnings reports. Things could change, obviously, but that's what they're all saying right now.

11:19SPEAKER_02:So the bubble has not burst, breaking down.

11:21SPEAKER_01:Definitely not. Uh I I think that you raised an interesting point there around like that money is ultimately going to look for a home whereby it comes back the quickest, right? And I think one of the other things that I've seen come up a fair few times over the course of the past few months is shifting market dominance, especially in Europe, right? So Flap D, historically, the dominant markets. Not necessarily the case anymore, given the trends. Um why is that?

11:53SPEAKER_01:Well, like the UK, for example, grid connections, the time that you have to wait to even get plugged into the grid before you can even start powering things up now is absolutely massive, right? And if you've got a big old pile of cash that you need to do something with, the longer you wait to not do something, one, the market's gonna change in a way that you probably can't anticipate, and two, you can't have big piles of cash sat doing nothing for extended periods of time.

12:24SPEAKER_01:So you're gonna go wherever, one, power is cheapest, and two, you can get access to that power as quickly as possible because that's gonna reduce the amount of time it takes you to start seeing a return, right? And I think that the the historical markets, the flat D markets, they're incredibly built out at the moment, which means that you're looking around the edges for whatever little bits of power you can get here and there versus Norway, for example.

12:50SPEAKER_01:Again, starting Norway. Why? Well, there's 230 megawatts that they're starting with, with the option to potentially extend that up to 520. I think it was good luck finding 520 megawatts anywhere in the UK. I mean, even in Aggregate, good luck finding that much power.

13:08SPEAKER_02:Yeah, I think the the AI growth zones in the UK is interesting, isn't it? Because you had Manchester, I think, was one that was announced fairly. Teesworks. There's Teeside, there's Yorkshire, but the Manchester one is I think around 1.2 gig or something. Again, don't quote me, it's one point something, but yeah, you do think when, you know, like yes, but when, yeah, and that's and that is a challenge, I think. But you but what you've got, I had an interesting conversation with uh a CEO uh developer platform yesterday, and the the way that customers are looking for different size, so you've got multiple customers looking for different things.

13:45SPEAKER_02:So you've got customers like OpenAI, Stargate looking for that in Norway, that's what their market is, but you've also got customers looking for the the larger city-based facilities, you've got customers looking for new regions, you've got customers looking to develop in regions they're already in. So it's quite a dynamic market at the moment. So it's um, I guess having that portfolio, the ideal portfolio would be across multiple regions, multiple sites that can sol that can serve all the customers.

14:15SPEAKER_02:That would be the ideal if you're an investor to have all these different portfolios of land that can say, you know, open AR, yeah, I've got this one over here, you know, Microsoft. Yeah, I can I can do that for you here. But obviously that's not realistic. But I think you've got so many customers now looking for different things, which has caused quite a lot of dynamics in the market. See, that's why you're seeing more and more developments, because um, it's not just the obvious that are developing, there's multiple new customers coming in, like again, like a Stargate that have a demand for a high capacity in a different region, and now all of a sudden, those are key customers in this in the data center world.

14:50SPEAKER_02:Whereas go back, you know, however many years you want to go back, two, three, four, five years, it was quite a small select bunch of customers. So we're seeing that dynamic now play out, I think.

15:02SPEAKER_01:Yep. And I think how the market evolves in terms of the kinds of workloads that are being run as well is also going to have a massive impact on what the market looks like going forward as well. So up until this point, a lot of it's been dominated by training, right? Which is a massive cluster somewhere that has a massive amount of power and latency, proximity to urban centres, and ultimately proximity to a user base isn't necessarily quite so important.

15:28SPEAKER_01:Um, but from an inference standpoint, and especially as we start seeing use cases coming to market, they're going to need to be that much closer to urban centres because ultimately that's where the user base is going to be, right? And inference from a revenue curve standpoint versus training, it's potentially massive depending on the model that you're taking. Um, if you've got a thousand customers who are all connecting bar and API, the the ROI metrics, depending on the model that you're serving, are absurd compared to just charging per GPU per hour, right?

16:04SPEAKER_01:And for that, you want to get as close to the customers as potentially possible, um, which then means the little one megawatt here, two megawatts there, 500 kilowatts here, then becomes really, really, really interesting because you can get basically in the basement, right? Um, and there is a not insignificant portion of very large portfolios of commercial real estate at the moment that are in a bit of a hole because rates have changed, because work dynamics have changed, people aren't coming back to the office, and all of a sudden people are losing money hand over fist, but they've got a building that has a massive amount of power being delivered to it that just isn't being used at the moment.

16:48SPEAKER_01:Couple this with net zero, right? Um, and you're looking for ways to reduce emissions. Well, if you've got effectively a heating plant in the basement in the form of a few racks here and there, and you're still paying to heat the building as well, that doesn't make sense because you're producing two lots of heat for the same purpose, potentially, right? So there's ways that you can close that gap along with bringing the capacity closer to the customers.

17:17SPEAKER_01:That's a very, very different market again. And then it's not a case of right, I need 1.2 gig here or 500 megawatts there. It's I want every single basement in this particular part of the city, and I'll figure everything out there after. Now, is the planning system set up to do that at the moment? Absolutely not. And this isn't just in the UK worldwide, it's a very different paradigm. Um but that's where it looks like the market could potentially be heading.

17:42SPEAKER_01:Because if you've got self-driving cars, if you've got robotics, if you've got other real-time AI applications, 30 milliseconds of latency is huge. Um and 30 milliseconds isn't very much, right? But if you've got real-time applications that are dependent on effectively data streaming between wherever the models are being run and wherever the models are interacting with the real world, every millisecond of latency is going to have an impact.

18:12SPEAKER_02:Yeah, and if I want my robot to do something, I want my robot to do something. I don't want it to large. I don't want to have to wait for my robot to go and get me a cup of coffee. I'm not going for it. But no, it's true. No, it takes side, it is it's true, and I've talked to some uh investors around that as well, like looking at hotels, and you know, there's hotels now being designed for with data centres in the basement, basically. And like you say, the the heat, um the heat aspect of it, the energy aspect of it, you know, the water aspect of it, there's loads of positives that can come out of that once it's a feasible way of developing, basically.

18:51SPEAKER_02:But yeah, definitely need edge. It's not all about the the massive one gig, two gig, three gig, four gig campuses that are now being announced. The um that edge, or whatever you want to call it, closer to source is going to be even more important as we move into that technological revolution which is uh happening. And I want a robot. I want a robot as quickly as I can. Yeah. I saw the ones that is it in China. I saw some robots the other day that were on um I saw them on X or somewhere, robots that are literally like$20,000 lifelike robots now.

19:24SPEAKER_01:But honestly, cheaper than that. So this was the um the unitary one they've just done. It might be, right? It was doing like cartwheels and stuff like that. Yeah, that was it. 6,000. 6,000 US. So it's cheaper than cheaper than a high-end Mac. And who doesn't want a robot that can do cartwheels? Exactly. I mean, it can't do my washing, it can't do any of the jobs I don't want to do, but it's basically one of my children.

19:47SPEAKER_02:They can't do any of the jobs I want to do, but they're very good at cartwheels. Maybe that maybe that's what they base it on. Another um one of the large ones that I sort of Meta, I think, is quite a cool company to pay attention to. I I find Meta quite fascinating. You know, Facebook is a very simple um product when you actually think about it, but then they're making more and more from advertising now, which is as a result of AI. But then people always say, Why are they investing all this capex in AI?

20:13SPEAKER_02:What's the use case? Well, think about how they're making more and more profit now through AI, through their advertising. But they had that one gigawatt development, is it Prometheus? I think it's called, and also the Hyperion.

20:26SPEAKER_01:Hyperion, yeah.

20:27SPEAKER_02:I like I liked the um the way that Zuckerberg announced it by plonking it on Manhattan for as an image just to show the scale of it. And and I still get messages from people now because that post still doing the rounds on LinkedIn saying, are they really building that in Manhattan? And I just leave it. I think I can't I can't comment on it. But the way they marketed that was pretty cool, it got loads of interest.

20:50SPEAKER_01:Absolutely, and I think also the naming of those sites is really, really, really good as well. So, for anyone who's not familiar, both Prometheus and Hyperion were titans in the Greek Pantheon. So they're the the previous generation of gods prior to Zeus and that lot of gods. Gods of the gods, gods of the gods, exactly, right? So Prometheus stole fire from the gods to give knowledge to man, right? Which is then why he got tied to a rock and was damned for all eternity to have his liver eaten on a daily basis while he was still alive.

21:25SPEAKER_01:Um so there's some very, very clear, yeah, right. Very, very clear naming there, like stealing fire from the gods in the form of knowledge. That's okay. I like that. Hyperion, if memory serves, was Prometheus' father, and Hyperion was the light bringer as well. So you've got a five gigawatt cluster called the lightbringer, and Prometheus, which is C5 from the God. I I've got moral issues with Meta as a company, but whoever came up with that naming convention in this particular context is Prometheus.

22:00SPEAKER_01:They deserve a raise because it was it was good.

22:02SPEAKER_02:Yeah, they're all they are fascinating to watch, aren't they? Meta. They're they're an interesting one. I always read their earnings calls and the light, and some of the some of the things they come out with, you know, it is interesting from uh understanding where these people think the market's going. And that's kind of what I take from stuff. It's trying to listen to these people that are you know obviously hyper-intelligent, they know what they're doing, but to see how their views differ on where the market's going.

22:27SPEAKER_02:Because they do differ, but they're all they're playing with you know tens of billions of dollars to try and predict the future before someone else.

22:37SPEAKER_01:That and the paychecks for the oh yeah, well, that's the other thing that's come out, isn't it? That's the newly formed superintelligence unit. So I saw a recent video from Zagobo. The the big picture play is an individualized superintelligence for every single person, right? Um, which will know you better than you know yourself. Um I'm sure there's probably some commentary that you could have there around creating a product which destroys human connection and then creating an alternative to fill the gap to solve the problem that you created, but that's a different conversation.

23:07SPEAKER_01:But the fact that they're going around offering like one billion dollar, 1.2 billion dollar packages for someone for a four-year period, right? Like the only numbers that we have a kind of comparison with are footballers at this stage, right? It's like it's a four-year contract for X number of hundreds of millions of dollars. And but these are now going to like AI researchers, right? So it's it's insane to be quite honest. I have I have no idea what to think of it.

23:37SPEAKER_01:But I mean, fair play, if you're an AI researcher and you get someone knock on your door saying, right, do you want a billion dollars to come work for me for four years? Like, that's a difficult thing to say no to, unless apparently you're someone who is working at thinking machines, um, who's the previous so's the name? Previous CTO of OpenAI spanoff set up thinking machines. Apparently, um Zuckerberg has offered basically competitive offers to like 25% of the team or something like that, which are all in the hundreds of millions, millions of dollars, and every single person has said no.

24:12SPEAKER_01:Every single one has said no. Um so there's a bit of that going on as well. And I mean, fair play. But in fairness, they've just raised two billion at a ten billion dollar valuation with for a seed with no products and no roadmap. Um, so hey, why not? We could try that.

24:31SPEAKER_02:Like I said, the news and GPUs, news and GPUs, C funding. We're out, we're public, we're going public right now. Contact either of us.

24:42SPEAKER_01:We haven't got any ideas, but that's fine because we've already validated that that's an idea and in the workself.

24:47SPEAKER_02:Exactly, exactly. We we can we can help you predict the future. Um, but yeah, no, like I I always just think of recruitment fee when I see when I see things like that. I think Imagine. Imagine someone's someone's doing very well out of it. But it also does um again paint a picture of where the skills are going. Because again, skills are really important, and like the future skills, understanding what my children, what skills they'll need is really, really important.

25:13SPEAKER_02:And I think the the salaries always give you a picture of that, because the the reason why the the money is so high is because there's so few of these people, because it's a new skill. It's like you know, they they have whoever these individuals are, managed to learn this skill ahead of the ahead of the the world, really, ahead of it, the industry. Therefore, now they're you know they're the only ones that can do it.

25:33SPEAKER_01:And yeah, yeah, and it also speaks to the power of specialization as well, right? Like these individuals are so deep in their little niche that no one can do what they do, no one can do what they do, right? Um, and the deeper you go into a niche, the less people you're competing with, right? Until ultimately it becomes a market of one. Um and that parallel also works from a platform differentiation standpoint as well, right? Um I've been pretty vocal about how I think bare metals are commodity market.

26:04SPEAKER_01:Um, and there are two ways that you can deal with that. You can either go down the stack and you can operate your data centers so you've got a little bit more play in terms of your pricing, or you can go up the stack and start vertically differentiating and go into a niche, right? Start serving a particular customer base, solve problems for that specific customer base in a way that no one else can do it, and then all of a sudden you're not necessarily competing in quite the same way.

26:26SPEAKER_01:It's the same thing across the board. Um, go deep enough into a niche, and then life becomes a hell of a lot easier. Totally agree. I remember with data centers with a niche.

26:36SPEAKER_02:And the niche now the the funny thing is, some people still say to me, Yeah, that's quite niche. I think well, do your research, do your own research, as they say. But yeah, it's not uh not quite as niche as it once was, and I'm sure some people listening understand that as well. It's uh again, the scale is mad. Other thing I want to talk about, Scotland. Um finally, in the last quarter, we've actually seen announcements of developments going in for planning.

27:03SPEAKER_02:Uh, Apertura with a I think around 300 megawatts it was. So Scotland is actually rising to where it hopefully should be, uh, which is a pretty key player in the industry based on everything in Scotland with regards to the power, but also talent. There's so many great people that work in the industry from Scotland. It's also got oil and gas talent, which again needs repurposing. So that's an interesting move from a UK perspective to see uh finally some developments actually going through.

27:32SPEAKER_02:Whether they will happen, we'll have to wait and see whether the Scottish Government will push them forward. We don't know, but it's good to see that it's uh now hitting the press at last.

27:41SPEAKER_01:Especially from a power infrastructure standpoint up there as well, right? I I've seen that there's a few conversations going on around, I think it's from Octopus Energy driving it around zonal electricity pricing. Yeah. In that if you're consuming power close to where the power's being generated, you get a better rate, effectively. I mean, if if something like that goes through, that's a great way to bolster economic development in places that have perhaps not seen the case same benefit of the past 20, 25, 30 years or so, right?

28:14SPEAKER_01:But then that very much revolves around the um the assumption that the UK government is going to be willing to um relinquish some power from London. Um and I I don't know, I don't know about that.

28:30SPEAKER_02:It's a nice nice idea in principle. Yeah, but with the UK, you see again you it goes full circle all the time. Like we want to move all as in data centres, we want to move away from London, yeah, or we'll start announcing them elsewhere, then it's oh no, actually we want to go back towards the London region. Yeah, exactly. And it goes back and forth. So it's good to see these conversations happening, and as I say, and then whether it's uh Scotland or the rest of the UK actually announcing, you know, that there's developments being put into planning, but like you know, the the proof is in the pudding, yeah.

29:00SPEAKER_02:Like when these happen, is there is a demand for them as well? You know, are the customers out there, do they want these facilities? I know those conversations will be happening with the developers, but until we see that move, we we have to just assume that it's um you know could be or couldn't be.

29:14SPEAKER_01:That's honestly that's the biggest issue I have with the recently announced Teaseworks um AI growth zone in the it's a massive site, but one BP we're also supposed to be doing something from a hydrogen standpoint on the site. And I know that there are some clashes going on as to well, we can't do both with this site, right? So which one do you want? Do you want net zero um with the hydrogen production, or do you want a data center? Because it seems like there's some incompatibility there.

29:45SPEAKER_01:It's a really long way away from large amounts of people, right? Which ultimately are going to be the ones that are consuming like a site of that scale, the number of users you're gonna have to have if it isn't for training, if it's for something else, or if it's for another paradigm entirely, then it's just it's a long way away. Like the latency, it again, it's not huge, but is it enough? Is it slow enough that all of these real-time AI applications end up having an issue?

30:18SPEAKER_01:And like I get it, because it's a big site and there aren't very many of that scale, but it's a long way away from everybody. So it is it potentially commercially viable in the same way as some of the other sites that are going to be that are closer to urban centres are going to be. I don't know, to be quite honest. Um, I would love to have been a fly on the wall to understand how that decision making was made, because like Cullum makes sense, right?

30:50SPEAKER_01:It's in the the arc corridor with Oxford Cambridge, it's all in that area, and that makes sense. But like that's a long way away.

31:00SPEAKER_02:But yeah, Cullum, like you said, perfect sense. Power station. And again, and you start to see those ones move through, and I think uh a lot of developments around power stations or you know similar industrial facilities is is where you're gonna see where the power's already there, it's it's obvious, but it's a big redevelopment project, but it's definitely the definitely the right move. Other thing I was thinking, what have I been up to in the last few months?

31:25SPEAKER_02:Went to Cannes, went to Data Cloud, which again seems like ages ago. I don't know if you've been there.

31:30SPEAKER_01:I saw the picture of that. Look like you've been working really hard there. Yeah, looks like a really, really difficult conference.

31:37SPEAKER_02:I did work really hard. I actually worked so hard I pretty much lost my voice. Um, but yeah, it was Manic, absolutely manic. I think again, I always say it, events paint a great picture, ready. Industry sits um more and more people every year. They they broke records as an event for attendees uh into the event. You've obviously also got thousands of people outside of the event. So again, another another example and one of those moments when you think, wow, like this this industry is is booming right now, but it was a it was a fun event.

32:08SPEAKER_02:Have you been anywhere?

32:09SPEAKER_01:I know you've been in the US, but being anywhere, GTC Paris was the um was the big one, the last big one, and that that was really, really interesting for a couple of reasons. Um from an announcement standpoint, relative to GTC in March in San Jose, nothing massive, really. Um nothing game-changing to use the overused AI terminology. Um but the it it became obvious just how dense the talent in Europe is, right? Like, because VivaTech that it was bolted onto is massive, absolutely massive conference, and then GTC was kind of bolted onto the side of that, and then seeing the density of talent, the density of companies that are actively exploring opportunities in Europe and the scale of the opportunities that are now coming to light, right?

33:03SPEAKER_01:Like with the the Stargate um projects in Norway with NScale that was announced today, massive, right? So progress, I think, is the biggest thing there, and like you've got this massive talent density, and now it's clear that like the market is validating the quality of that talent as well. But I think honestly, the more interesting thing there was the timing of GCC Paris because on the same week AMD's Advancing AI conference happened as well. Um and GTC Paris didn't exist like six or seven weeks prior to it happening.

33:38SPEAKER_01:And then I think on I think it was the Tuesday of GTC Paris week, NVIDIA put out 18 press releases. Um, and then Jensen then spent most of the day on the show floor taking photos with absolutely everybody, um, which meant that by the time Thursday rolled around, which is when all of the AMD announcements started coming out, everyone's social media feed was just full of Jensen with pick taking pictures of people.

34:06SPEAKER_02:He is a marketing genius.

34:07SPEAKER_01:I say, I 100%, 100% agree, but I think that the nature of AMD's announcements, um, especially with the the rat scale systems. So there was the ZT systems um acquisition that went through relatively recently, and then that's ultimately going to feed through into the MI400, MI450, Rat Scale solutions and the Helios solutions, right? Which are big, big systems. We're talking like two kilowatt cards potentially, some of these, which is mad. But the the benchmarking and the performance, and ultimately Rockham, the software ecosystem, um, that AMD have been slowly updating, and now it's starting to get to a much better level of usability, um, is getting there, right?

34:52SPEAKER_01:And you've got NVIDIA with 92% market dominance, AMD with 8% market dominance, Intel not shipping enough GPUs to even be seen as a rounding error in terms of the overall percentages. I mean, 92%, there's realistically only one way that's going to go. And I've probably said this before on the previous ones, but I think with where AMD is at, that's what the most interesting thing is going to be. So with GTC Paris, my biggest takeaway was that owing to the PR strategy and the fact that there wasn't any massive, massive announcements.

35:29SPEAKER_01:Maybe there's some internal discussions going on around like what does AMD mean for the market? And then also challenges silicon, like the alternative semiconductors as well, right? Saw the Furiosa AI, the um the South Korean semiconductor manufacturer who Mesa tried to buy back in, I think it was April for$800 million, and they said no to. Um, they've just landed LG as a customer as well. Um, and I think they've also just closed a funding round as well.

35:58SPEAKER_01:Um so they're obviously doing something quite interesting. And they're 150-watt PCIe carts with better token throughput performance than Nvidia's L40S. That's quite interesting. And if inference is ultimately power constrained in the same way that training is power constrained, but you're getting better performance from a token throughput standpoint with some of this stuff. That's very, very different as well. Now, that is not to say that NVIDIA are going to be struggling in any way, shape, or form at the moment because the market has spoken and the market continues to speak in a very, very loud manner.

36:31SPEAKER_01:But I think over the course of the next probably year or so, we'll start seeing a little bit of a change. Um I mean, if they can maintain 92% market dominance, then fair play.

36:42SPEAKER_02:And think about the market as the size of the market, isn't it? I know it's obvious, but you know, the market is is growing and growing and growing. So maintaining a high high percentage.

36:53SPEAKER_01:$4 trillion company. NVIDIA's a$4 trillion company. Um Microsoft also, I think, crossed$4 trillion, I think it was last night, um, in after hours trading as well. So we've gone from low$4 trillion companies to two very involved in AI over the course of like a week, two weeks or so.

37:12SPEAKER_02:And you've still got people asking five in the next six months. And when you listen to the earnings calls or read the earnings calls, you've still got people saying about well, how are we going to make money out of AI? We think, well, if they're not making money out of it now, it's only going one way, isn't it? Like if if and obviously there is the question, will they? But you know, I think we're pretty confident there'll be multiple use cases that'll be profitable.

37:35SPEAKER_01:But yeah, yeah, and it's not just chatbots as well. No, that's what I mean. That's the that's the really, really big thing, and like things like GPT-5 potentially, which is supposed to be coming out in August, right? That's potentially going to be another step change. And yes, there are problems with the LLM paradigm, and then there are some unsolved technical issues around things like prompt injection and that sort of thing, which mean that from a security standpoint it's not great, but it's not just text that this stuff works for, right?

38:05SPEAKER_01:So have you heard of a company called Basecamp Research? No, Basecamp Research have basically been out in the jungle, literally, just out there collecting biological data. Um, they have the biggest, deepest, most comprehensive data set of basically the largest chunk of the entire tree of life of anyone out there at the moment, and this is like baseline genomic data. They're now starting to look at ways to monetize that data as well. And if you start throwing that kind of data into model training and that sort of thing, that's that's absolutely massive, right?

38:47SPEAKER_01:Like from a healthcare, from a medicine, from a just general understanding of the world around us perspective, that's absolutely huge. And like it's it's stuff like that that's just starting to come into the I guess into the media ecosystem for lack of a better phrase. It's been going on for years, right? It's just people haven't been talking about it because everyone's thinking, oh wow, this chatbot's amazing, right? Um and chatbots are interesting, yeah, but it's not necessarily the most exciting.

39:12SPEAKER_01:Um, there's other really exciting stuff going on in the background, and I think that that's the kind of stuff which we're now starting to see come to light. And also like AI today is the worst it's ever going to be.

39:24SPEAKER_02:I know, and I always think it just gets better. I always say that, and I still think most people don't even know what AI means. They just say it.

39:33SPEAKER_01:I mean, in fairness, it is a marketing term, right? Um go it goes way, way back. Um, I've read was it um Karen Howe, she wrote the Empire of AI, I think it was the kind of expose um interesting book, and artificial intelligence was basically a rebrand of something like automata studies or something like it was some generic thing that the public didn't understand. They rebranded it to AI, and lo and behold, everyone starts talking about it, right?

40:01SPEAKER_01:Like, same with AGI as well. It's uh what does that mean? Or artificial superintelligence, and it's you've got competing definitions for these things which ultimately are PR for a given company, which is pushing their vision of what they think is going to be. And I think it it's really difficult to understand what does it actually mean, right? Because the the lines start getting blurred, but whatever it is, it is getting better. Um, and it's getting better quickly.

40:27SPEAKER_02:Exactly. It's amazing, it's it's ridiculous what it can do already. And like you say, it's only the it's only the start of a very, very long journey, I'm sure.

40:35unknown:Yeah.

40:36SPEAKER_02:Anything else before we cut this off, anything else over the last three months that should be a month that's that's taken to hit you in the news?

40:46SPEAKER_01:It's I think data center adjacent. I think one of the things that's interesting at the moment is what's going on in the power space. Um in terms of There is not enough. There's it well, there's not enough for one, but then also the the speed of development of new technologies in that space as well, right? So you've got, I think it was an AIDC grid in Korea, um, which was commissioned or announced as a project recently that's looking into uh superconducting grids, right?

41:19SPEAKER_01:So superconductors for a grid for AI data centers, and that means I mean, lower lower opex, lower capex, higher efficiency. That's huge, right? Especially if you're someone who's looking at grid upgrades at the moment, that's gonna be absolutely massive. Also, Hedeon, um, they're a fusion reactor manufacturer. Um was it Microsoft they signed a deal with recently? Yeah, they've just started building the first commercial fusion power plant on the planet, and like that's huge.

41:47SPEAKER_01:Um, and the fact that Microsoft have come in as a customer or whoever the hyperscaler was is I mean, it's pretty validating in terms of what they're building and how they're building it seems to be interesting because I mean fusion's one of those technologies that everyone jokes is always like 20 years away, right? Apparently not, it's 2027, that's when it's that's when it's hitting. So all of this development that's happening as well, and then obviously the resurgence in nuclear and then all of this stuff as well.

42:15SPEAKER_01:It's it's all in support of ultimately AI, right? And this massive, massive infrastructure build out where that goes.

42:24SPEAKER_02:Yeah, AI has caused the governments to it to move. The problem was there, wasn't it? You know, like we we we've experienced it just from paying our bills, and I'm sure that's no different to most people. You know, there's a huge demand for it, populations are increasing, but AI has made them actually have to do something because they now see that they're gonna miss out economically significantly if they're not able to move move fast enough. There's a few um, I know a few off-grid developments that will be announced fairly soon for data centres that are in in Europe.

42:56SPEAKER_02:So yeah, it's definitely um moving in the right direction. There's some pretty cool stuff happening out there, but like you say, it's um they have no choice. You know, if you want to build at this scale, you need to find a power solution because it's not as easy as just finding the power. If it if it was that easy, we'd we'd all be doing it.

43:12SPEAKER_01:Yeah, right. Predictions for next time for you, what do you reckon?

43:17SPEAKER_02:Yeah, I was thinking about this earlier when I was listening back to the predictions from from last time. I I don't really see anything specific. I think we'll just continue to see what we're seeing right now, which is major campus development announcements across the globe. And they will similar to the Stargate Norway one that we've mentioned, similar to some of the ones in the US that are now in new regions, um, that scale of development, let's say one gigawatt plus in the US, two three hundred meg plus in Europe, I think you'll find a lot of these are going to be announced in newer regions.

44:00SPEAKER_02:um Scandinavia, Iberia, obviously in the US, it's you know pretty much anywhere where there's not already a a big amount of data centers, which is Virginia largely. But you know, Texas continues to see the significant growth. But I think again we're gonna see a move away from Texas soon because it's becoming a bit saturated uh with with announcements. But yeah I don't I I don't see anything changing a lot. I think in when we talked about it in April it was pretty clear market sentiment where it was.

44:33SPEAKER_02:A lot of what I'm talking about now or hearing in the conversations is just more of the same. I think the CapEx is filtering through for next year already that it's going to be similar. You started to see the hyperscalers move on the projects because a few have been delayed this year so that'll start to move. But other than that I shouldn't say it but I don't see any massive curveballs coming imminently so expect a curveball next week.

45:00SPEAKER_01:Alright yeah you anything I I tend to agree I think the the thing will be as I said earlier I think we'll start seeing some policy and ideological alignment starting to happen as to write which other countries are buying into to which vision for the future um I think that's what is probably going to be the next thing because the the AI Act is now enforced but enforcement starts on the 2nd of August I think it is it's article 51 I think it is enforcement starts on the 2nd of August.

45:35SPEAKER_01:Google have just announced that they've signed the voluntary code of conduct um Meta on the other hand have just outright refused and said no we're not doing it. I think we'll start to see an impact um with enforcement and I think that might change dynamics a little bit because I remember six months ago having conversations with various neo cloud providers or anyone on the enterprise at certain point it was on the one hand we don't need to worry about certifications because it's expensive it's not necessary at the moment and then the enterprise being like we can't engage because there are no certifications right those two things seem to be coming together.

46:14SPEAKER_01:I've seen that a lot of the bigger neo clouds have come up with um SOC 1, SOC 2, ISO 27001, FedRAMP all of these certifications have started to happen relatively recently and all at the same time so I wouldn't be surprised if we start seeing some bigger enterprise adoption um outside of the the standard um if we start seeing some trickle down bigger announcements into maybe the larger SME space outside of the the FTSE for example or the Fortune 100 500 um I think we'll start seeing some wider adoption in that regard.

46:50SPEAKER_02:And then also alternative semiconductors more more market share we need a lot of them yeah so you know worth it yeah we need them so yeah it's gonna cut like that's that's when innovation happens out of necessity isn't it so yeah it's gonna come it's definitely coming and I think um the other thing I was gonna say is I put a post the other day about doing a podcast inside data center podcast you know I've great name I've decided that uh podcasts are much more fun with a beer in a pub um so we're looking at doing that end of August I'll get a day hoping can get Ben along for a for a chat as well but come back on that one and obviously then it's event season again September bang back at you DCD London on platform on Teeves Capacit Data Cloud in the USA DCAC in the USA we could go on and on and on so I'm sure we'll be out on the road somewhere and we'll catch up with with some of the listeners but hopefully we can get a we'll do the pubcast and then hopefully we can get another episode in probably I'm not even gonna say a month anymore because it's been in the future end of September I reckon end of September so in a couple of months but um I know a lot will happen in that time period so we'll have a lot to chat about but if you want to reach out to us you you listeners you know where we are come and find us always happy to chat happy to help if we can if you've got any news fire it our way we'll get it in the in the newsletters keep the GPU and weekly data centre news uh they're always out every week without fail pretty much unless unless nearly at 2000 subscribers now it's not far off nearly that's it yeah keep keep growing keep growing but yeah definitely go and have a look if you're interested in the GPU side you know check out check out Ben's newsletter well worth the subscription cool I'll leave you to it for the day it's Thursday today 31st July we'll get this out early August so it'll be out next week so hopefully not a lot's happened between now and then but if it does we apologise we missed it we'll we'll get we'll get it in the next one beautiful thanks for having it's been a good conversation and again like there's never a dull moment this day I'm sure we'll have plenty to talk about the next one.

49:03SPEAKER_01:Exactly always something happening have a good day.

49:06SPEAKER_02:You too all good