Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Inside Data Centre Podcast

Inside Data Centre Podcast artwork

Data Centre News & GPU's - Episode 1

Feb 21, 202536 minAudio

Send us Fan Mail Data Centre News & GPUs - Episode 1 Welcome to Data Centre News & GPUs, the new monthly podcast where Andy Davis & Ben Baldieri break down the latest developments in the world of data centres, GPUs, and AI infrastructure. In this episode, they dive into: NVIDIA - The latest moves from the GPU giant. Deepseek - What was that all about? Stargate - A closer look at this groundbreaking project. Paris AI Summit - Key takeaways from the event. LEAP 2025 - The major announcements. Hyperscale CapEx Spend - The numbers behind the data centre boom …and much more. Tune in now for expert insights, industry trends, and everything you need to know about the future of AI, data centres, and GPUs. Listen now on your favorite podcast platform! Support the show The Inside Data Centre Podcast is recorded in partnership with DataX Connect, a specialist data centre recruitment company based in the UK. They operate on a global scale to place passionate individuals at the heart of leading data centre companies. To learn more about Andy Davis and the rest of the DataX team, click here: DataX Connect

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

SPEAKER_01Welcome to the Inside Data Center Podcast. A bit of a different one today. The first edition of a new monthly podcast that I'm going to be doing with Ben Vallieri, who's on the call. I've called it Data Center News and GPUs, but I am open to other ideas. So if any listeners have got any ideas, and I'm sure me and Ben will talk about it as we work our way through the news over the next few months. But what I wanted to do is create a podcast for news. And Ben's very active in the GPU space. I'm obviously on the data center space, and I thought it'd be good for us to get together, try and share some of the news because there's been a lot of news this this month, and cover that and obviously directly towards the newsletters where you can get more detail on it.

SPEAKER_01Ben, do you want to give a quick introduction to the audience of who you are and why you're my chosen GPU expert?

SPEAKER_02Absolutely, thanks, Andy. So Ben Balvieri, I'm the founder of a newsletter called the GPU. I recently managed to snag the domain the GPU.ai, so it's nice and easy to navigate to as well. I'm also technical sales business development marketing at Panche Air Limited, and our customer base is the GPU as a service platforms. That's who we're targeting. That's who we've been working with over the course of the past couple of years. So from that position, um got pretty good understanding of what's going on in this niche within a niche, within a niche, within a niche.

SPEAKER_02And I mean, as you said, Andy, right, it's been a it hasn't been a slow start to the year, to be quite honest.

SPEAKER_01It's been absolutely crazy. And I think me and Ben have been talking about doing this for a for a few months, and I think we're trying to do it at the end of January, and the news kept coming, and it was we just had to find a point when we cut it off. So it's the 14th of February as we release this. Happy Valentine's Day to the listeners.

SPEAKER_02Happy Valentine's Day indeed, yeah.

SPEAKER_01We'll get this out next week, so it'll be out relatively quickly. But there's been so much news. Um, how we're gonna do it is me and Ben have picked five-ish stories from the data center world and the GPU world. We're gonna just run through them, whistle through them, have a bit of a chat. We're open to comments. And you know, if anyone's got any news that they feel we should mention next month, feel free to reach out to us. But that's how we're gonna play it. We'll see how we get on, and hopefully this will become a regular thing. So what I did when I was trying to work out five stories from the 5,000, multiple hundreds of billion dollars of announcements that have happened so far this year, I thought I'll go with my LinkedIn page and the five most interactive posts that I've done, taking out all the ones about the images that everybody loves uh interjecting on.

SPEAKER_01Top of the list was a recent announcement. So the UAE to invest between$30 and$50 billion in France to build Europe's largest AI campus. But then when I was looking at France, obviously we've just had the AI summit um this week. We've also seen new build announcements from Evrock,$20 billion an investment by Brookfield Data 4, Prologis, 3.5 billion euros on powered shells, Fluid Stack, a one gigawatt AI supercomputer, mistral to invest several billion in a data centre.

SPEAKER_01And I could go on, but I can't because it would take too long. But I guess what we're saying is in the last, I think we're only at two weeks or 10 days, we've seen well over$100 billion of announcements alone for France. So that gives you an idea of the scale of the news that's out there at the moment. And I know some of them are linked to AI as well, Ben. So I don't know if you've got any opinion on any of those.

SPEAKER_02Yeah, I mean it's it's interesting, right? Because if you look at what's happened over the course of the past year or so, the criticism that tends to get levied is that Europe's a bit slow, Europe's not necessarily catching up, Europe's not competitive, and this tends to revolve around outside of Scandinavia, to be quite honest, and the power pricing, right? Because power is not the cheapest in Europe, which makes it really difficult from a GPU as a service standpoint, certainly, to manage the OPEX effectively, especially when margins are being compressed as quickly as they are if you're offering GPU per hour.

SPEAKER_02But then, I mean, yeah, as you say, the AI Action Summit happens, and then all of a sudden it seems like there's a bit of a shift in the center of gravity, and I think that's pretty much how you could summarize the past few weeks, to be honest, right? Because start of January, you've got lots of things going on in the US. Trump administration is coming in, and you've got the Stargate project which gets announced, so you've got$500 billion, which is being thrown into just phase one there, and then Deep Seek happens, right? And you've got trillion dollar stock market routes, so you've had a central gravity shift to China, and then this week has happened, and you've got multiple hundreds of millions, billions of dollars being announced of investment in France, and it's just we've almost gone full circle.

SPEAKER_02It's like one rotation around the globe in one month when you think about it, and it's where does it stop? Right? Like, is this really kind of just the beginning? Because if you look at how the cluster sizes have increased, I mean, things are happening logarithmically, it's not even exponential, right? And if this is what's happening now, and this is hundreds of billions of dollars, is it gonna be another order of magnitude over the course of the next couple of years? And if so, where's the liquidity coming from, right?

SPEAKER_02And I mean, I think we've probably got an answer for that already, in that the UAE are doing a hell of a lot at the moment. Saudi is starting to wake up a little bit as well in terms of where the investment is coming from, and it's just I mean, it's it's exciting if nothing else. But something it's already cost me my hairline, right?

SPEAKER_01So I hope. And then this week I saw a video of Sam Altman saying, you know,$500 billion, and he said, give it a couple of years, we'll be we'll be raising five trillion dollars. And again, I think if whether that's true or not, I don't know. You know, people much more intelligent than me can can have an opinion on that, but that gives you an idea of the scale. And then on Saudi, I also jotted down quickly before I before we did this call. You know, we had Leap this this week, Leap 2025. So you've got$20 billion of investment in Saudi, including a$5 billion data vault data center.

SPEAKER_01You've got Equinix, developer$1 billion data center, and various others that have had that again that have been announced this week. It is mad the week that we've just had, but every week this year has been like that. So what about GPU world? What's your what's your main one?

SPEAKER_02Yeah, so I mean, I think the the fluid stack announcement is obviously a really big one, right? Like one gigawatt campus is huge from a GPU standpoint. Um it's what was the announcement? They had like 500,000 GPUs online by 2028, but then that wasn't the only one that came out this week either. You've got Cistus also announcing that they've got one and a half billion dollars of investment planned for coming in as well. And that's an interesting one because that's uh it's also a French company, right? So it's two and a half billion dollars in France, um, or more than that one we're talking about.

SPEAKER_02500,000 GPUs is gonna be a hell of a lot more than that. Um just a massive amount of liquidity going into France. But then this is also off the back of like the AI um summit in the UK, right? And the action plan that came out with Matt Clifford and everyone, and then Nscales getting involved there too. And then Applied Digital are also, I mean, they've secured five billion debt facility from Macquarie too. It's just it it seems like some of the platforms are really starting to kind of step up now, right?

SPEAKER_02Because last year it was kind of dominated by a couple, really. So you've got your core weaves, your crusa's, your lambdas, and they were taking market share when it came to I mean, well, just mind share, right? You say NeoCloud, those are the ones that came to top of mind, but then January's rolled by, and then all of a sudden there are several more large players who are now entering the game at scale in regions that are outside of the US. Um and it's gonna be really interesting to see how it all progresses.

SPEAKER_02I mean, I think my favourite one off the back of that was the N-Scale announcement came out with the AI um action plan for the UK. Same day, press release from Core Weave. Our first classes were online in London as well, right? So it's it's all kind of happening at the same time. Um, it's just trying, I guess, maintain that mind share, right? Try and stay front of mind. Um just to see how things heat up, really, because it's all so new. And I I guess from like your perspective with the data center side, like a lot of these neo-cloud players, everyone wants them as a tenant, right?

SPEAKER_02Because the the rack densities alone, if you've got an 135 kilowatt rack and you're charging a hundred dollars per kilowatt per month, plus the power costs on top of that, that's potentially quite a lucrative contract to land. So getting those guys in is interesting. It's just gonna be I mean, who can support them? I guess it's gonna be the next big thing, right? Like which facilities are actually going to be able to land those customers, and if you can't support them now, how quickly can you update to actually be able to support them, right?

SPEAKER_02I mean, I'm sure you've probably got some more insight there than I have.

SPEAKER_01No, it's a good question. And I think um AI, there's still a lot of um new players that are going to come in the market fairly soon as well. Still a lot of capital um trying to mobilize into the market, and also the customer element's fascinating because nobody really knows who the customers will be. There's obviously the obvious, but then long term, some new AI platforms and AI software will come in and be a significant customer. So they're trying to evaluate the landscape of all these little startups as well, and think which one of these is going to need the capacity?

SPEAKER_01But yeah, it's uh it's an exciting time. And I think as well, that that you mentioned it earlier. My next one was Stargate, which brings us on to it. You can't really not talk about it. You know, we had um our friend Mr. Trump um stood on a platform, and I think uh some others stood fairly close to him, announcing a$500 billion uh investment from OpenAI, Oracle, and SoftBank to develop AI infrastructure in the country. And there's been a lot of um further announcements around the locations and they're looking at multiple areas and the scale of these, as saying I listened to quite a lot of Sam Altman and around you know five gigawatt sort of capacity that they're looking at for some of these campuses.

SPEAKER_01So significant investment, but you also touched on it as well. What it's done is it's triggered a kind of the snowball effect where everyone's in this AI race from a government perspective, everyone's in trying to win this race, and someone announces 500 billion, you suddenly see that kind of snowball around the world. And I and I think that's what we've seen from my perspective, anyway.

SPEAKER_02Yeah, definitely. And I think the the Sarget announcement's interesting, right? Because the the first campus is in Aberdeen in Texas, right? It's the the Lance and Clean campus. Um and the first data center operator, as far as I can tell from the information I've seen, is Crusoe, right? Who are previously Bitcoin mining, capturing energy from flare gas, moved over into the GPU compute space and the AI space, similar to Core Weave, similar to Applied Digital. Um, and it's interesting that ultimately a NeoCloud, right, has been chosen as one of the first providers over a more traditional player.

SPEAKER_02Um, whether we're going to see more of that kind of dynamic happening, I wouldn't be surprised if we did, really, because I think that's what the NeoClouds have shown they're very good at. It's sanding up massive amounts of compute incredibly quickly. Um Core Weaver is excellent at it, right? I mean, if you look at who their biggest customer is, it's Microsoft, right? Microsoft stand up to things pretty quickly, but then also they need more capacity. They're going out into the market, and then Core Weaver have got amazing track record in terms of deploying large amounts of compute, supporting these data centers.

SPEAKER_02They're very, very good at what they do, right? How does the more traditional market respond to something like that, right? Because from what I've seen from the GPU space, looking into the more traditional data center space, there tends to be quite a set way of doing things, right? Like this is the way that the industry operates, it's operated this way for a very long time. But I mean the tech alone, never mind the companies that are doing it, is the disruption here, right? Then you get the disruption on top that's made possible by the companies.

SPEAKER_02How does the rest of the industry respond to effectively these upstarts that are doing both sides of the equation? They're running the cloud and they're operating the facility. What happens?

SPEAKER_01No, I think it's um it's fascinating. I I've sort of been looking into the mining side of it for quite a few years and how that's transitioned into AI and HPC, and it was always the sort of ugly sister of the industry, Bitcoin mining. And I I understand why. And the facilities are very, very different, obviously, but making one of those facilities a tier three facility is not as complicated as it was for making it a tier one, and that's kind of where you've seen it. But I also guess it's the power they have as well. Like these facilities have you know multiple gigawatts of power allocated, and power's king at the moment, you know, or queen, depending on on what side of the fence.

SPEAKER_01But it's if you've got the power and you've got the ability to scale quickly, obviously that puts you in a much better position than if you're starting from scratch and you've got to allocate power, etc., etc. It's obvious, but that's kind of the the niche they've seen. I'm actually interviewing the um the CEO of Applied Digital next week on a pod. So it'd be quite interesting to see the journey that that they've been on from the early days of establishing an organization as a as Bitcoin mining, basically, to suddenly Oh no, it's that it's earlier than that.

SPEAKER_02Um they started with they started with potatoes.

SPEAKER_01There you go.

SPEAKER_02Something to ask him about. They started with industrial farming and potatoes way, way back when in the early 2000s.

SPEAKER_01And you go from there to mining into the AI HPC, and like you said, raising multiple billions of dollars and building a ridiculous scale facility just for that market. So, yeah, fascinating, fascinating world. Um, what's next on your list?

SPEAKER_02Well, so I think that leads quite nicely into the the first MBL72 systems coming online, right? The blackwells, the GP200s, USUs with the thermals to start with, with the first generation, from what we've heard. Um, but then let's be honest, it's not entirely surprising when you're deploying an 135 kilowatt rack. Um, that's a pretty significant power jump over the the previous generation, which was the H200, right? Like 10 to 11 kilowatts per box versus 135 kilowatts of rack at 1.4 tons, and it has to be liquid cooled at the same time.

SPEAKER_02It's it's going to be really interesting to see what that does to the market. Um, because what we saw last time, I mean, when I say last time, this is with every new hardware release cycle that we see from NVIDIA, not so much AMD at the moment, but potentially in the future with some of their new hardware that's coming out with the MI325X and the 345X, is that whoever is first to market with a new generation of hardware is in this really enviable position in that the market doesn't really yet know what the rate is supposed to be per GPU per hour.

SPEAKER_02Um, and if you have track record and you have the facility and you can meet your customer requirements, which is, to be honest, at this stage, probably going to be one of the Frontier AI labs, and you can get those systems online and you can get the allocation, and you're credit worthy enough to be able to unlock the financing to be able to deploy those systems at scale, you're gonna hit payback incredibly quickly. Um, we've been looking at a few opportunities in Europe at the moment, and we're starting to get some indicative pricing as to what the GPU price is going to be on an hourly basis, and the a one-year contract seems to be coming in at around about between$3.40 and$3.60 odd um per super chip per hour.

SPEAKER_02Right? 72 chips, 72 GPUs in one system, 36 combined chips um that are chargeable. At that rate, the um ROI time is very, very attractive. If you're able to lock someone in potentially for a two-year contract, you're gonna hit break even well before the end of that contract, from what we're seeing. If you're also then factoring in storage revenue, too. Um, so your core weaves who have got some stuff online, right? We saw that there was an announcement last week, they were the first provider to make general instances available, although technically Microsoft bought them online last year.

SPEAKER_02Um, that was probably the first generation, and I wonder if there was some thermal issues with that generation, which is why there wasn't necessarily quite as much press as there would have been otherwise about it. Um but those guys who get them online now are going to be in a really, really interesting spot from a cash flow standpoint. Um, and then also from a model capability standpoint, right? Especially in the context of what's just happened with DeepSeek, um massive performance increases, massive efficiency gains on old hardware, right?

SPEAKER_02If you take those efficiency gains, if you take how they made those performance increases, and then you apply it to the latest generation of hardware, and the TB200 is what, 32 times more performant than the A100 or something like that, you combine all of those things together two years down the line when those contracts are coming to an end and the models that have been trained on that hardware science to be put out into the wild, what are they gonna look like? How performant are they going to be? Um, with all of these things coming together at the same time, it's gonna be absurd.

SPEAKER_02Um, so very interested to see what the GB200 system is gonna do to the market, definitely from a neo-cloud standpoint. Um, but be interested to see what your take is from a data center standpoint, right? Because an 135 kilowatt liquid cooled rag is not an easy thing to host. Um, what's this done on your side?

SPEAKER_01I think it's it's interesting from a data center side because you've got developers trying to build a data center for the future technology that they don't know what the future technology is. So I think I was I went to Kickstarter and I was listening to some of the panels, and it's around trying to design that hybrid facility where you can adapt to the technology. Um, so you kind of you obviously liquid cooling is a is a the way that a lot of these are going, but sometimes you also need the ability to for air cooling as well, like short-term, if not long term.

SPEAKER_01So I think their biggest challenge at the moment is trying to create that facility. Like, what does it look like? What does the data center of the future look like? Nobody really knows, but that's the challenges they are having. And obviously, when like we were talking about before we come online, when you increase the rack capacity and you've got these mammoth facilities, are we looking at the demand is just going to be so big that the facilities will continue to scale? Or will there become a time when the facility is kind of obsolete because the capacity of the rack is is so high?

SPEAKER_01I I don't know, it's like paradox law, isn't it? They say like the easier it gets, the more people use it. But is that the case or not? I yeah, I wish I knew I'd um I'd hedge my my investment bets if I if I knew the answer.

SPEAKER_02If only we knew what was going to happen in the future, and the portfolio would be doing slightly better, right?

SPEAKER_01Yeah, it's exactly some do well, some don't.

SPEAKER_02Yeah, right. It is an interesting one because like I think to your point with like how do you design a facility for the future, um the the traditional build-out methodology whereby I mean if you look at what the average rack density is in Europe, right? It's eight to twelve kilowatts air-cooled, right? Um, and there are still new facilities that are being brought online that are like 45 to 50 kilowatts air-cooled racks, and like those that's that's a massive amount of power.

SPEAKER_02But it's not enough, right? It's not enough rack density, and I mean semi-analysis yesterday put out a piece on um cooling the data center of the future, and there's one little tidbit of information there, um, which was the proposed rack density of the the Rubin Ultra systems from Nvidia, which we're anticipating are probably going to be released at um at GTC this year.

SPEAKER_02And for context, right, you've got the MVL72 system, which is a 135 kilowatt rack, and that's that's absurd, and people are already struggling to get that installed. With the burn-in phase, there have been reports of it jumping up to 200 kilowatts, right? Which is even more absurd. But the Ruben Ultra in this semi-analysis report is an 819-kilowatt rack. That's beyond ridiculous, right? Because you could have two racks, you've got a 1.6 megawatt data center.

SPEAKER_02And I mean providers are gonna have to pivot towards it as well, because we were saying before we can hit record as well, like the the the Revenue potential from a single rack at that density is absolutely ridiculous. Because, like, as I say, hypothetically, you're in a position where you somehow manage to secure some nice Scandinavian power pricing, achieving you're paying seven cents per kilowatt hour, you've got an 819 kilowatt rack, you're paying$100 per kilowatt per month, and you're paying that seven cents per kilowatt hour, your monthly cost at 100% utilization for a single rack is going to be$123,000 seven hundred and fifty per rack, right?

SPEAKER_02I mean, if you're a data center operator, you're probably gonna want a piece of that action for one. But then if you're deploying those GPUs and you're a GPU provider, you're gonna want to make sure that whatever facility you're deploying that into is able to support it, right? You're gonna want you're gonna want something good for that amount of money. Um we only know the power at this stage, right? We don't know anything else. Um what is a facility going to look like that's able to support that? How are facilities going to be able to be updated to be able to support that, right?

SPEAKER_02If you're bringing something new online and you're you've got what I don't know, eight megawatts or something like that. Ten of these.

unknownYeah.

SPEAKER_02Ten racks in your facilities done, right? What do you do with the rest of the floor space?

SPEAKER_01No, it's like we were saying, yeah, yeah, five or side pitches in all the all the data centers, indoor golf. Um there's there's definitely a market there for for entertainment within these facilities. Or maybe they could have education areas to train some talent for the future, but that's another that's another point. Right. My next one was something you touched on earlier. UK news, um 14 billion. The UK government announced$14 billion of investment in the UK data center sector. Seems like a lifetime ago, but it was the start of January. It was about four or five weeks ago.

SPEAKER_01Um, that included$12 billion from Vantage, uh,$2.5 billion from NScale, as you touched on. Also in the UK, we saw DC01 UK approved, um, which was a$3.5 billion 400 MBA um site in Buckinghamshire. So there's been a lot of UK announcements this year to date already. Again, ties in with our comments earlier around the US announcing Stargate, the Paris summit, everyone trying to get a piece of this AI action.

SPEAKER_01Um, but it does seem that the UK, from a government perspective, has changed their viewpoint on data centres, you know, national critical infrastructure, and basically the importance of the facilities in order to enhance and attract the correct organizations and the economic value for the future.

SPEAKER_02Yeah. Yeah, and and it's great to see, to be honest, because I think it's very easy to end up feeling pretty pessimistic when you're on this little island in the North Atlantic, especially in like December and January when you haven't seen the sun for a couple of months. But it was it's an interesting announcement, right? It's good to see that those in government are starting to take things that little bit more seriously in terms of what's publicly facing, right? I'm not saying that there aren't more conversations happening internally, and I'm sure that there are, but putting something like that out does show that we're taking things that little bit more seriously here.

SPEAKER_02But then on the other side of that, 14 billion. It's not enough, is it?

SPEAKER_01Like everyone always says to me, it's um the numbers are announced, but does the money ever come? And I think you have to ask that question sometimes. You know, these big numbers, these big scale facilities are announced. Um, not all the big scale facilities happen, but you rarely see a announcement saying, you know, that one we announced last year that it's 400 mega, we're not actually going to do it anymore. It just kind of brushed under the carpet. Um and it's the same with the capital, you know, the the investment, hopefully it comes for the, you know, and I'm not talking about from my own perspective, I'm talking about the the value it adds from uh, you know, social value really and the importance of technology to the to the growth of individuals and and and the collective is is massive.

SPEAKER_01So hopefully it does come, but we do have to say let's see, let's see what washes through, especially in the UK, where not not a lot washes through.

SPEAKER_02No, and I mean, given the number of times we've had a change in Prime Minister over the course of the past few years or so, right? It doesn't exactly inspire confidence. Um, I mean, sure, it's a different party now, right? But ultimately the system is still the same, and this might be a feature as opposed to a bug, right? Um it's as I said though, like they've got some serious people in a senior advisory capacity who know what's going on, right? Um got Maclifford of entrepreneur first, um and then CEO of DeepMind as well, serving in advisory capacities to the Prime Minister, and like these guys know what's going on, right?

SPEAKER_02So that's that's a big vote of confidence in terms of how how the plan has been put together and how much thought has gone into it. But then like the the hundred thousand GPUs of public compute capacity by 2035. 100,000 GPUs is nothing in the grand scheme of things now. And that's I mean, it's still what it's probably close to 5 billion in CapEx at some current rates. Um but it's not enough, right?

SPEAKER_02And this is by 2035. Um Elon did that in what six months total from beginning to end. Politics of it aside, I mean that's obviously one end of the scale, um, and not everyone is going to do that, but that kind of deployment capacity is it's a step in the right direction, though, right?

unknownExactly.

SPEAKER_02Is this a start of momentum? Hopefully.

SPEAKER_01Yeah, we're making the right noises, which is again a change. A change is more than noise. A change, in fact. Right. What else have you got? We've got I I knew this would take we're we're uh we're trying to do this quickly, listeners, but um, there's so much news, it's quite it's quite difficult to do it. So bear with us. We'll we'll keep going.

SPEAKER_02Yeah, no, I mean I guess those are the main ones for me. I think the my the thing that I'm most excited about, I think, over the course of probably the next year or so is what is how is the market going to change off the back of the DeepSeek news um in terms of the efficiency gains, and then also with the new hardware generations um starting to come online. Um those two things together, coupled with the fact that we're now starting to see a lot more serious, you know, and I mean that's not to say that this the investment wasn't serious beforehand, right?

SPEAKER_02But like serious investment happening now in various places around the world. Um, and with an administration, the US, that I mean, at least currently seems to be a bit more pro-tech, and things seem to be happening relatively quickly. Um, politics aside, stuff does seem to be moving a bit faster off the back of the election, um, and then the new administration coming in. So, do we see some acceleration coupled with things like I mean, obviously you've got like the the AI Diffusion Act that came in actually before um the Biden administration went out, like with the new hardware generation and then export restrictions coming in, and then also the tariffs, but then also the more investment coupled with the deep seat news, all of that coming together.

SPEAKER_02I'm just really excited to see how the market is going to change, to be quite honest, right? Because last year it was very much a US poker's party, all we've were making the most noise, Lambda were making the most noise. It was US neo clouds and US GPU providers that were making the most noise. Now we've got French GPU providers, we've got British GPU providers. Um things are getting a little bit more exciting, I think. So who knows what's going to happen?

SPEAKER_01Yeah, who knows? The only other things I had down were Spain, Merlin property Merlin Properties and Edge to announce two gigawatt scale campuses. We've seen a lot of action in Spain already. Again, tight into yeah, we could work our way around the world, to be honest. A lot of investment coming, yeah, exactly. A lot of investment coming in Spain. And my other final point was simply hyperscale CapEx. So Google, they've all increased. I'm not going to read through the percentages, but they're all significant growth on 2024.

SPEAKER_01Google$75 billion, Microsoft$80 billion, AWS$104 billion, Meta$60 to$65 billion. And I think on your point around sort of Deep Seek and you know that what that did to the market, it oh, we're going to see a slowdown. That's all been announced post-deep seek. And I think Deep Seek's had a had an impact, don't get me wrong, and it it changed the way people are thinking. But I don't think we're going to see a slowdown in in CapEx for a considerable time.

SPEAKER_01You know, if you read a lot of the earnings reports from those earnings calls from those um recent announcements, you know, they're they're significantly talking about building for the future. They're scaling for the future because they can see the demand that's on the horizon. They're not building for today from an AI perspective. Cloud obviously continues, but from an AI point of view, they're building to ensure they're in a position to make the most of the once-in-a-lifetime opportunity, as they all say.

SPEAKER_02Yeah, no, absolutely. And I think like from a model capability standpoint, if you look at how the performance gains have been made over the course of the past few months. One, there's the efficiency side from Deep Seek and then fully open sourcing everything. So the the closed model providers can now take those in. The other half of it is it just seems to be a case of if we throw more compute at the problem for longer and we make it think harder using more GPUs, then it gets better, right? The facilities are the future, and then with like an 819 kilowatt rack, right?

SPEAKER_02Like what is the computational capacity going to be if it's using that much power? Um it's I mean, as you say, like they're building for the future, right? And I think if if that level of CapEx has been approved, right? Assuming that there's not an unmanageable amount of FOMO going on, um there's clearly something on the horizon that we're not necessarily privy to right now.

SPEAKER_02And I think whatever that thing is, it might end up being a little bit further away than people end up thinking. Perhaps there's a bit of brinkmanship happening at the moment, right? Because if you're a hyperscaler and one of your competitors starts deploying huge amounts of data center capacity all over the world, you can't not do it because your moat is making sure that you've got the PPAs, you've got the land, you've got the facilities locked down as well. Maybe those guys make a little bit less money for a few years. But then whatever is coming, I think the wave is going to be much, much bigger than people think.

SPEAKER_02But because it's so much bigger than people think, it's probably a little bit further over the horizon than people think, as well, right?

SPEAKER_01Yeah, definitely. I think I'm one point I was referred to is I was Elon Musk talking about the compute required for humanoid robots was I think 10 times as much as autonomous autonomous driving. So and he's he's um you know, I think in the same kind of statement he was saying like everyone will have a humanoid robot one day. And again, whether that's true or not, I don't know. Maybe it is, maybe it isn't, but gives you an idea of what they're thinking from a compute perspective, it's it's massive. The demands of everything we do now, just more compute.

SPEAKER_01So fascinating. Right, we'll we'll end it there. I think I loved it, and we'll do it. We're gonna do this every month. Um we we would love to have guests on as well, you know. So if any of the big announcements we're talking about, we are gonna reach out to companies. But if you've got a huge announcement and you want to come on and chat about it with myself and Ben, you know, reach out to us. I say anyone's got any ideas on names, data center news and GPUs, rhymes, and uh it was very, very easy. So uh I've gone with that for today's episode. But who in the future, who knows?

SPEAKER_01So, yeah, enjoyed it. Ben, thanks. Thanks for the input.

SPEAKER_02Likewise, Andy, thank you very much. Um, yeah, it's it's always nice to go down a rabbit hole, right? Where you've got areas of separate, but then also areas of confluence, and then understanding where those areas are, um, especially in this niche within a niche within the niche, right? Like there's so much going on, um, and having an opportunity to talk about it, and then hopefully also provide some value to the listeners as well, right? Um, of two dudes within the niche within a niche within a niche, right? Um, just trying to keep up.

SPEAKER_02It's part of the fun. So looking forward to doing the next one of these and then seeing how many more hundreds of billions of doppers have been in the mounts in the next month, right?

SPEAKER_01Exactly. And we've got a time, we know now, February the 14th record. So the next time we record, we we we'd be able to say, well, the last four weeks or whatever it is, this has happened, and it does give you it does give you an idea of the of how much is that again, like you write a weekly newsletter of news, and you don't really realise how much is on it until you're writing it and you're like, oh, this is only a week. Yeah, and I've got and I've got to try and collate all this information. So much happening. That's why we love it, that's why it's such a crazy industry, and yeah, um, yeah, it's fascinating to be part of. But yeah, but all good.

SPEAKER_01Thanks for your time. So hope you enjoyed it, listeners, and we'll um we'll do it again in uh in March.

SPEAKER_02Absolutely. Thanks so much, Andy, and thanks everyone for listening.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Powering the UK’s AI Future: Inference, Energy and Local Growth | Live at The Xcelerated Compute Show 2026

    Nov 13, 202627 minAudio

    Send us Fan Mail In the third and final episode of our special series recorded live at The Xcelerated Compute Show 2026 in London, Andy Davis and co-host Rory Flashman from Spa Communications explore what the UK needs...

    Credit the people on this episode

  2. The Truth About Data Centre Talent

    Nov 6, 202638 minAudio

    Send us Fan Mail In this episode of the Inside Data Centre Podcast, Andy Davis is joined by Oli Monks, Head of Talent at AVK, and Sophie Parker, Head of People Development & Talent Acquisition at Pure Data Centres...

    Credit the people on this episode

  3. The Value Gap | Critical Careers

    Oct 30, 202639 minAudio

    Send us Fan Mail In this episode of the Critical Careers podcast, Lizzy McDowell is joined by Andy Davis from DataX Connect and Sarah Davenport from Capstone Recruitment for a conversation about the value gap, and why...

    Credit the people on this episode

  4. Can the UK Win the AI Infrastructure Race? | Live at The Xcelerated Compute Show 2026

    Oct 19, 202622 minAudio

    Send us Fan Mail In the second episode of our special series recorded live at The Xcelerated Compute Show 2026 in London, Andy Davis is once again joined by co-host Rory Flashman from Spa Communications to explore what...

    Credit the people on this episode

  5. Building AI Infrastructure at Speed: Europe’s Advantage | Live at The Xcelerated Compute Show 2026

    Oct 9, 202617 minAudio

    Send us Fan Mail In the first episode of a new special series recorded live at The Xcelerated Compute Show 2026 at the Business Design Centre in London, Andy Davis is joined once again by co-host Rory Flashman from Spa...

    Credit the people on this episode