Name: Karl Havard, Era4
Uploaded: 2026-09-29T16:00:00.000Z
Duration: 42 min 28 s

Clips
More from Sovereign AI: From Strategy to Reality

Karl Havard

Chief Commercial Officer

Era4

play_circle_outline Era4: Building UK Sovereign AI Beyond Borders with Land, Behind-the-Meter Energy, and Data Centers

play_circle_outline National AI Control Plane Sovereignty: UK/EU Consortia, Orchestration, and Reducing Foreign Silicon Dependency

play_circle_outline Data residency trade-offs: replication, telemetry, backups, air-gapping choices

play_circle_outline Sovereignty is not purity: compensating controls and customer risk trade-offs

play_circle_outline Legal and regulatory risks: CLOUD Act, export controls, US legislation impacts

play_circle_outline Auditing and standards gap: lack of sovereign audit frameworks

play_circle_outline UK AI: From Training to Inference — Rising Tokenization Costs and Retaining Revenue in the UK

play_circle_outline Newport Steelworks Redevelopment: Vocational College, Heat Reuse, Liquid Cooling and Low-PUE Closed-Loop Systems

29:29

31:57

Info
Transcript

Karl Havard, Era4

Karl Havard

Chief Commercial Officer Era4

Karl Havard, chief commercial officer of Era4, joins Dave Vellante and Amit Eyal Govrin of theCUBE Research for a discussion on sovereign artificial intelligence, UK-controlled AI infrastructure and the economic benefits of keeping AI investments within national borders. As part of HPE's "Sovereign AI: From Strategy to Reality" program, Havard explains how Era4 is building a network of energy-powered AI data centers across the UK, combining local infrastructure ownership, operational control and community investment to support the country's AI ambitions.
<... Read more

explore Keep Exploring

What is Era4, what is its mission, and what is the role of its Chief Commercial Officer? add

How can the UK and EU achieve sovereign AI infrastructure and offer AI services in a sovereign manner given their reliance on overseas silicon (GPUs/TPUs), geopolitical risks, and other control-plane vulnerabilities? add

When you say "leaves our territory," do you mean that the data leaves our territory (as opposed to the servers/compute or other assets leaving)? add

How should organizations and nations think about AI/data sovereignty when using third‑party or foreign models — what does sovereignty mean in practice (data residency, who has access and control, personnel clearances, operational measures), and how pure must that sovereignty be? add

Which, if any, of the US-based vendors in your stack (for example NVIDIA, HPE, VAST, WWT) could be compelled by a US production order or the CLOUD Act to disable or "switch off" your service? add

Has an independent third party audited your sovereignty claim (for example over a UK AI tech stack), and would you be comfortable providing evidence of that to a regulator? add

What are the financial and sovereignty implications of relying on externally hosted, token‑based AI models—who benefits from token spending, and can a nation maintain operational AI infrastructure (e.g., if physical links are severed)? add

How will your planned data-center deployment at the old Newport steelworks be carried out, and what community, environmental (power and water), and heat-reuse benefits will it bring? add

bolt Powered by CUBE AI

Karl Havard, Era4

(INTRO)

Dave Vellante

>> Hi, I'm Dave Vellante. Sovereign AI is one of the hottest topics on the docket these days. But to many, sovereign means just making sure your data is resident in a local jurisdiction. But there's much more to sovereignty beyond territorial. There are legal issues, operational controls. There's, of course, a technology stack dimension. And the most important is financial sovereignty. My co-host and theCUBE Research analyst Amit Eyal Govrin calls these the 5 pillars of AI sovereignty. Welcome to Sovereign AI: From Strategy to Reality, made possible by Hewlett Packard Enterprise. Joining Amit and me today is Karl Havard, Chief Commercial Officer of Era4. This is a company building a localized, sustainable network of data centers to power the UK's independent AI infrastructure. Karl, welcome. Thanks for taking some time out today.

Karl Havard

>> Thanks for having me. Delighted to be here.

Dave Vellante

>> You bet. Tell us a little bit more about Era4 and your mission and your role.

Karl Havard

>> Yeah, so I'm the Chief Commercial Officer at Era4, which means responsible for virtually everything that's got a number associated to it, coming in and also being spent by the company. So what Era4 is, it's a UK sovereign AI infrastructure provider. And what we do, we have two parts of the business. One is where we generate our own energy from behind-the-meter power in the UK, and we then deploy AI infrastructure on top of that energy. But what we also do is we regenerate old industrial sites across the UK, such as manufacturing plants, old steelworks, and convert those into areas where we can deploy AI infrastructure, but more importantly, we do it in harmony with the local community. So AI infrastructure gets deployed, but what it does is it reinvigorates that community, creates a new destination, and in partnership with that region, we will also stand up a tech college, maybe community centers, even conservation areas that we're doing in the south of Wales. So our aim is to be in full control of the energy all the way up the stack and have that ability to truly deliver UK sovereign AI solutions.

Dave Vellante

>> Thank you. So you heard my introduction there. I think that many in, certainly in the US, are not as tuned in to the sovereign mandate as they are certainly in the EU and generally in the UK specifically. What would you say are the drivers for sovereign in the UK?

Karl Havard

>> So it's really interesting wherever you go in the globe. I've just come back from the US actually, having spent a week in both mid and West Coast. Sovereignty really isn't a thing because if you go back 20 years plus when cloud computing emerged, it all came from the US. So, your usual suspects— Microsoft, AWS, Google, Oracle— they have come from the US and offered their cloud services across the globe. So what's happened, with the emergence of AI and the need to sort of, rebuild different types of data centers to house AI, it's been a moment where UK and EU, regions, nationalities have sort of looked at themselves and thought, well, if we're going to create AI applications for the use of our citizens, for the use of our European, UK companies, do we really want to have our data, of course, in petabytes and more in order to be able to train AI models and tune them? Do we want that data to reside offshore? And the answer to that question is typically a no. And the reason that is, is because there has been a behavior of the last 20 years where applications have run on US hyperscale clouds and credit to them because there's no alternative over here. But now with the emergence of AI, there is an opportunity for the EU and the UK to sort of stand on their own two feet and stand something up.

Dave Vellante

>> And I'm going to ask Amit to jump in here, but I just want to clarify. So your role, your firm's role is the land, power, and shell as well as the tech stack, or do you stop at the land, power, and shell?

Karl Havard

>> No, we go further up. So it's land, power, shell, and the CapEx of the AI infrastructure, which we then own as an asset inside the UK. And then we offer that as a UK service. So there's a UK control plane on that. And then above that, for the orchestration and the application stack, we're partnering with a, well, creating a consortium of UK and EU organizations to sit on top of our sovereign engine.

Dave Vellante

>> Great, thank you for that. Amit, I know you wanna jump in here, please chime in.

Amit Eyal Govrin

>> Sure, and Karl, before I even ask any questions, I'll commend you first and foremost about fighting the AI doomerism. We've experienced this quite often in the US, and really what Meta has done has really helped spearhead that, showing how communities can grow and prosper from having and working hand in hand. And I think what you're doing there in the UK is just as important. So by all means, I'll give you that credit. And I'll also challenge you a bit just to kind of tease out a few of these T's and C's. I'm sure you're familiar with our 5 pillars of sovereignty, the 5 pillars that we define. It goes beyond territorial. So maybe I'll start there. How do you define sovereignty in your own mind? And if we had to, uh, put a gun to your head type of analogy, where do you feel you're least sovereign within your whole stack? Maybe I'll go there.

Karl Havard

>> I can start there first, where we're least sovereign. And, of course, the UK and the EU doesn't really build its own silicon, AI silicon. So, GPUs, TPUs, certain types of accelerators, etc. We are hugely reliant upon overseas infrastructure as an asset. So sovereignty in that basis, it's a case of I ride a motorcycle, it's an American motorcycle and I bought it, it's mine, sits in my garage, and then I ride on the road. It's in my control. So we believe that once you've acquired the assets and you've installed them into your UK facilities, whether it's UK energy, UK land, UK control, then there is the ability to offer that AI service in a sovereign manner. I saw the 5 pillars that you discussed. Actually, the way Era4 and myself define it is we have 2 key areas. The first one is the control plane, and that's at every layer of the stack. So geopolitical situations, internet cables being cut under the oceans, potentially Middle Eastern, unfortunately, conflict has shown that data centers are a target. Also export controls, tariffs, Patriot Acts, all those pose a risk to, inside the US, all those pose a risk to the UK and the EU sovereign stack. So what we're trying to do is make sure that every level of that stack where the control plane resides, then we can have the ability to make sure it is on or off based upon the decisions that we take as a nation. So I think we're getting there, certainly at the lower ends of the stack. So we control the energy, we control the land, we control the data center. Once we've bought the asset of their infrastructure, we control access to that through the control plane. And then in partnership with the UK Sovereign Consortium, people like Cosine and the new model that they're developing, the EU organizations, Mistral's, a great example. We can actually stand those guys up on top of the stack and there is a non-US control plane, let's put it that way, outside of that. The area that I would define is, and I can see it on your stack behind you, and that is the economy, the financial elements of sovereignty, the way we are building and the way we're funding our business is every, I'll say, pound but dollar spent on our system or on our AI solution infrastructure is revenue that stays in the UK. So it benefits directly the GDP and the UK economy, not just through tax. Actually, the revenue stays, it supports the community because it's invested into that community. Supports the employees of the ecosystem of those, whether they're developing models or sending them out to work for various PLCs across the UK. That money stays here. What's interesting about other inverted commas, sovereign solutions, we have the AI Growth Zones in the UK, is that actually is US-based. So if I take in the South Wales area, just over the estuary from where we're building, actually, we have an AI growth zone nominated by the UK government. But within that, 5 data centers owned by Vantage with Microsoft using those facilities. Now, if I'm a customer and I'm going to use the Microsoft platform in that area, if I'm UK, non-UK, whatever it might be, that revenue touches the shores of the UK and then goes offshore back to the Microsoft and Vantage P&L. So we want to make sure that everything we do benefits the UK so then we can start to reinvest this money and start to expand accordingly in the way that I think a sovereign nation should do.

Dave Vellante

>> So Amit, if I could interject, I was meeting with a company yesterday who, while there was less of a concern about US silicon. There was a concern about getting locked into silicon. And what they've done is they've built their own abstraction layer to be able to, hide the underlying primitives of whether it's, NVIDIA CUDA or AMD ROCm or Arm, etc., x86, and to be able to have the ability to switch, swap in and out of silicon. So even in the US, they're aware of that. It's interesting, Karl, to hear you talk about it 'cause you're not mandating, for instance, Arm, UK-based Arm, like we've seen some efforts in France to sort of mandate Mistral. But you are putting in, it sounds like, compensating controls, perhaps like this company, or maybe your customers are doing that to be able to sort of minimize the switching costs, which, Amit, is something you and I talk about all the time.

Amit Eyal Govrin

>> 100%. Yeah, actually, let me have a follow-on. You answered very, very honestly, Karl, so I want to give you exactly credit for that. I think anybody, even any Fortune 50 company in the US, if they had to go in to introspect, very few could say we have the fab, we could go ahead and replace AMD or NVIDIA or TPUs tomorrow. So Nobody should shake a stick at the fact that you are still reliant to some degree on a supply chain that tracks back to the US and even the rare metals beyond that to China. Right. There's always going to be some dependency here. My question is actually a little different. If— let's tease out the 5 pillars throughout this questioning and hopefully you could— we can go down this list fairly easily. If we have to go to that territorial claim, the real question becomes what leaves the UK today if your customer is on these data centers? Is it telemetry, support tickets, backup models, downloads, license checks? I'm just curious if the territorial piece of it will— what, if any, of this leaves your territory today?

Karl Havard

>> When you say leaves our territory, do you mean the data leaves our territory, or—

Amit Eyal Govrin

>> That's really the question, is where that goes.

Karl Havard

>> Yeah, yeah. So interesting. So I used to work for a couple of hyperscalers in the past, and they had solutions which are still available today, Outposts and local, where they would put their servers into the territory that they would deploy within and claim that because the data resided there, they would claim that would be data sovereignty. Well, data is not a physical asset, is it? It's not like sticking your gold in a single place, putting it in a safe and keeping it protected. Actually, data is fluid, it's replicated. It can, inverted commas, leak anywhere if you haven't got the right air gap controls. So it depends on the sensitivity of the data, but the way that we are building our mesh, as we call it, so where we've got our behind-the-meter energy being generated ourselves, actually from landfill, which is an interesting concept, which over the years has been sold back to UK Grid. We've stopped that now and we're using it to power AI infrastructure. We've got like 5 to 10 megawatts dotted around the UK. We're creating a sovereign mesh over the top of that in order to make sure that it's in the control of the nation where it resides. For higher sensitivity stuff, well, then we will air gap it naturally. So there'll be no back doors where things can move around. But if we're supporting, say, the universities across the UK as an example, then various universities may want to collaborate with each other. They may want to move models around. They may want to use Token Factory, etc., applications around different areas. We will give them the security to be able to do that within our network. Where I think the challenge does lie is when the hyperscalers do say, yeah, we've got these outposts, we've got this local, but actually the backup is offshore, or there's some need for some part of the application or some subset of the data to make the whole training of the model work, which is drawn from other areas. And that's where it becomes fragmented. And I think that's where they see the sovereignty and security issues may creep in.

Dave Vellante

>> Well, and there's trade-offs, right? 'Cause there might be a phone home there, there might be a patching. And those are trade-offs that many customers are obviously willing to make because you don't want to be air-gapped and not be able to patch. Or maybe you do. That's a choice. Really sovereignty is about customer control, isn't

Karl Havard

>> it? Completely. And I think there is a negative, element that sort of bleeds into the definition of sovereignty, and that's we've all got to stand isolated from each other. And actually, I'd like to think we are allies

Dave Vellante

>> in

Karl Havard

>> most things. But we have a strong partnership with the EU, the UK, and, and the US, whatever the media may try and imply in certain areas. But there is that element of collaboration there, and that supports global companies as well. So, organizations that have a presence across that area of the globe, would want to be able to distribute their data accordingly when it comes to more sensitive data. So things that need to sit on what I call the high side. So at secret or above secret, that's when you need the true sort of air-gapped capabilities. There can be no leakage. And whether that's for defense primes, whether it's for certain research or health companies, whether it's for people in the Ministry of Defence, then there's a different set of criteria that are needed in order to adhere to that.

Amit Eyal Govrin

>> Sure. Actually, you said something that I want to double down on. I have a saying that goes, sovereignty is not a purity test, right? It's control plus a sidecar, a risk sidecar priced into it. So you get to decide whether you're comfortable as an organization having a Frontier model in your request path. You don't have to. You could always use open-source models, but it's really up to you and really on what workload specific you're referring to. So the way I view it, there's no purity test here either. There's data that can leave the UK. There's data that can stay dormant within the UK. But if you're going to talk about the operational layer, maybe I'll take one step backwards and say, who gets paged at 3 AM? I assume these are people within the UK with UK passports, right? Who holds the keys, the log sources, maybe get into that a little bit.

Karl Havard

>> Yeah, exactly. So when it comes to, I guess, defense and security, national security associated with sovereignty, then it's not just the technologies, it's the actual ecosystem that sits around the stack, and that includes the people that engage. So as well as being UK citizens, you also have necessary clearance requirements, whether it's SC or DV, as we call it over here. And therefore, those people that are touching the various elements of data or introspecting it, have to do so in a controlled environment, maybe under CCTV, and they've had their background checks to some, the detail around them, and therefore they are able to do that. But that's probably an extreme and very specific example. If we move further down where, the data may not be that sensitive, obviously it still needs to be kept under control, but where you have normal working people operating that's when the true collaboration is required from a sovereign point of view. And that's a partnership agreement. I actually think there's probably only one, maybe two nations in the globe that can claim they are truly sovereign. One is the US because they own everything in the stack. But then even then you're still reliant upon, developing the silicon potentially overseas. You've got ASML, a Dutch company that in every single fab, they control it. And then the other one you could probably get to the similar sort of level would be China. Everybody else is probably nowhere near that capability.

Amit Eyal Govrin

>> Butyeah, I agree. supply chain goes way beyond just the silicon. So you brought this up and that's usually music to my ears because my next question naturally is you name the who's who in your stack, right? NVIDIA, HPE, I believe VAST, WWT. these are all US companies. So if a US production order comes in and invokes the US CLOUD Act, which one, if any of these, can switch you off?

Karl Havard

>> Interesting. So we are— our Chesterfield site is now live. We have— you're right, we've got B300s liquid-cooled deployed on site there. They were provided to us by HPE and WWT did help with the installation work. You've also got the sub-supply chain below that with Schneider and Vertiv in the mix as well. But we now own it. And that's the thing, because we own it, then I don't think— and I'm no legal expert, but I don't believe because we manage the control plane, we have the keys for the energy, the data centers, obviously MDCs, the AI infrastructure, we manage the portal and access to that. That's all in the control of Era4. I don't think we can now have an overseas legislation applied to us. If it was a US service, then absolutely. So if we decided to say we're going to build in Chesterfield sites, stand up the NDCs, but we'll allow AWS Outposts in there, then we would be susceptible to the Cloud First Act or any other export control potentially. But we haven't done that. We've bought the asset, we own it, and we're now deploying that. As a service. Interesting, your thing about VAST. At the moment, we're not deploying VAST on that site. It's self-contained. So there is probably no overlaying service from a US nature on top. But I hope that answers your question. Yeah,

Amit Eyal Govrin

>> it does. And again, there's no purity test whatsoever here. When I do my homework, I go a little deeper and I'm trying to see what's in the stack. And that actually brings me up to my next question, because you mentioned HPE Private Cloud AI sits underneath, right? You have the customer audit. Could you— you mentioned you own it, but there's areas that you do license, right? So would that still be the case, or did you correctly mention that you own the whole stack? Maybe that's the part that I missed in

Karl Havard

>> the—Yeah, no, that's a good point. When you get to the orchestration layer, then yes, we do have service agreement there. We license the software on top of that. So there is an argument to be said that if that software, interesting, it's not US software, but if an overseas territory was to say, we want to introspect that or whatever, then we'd probably have to acknowledge that. But we'd probably challenge it as well in order to protect the customer also. But I get your point, I get where you're heading to. There's no pure control from any single sovereign nation, I don't think, through the whole stack. And actually, I don't think our aim should be to try and reach that utopian point because then you just create an island of yourself. Right. And you don't get the best of both. So yeah, sorry.

Dave Vellante

>> And I think, no, no, I apologize for interrupting, but that really is the point here. Is the customer with obviously Karl, your help, has to evaluate the risk and the probability of that. In this case, yes, licensing the control plane, what's the risk of that versus maybe the CLOUD Act is more concerning or there's maybe other lock-in issues that lead to the financial pillar. But this is the point that we're trying to educate people regarding Sovereign, sovereignty, that it's not a purity test. It's about understanding where those risks are, putting in compensating controls, accepting that risk, as opposed to putting your head in the sand and then getting an invoice someday or a surprise someday. That's really the point.

Karl Havard

>> Yeah, totally. we've got a legal team which has constantly got the radar on looking at changes that are happening in US legislation, EU legislation. We're trying to help and advise the UK on the legislation as well. And yeah, it's an ever-changing feast. But I think you're right. Yeah, we have to be able to be— maybe the way to look at it is putting the customer's interests at heart to make sure that our customer that we onboard is protected, but also where we locate and the community around it, etc. They're protected as well in a similar way. But yeah, it's— yeah, anybody that claims to be 100% pure sovereign is not telling the truth or they're misinformed.

Dave Vellante

>> Or a vendor that claims that they can confer sovereignty to a customer, that's just sovereign washing. And so we kind of like to call that out.

Karl Havard

>> Yeah,

Dave Vellante

>> yeah,

Karl Havard

>> yeah. We should call it out. Yeah, absolutely. Yeah. Good.

Amit Eyal Govrin

>> Interesting you mentioned that because that goes into evidence and proof. So if, say, a third party, I'll, I'll reframe my question. Has an independent party ever audited your sovereignty claim? If you ever had to hand over evidence to a regulator, for say, on that, would that be something that you'd feel comfortable passing that? I'm just curious.

Karl Havard

>> Yeah, well, it's interesting we say our sovereignty claim. I think we haven't claimed that 100% sovereignty because we know it's not realistic. What we are claiming is there is a way to be more in control of a UK tech stack in the AI area. And there's also a way to make sure that the UK economy benefits from that. We would welcome an audit. We've not been audited. From a sovereign perspective. Clearly we've been audited in other ways, from ISO 27001 and all those good things. And from a security point of view, all that's been done. But when it comes to sovereign, I don't know if there is a body that actually would actually do so. And I think when it comes to auditing, though, they have to audit against some form of legislation, don't they? And I think the sovereignty, I guess, word is defined in many different ways in different areas. And therefore there's probably no standard to audit against, which will probably remain the case for a good few years, actually.

Amit Eyal Govrin

>> Actually, that's an interesting artifact you just mentioned because I agree with that statement. The truth is sovereignty is not a regulation. Right, it's— let's think GDPR, it's a guideline, right? GDPR is a guideline, it's not a regulation. Now you're either going to conform to GDPR or you're not, and everybody who asks you if you're GDPR, you better be telling the truth, because otherwise if you don't, they'll boycott doing business with you, right? You could easily find yourself out of a certain market if you're not GDPR, but it's not imposed by any type of, lawmaker. It's just something that you would want to do because that's the financial trade-off of being compliant,

Karl Havard

>> right?Yeah, exactly. Yeah. Yeah. Yeah. GDPR, it's clearly, absolutely needed. It was introduced, I guess, about over 10 years, 15 years ago now. And yeah, it's something that you do want to make sure you're acutely aware of and you adhere to. Otherwise, I think it's what is up to 15% of revenues you can be stung by, maybe more now. Yeah, but you won't go to prison for it, but you'll just end up going bust. So, yeah, but I think that came out of the EU, didn't it? And I think initially US organizations tended to turn a blind eye to it until the time came when the EU sort of bared its teeth and started to apply some fines, etc. But yeah, Sovereignty probably won't get to that point. I think there should be guidelines and frameworks. The way we're approaching it is to create consortiums, so organizations that have agreements to work together with the best interests at heart in order to maintain a sovereign standard wherever possible. But other organizations will have a different way of approaching it, as we, as we know.

Dave Vellante

>> But the EU is really leading in the thought leadership in this topic. And then, as you said, Karl, you were just recently in the US. I'm talking to US companies all the time, and it's just— there's the depth of thought that you've even conveyed here today is not evident in a lot of the conversations that we have in the US. And that's part of the reason why we're so happy about this program, because we're educating people on these very, very important issues.

Karl Havard

>> Yeah, it's interesting. There's also another flavor, and I'm not sure, Amit, you're aware of this, of sovereignty, which is, and this will sound awful, but it's non-US, as in we can stand up something which doesn't rely on the US. And I don't think that's the right approach at all, but that is one other flavor of sovereignty we're starting to see emerge. I always look at it from the other way around, and I think, What if the US were to really be focused on sovereignty and everything they did or the US did or US company did was to benefit the US economy? I think we wouldn't be in the position we'd be. We'd be far behind now where we are already. The only reason the UK and Europe have been able to stand up our infrastructure is because the US has developed this technology and made it available to us. So we need to be cognizant of that and sort of respect it as well at the same time. So

Amit Eyal Govrin

>> yeah. I think the framing should be sovereignty from who and for who, right? So sovereignty by definition, if it's a sovereign state, any state will have its claim for sovereignty from another state if they were to focus on that as a tenet. And the challenge is with the US, it's almost like us against the world from a battle-tested ground. We're sitting. So we don't need to be sovereign. But you are sovereign because everything you do falls under the US CLOUD Act. In many respects, you govern that. That's a boundary conditions everybody else has to go by. So it's a bit of a double-edged kind of sword here because eventually you fall where others will require to be sovereign of you and then the dependencies start mapping out.

Karl Havard

>> Yeah, agree. And then you can start saying, okay, appreciate from the legislation point of view, totally. But then you've got the physical elements of should, cables be cut and those sort of things. Can you still maintain an AI solution or infrastructure that is operational for the nation that you operate within? there's that angle of sovereignty as well, as well as the financial one. But yeah, we could talk for hours on this by the looks of it. So

Dave Vellante

>> yeah, this is a great— you're knocking it out of the park here, Karl. we have touched upon, some of the pillars. the financial one is probably the most important. And we saw a case where Canva had to rewrite its stack, because it was, blowing through tokens for its freemium models. We saw Uber chewed through its annual token budget in just one quarter. We've seen Microsoft make statements. We're certainly seeing this shift from the large portion of consumption being closed models to now probably 80/20 open closed. So you're starting to see awareness. Maybe people don't think of it as financial sovereignty, but that's ultimately what it is, isn't it?

Karl Havard

>> Yeah, completely. Yeah. there's two things there, isn't there? It's a case of how organizations have sort of not realized how much they're spending on tokens until the invoice hits. And, it's a scary place to be. But then it's a case of if you spent all that money, where does that then go? So, yeah, depends on the model you're using and where the model's hosted. But typically it would go back to the model maker. And which will be resident in the US. So, there's a lot of thought that needs to take place as the market is now fast moving into inference and tokenization. And what we started to see is more token factories emerging across the globe. I think a lot more thought has got to be put into— okay, so if someone is using all those tokens, great, there's revenue being generated, but who does that benefit? Where does that go? And I think there's probably further work to be done in order to agree actually that there's a proportion that the model makers keep, but then also a proportion that the hosting organizations in the nation where they're hosting from also keep.

Dave Vellante

>> Yeah. And we're starting to see this. I wonder if you are as well. There's a lot of talk about AI and the bubble and the circular reference, etc. But you're clearly seeing Monetization from the likes of NVIDIA, downstream companies like Hewlett Packard Enterprise, clearly we're starting to see real clear evidence of productive utilization, i.e., monetization at end customers, big banks, insurance companies, manufacturers, energy firms, et cetera. And I presume you're seeing that as well, which is ultimately what this is all about.

Karl Havard

>> Yeah, completely. it's really interesting. So we're now starting to see, even the financial services sector, banks over here incorporating AI. They've done the work, they've used the foundational models, they've trained it on their data, and they're now offering AI applications within the services that they provide to the customers beyond, call centers and chatbots, actually to make more intelligent decisions on mortgage applications or insurance applications, all those sorts of things. So absolutely. And the software industry as well, right? traditionally, years of developing software on CPUs, they've had to completely change their game and incorporate AI into their services now and are doing so, in anger now. At the moment, we're starting to see a real surge in that area, which is finally great to see. I think 2 years ago, we weren't seeing anything other than burning GPUs to train new frontier models. And there was the race to get the next model out and the race to AGI. But I think we know that marketplace has shaken out and UK PLC, global PLC is now starting to wake up and say we can incorporate this in order to enhance our products and enhance our own customers' experiences. So it's good to see.

Dave Vellante

>> Yeah, your point about inference is right on. We're seeing in our customer surveys that the mix has shifted from— it was 90/10 training inference, and the projection is probably now maybe 60/40 and headed toward 10/90. And the interesting thing is that both are growing. It's not like training is peaking, it is still growing. It's just inference is growing much, much faster. Amit, you have a thought on that, Karl,

Karl Havard

>> or—I was just going to add, as I said, this is the natural evolution of a new industry, isn't it? You need those foundational building blocks in place before you can deploy, and that the models are now in a good shape and now they're being deployed, and hence the inference market is starting to move definitely at pace.

Dave Vellante

>> You know, we should probably wrap here. You've been so generous with your time. There's obviously a big backlash in the United States about data centers and not in my backyard. Whether or not you feel like that's misplaced, let's put that aside for a moment. I think there is a lot of misperceptions, but what you were talking about is taking sort of existing facilities and using those existing facilities. That has got to be very attractive for UK citizens, UK government. That's a unique part, maybe it's not unique, but it's an alluring part of your strategy.

Karl Havard

>> Yeah, we're starting here. We haven't really talked about how we are deploying in the UK yet, but over the next few weeks you'll start to see us getting a little bit bolder in our messaging. But exactly that point. So the old steelworks in Newport, it's been dormant for around about 5 or 6 years. There's still generations of steelworkers in the area. And if you were to go and look at the city at this moment in time of Newport, it's not the most thriving environment to visit or even live in. Right. So our aim is we own a 230-acre site, the old steelworks there. We have access to now 800 megawatts of power, all behind the meter, and only about 20% of that site we will use to deploy our infrastructure. But what that does, it brings the community back to life because we're going to build a vocational tech college as part of it. Newport City Council have asked for sports facilities and a swimming pool. We're going to build that on behalf of them. We sit in the estuary and there's a rare species of newt and there's bird sanctuaries there. We're going to protect that and turn it into like a research area for universities and schools to come to. We're going to create a community center to really create now a destination that used to support that community, which has gone stale over the last few years. And the only way we can do that is by doing it in harmony with the local council, which we've got a great relationship with, and also by the deployment of AI. If we didn't do that, we wouldn't be able to bring it back to life. So, there's something here that goes hand in hand. And if every NeoCloud hyperscaler data center provider could do it this way, then I think we'd start to not only see positive reception to AI and how it's deployed, but also a true benefit to the citizens of where it is deployed to reduce tariffs of energy, subsidized council tax, fun and laughter back into the regions where they've been stale for ages. That's our blueprint that we want to take forward and we hope others will start to do a similar thing.

Dave Vellante

>> Such an encouraging story. A once thriving, community that has gone stagnant for decades because of, economic shifts is now— there's going to be a renaissance here. And as you say, the power is behind the meter, I presume. Well, are you able to bring closed-loop water systems to—

Karl Havard

>> Well, completely. Yeah. everything we're deploying is direct-to-chip liquid-cooled. So, I know the backlash in the US. We have the same in the UK. Some of it's merited because data centers are being done to communities. But some of it is misinformed about water usage. it's like your cooling system in a car or a refrigerator. What the way we're supplying is actually we're bringing our PUEs down below 1.2 and our water usage effectiveness. It's a closed-loop system. So absolutely there's hardly any water wastage at all.

Dave Vellante

>> Yeah. So you seed the system with maybe what it takes to water a golf course, 18-hole golf course in 2 days. And then you're done.

Karl Havard

>> Exactly. And the fact that Newport City Council, when we get back to Steelworks, they want a swimming pool, then there's a perfect offtake for the heat. You know what? What a great vehicle to actually take the heat from the data center and shove it into the swimming pool as well. So, yeah,

Dave Vellante

>> yeah, we got to get more of these stories out. We should wrap, but do you have any final thoughts?

Amit Eyal Govrin

>> I'll leave us all with the cliffhanger question. And Karl, feel free to only answer the parts you feel comfortable, but you've been inside Google, AWS, which hyperscaler who owns sovereign region claim fails your test of sovereignty? And if we're at it, which part of those tests would you feel Era4 would also fail, if at all?

Karl Havard

>> Yeah, I guess to answer that question as fairly and as accurately as I can, It depends on where I'm sitting. So if I'm sitting in the UK or the EU, unfortunately they all fail the sovereign test. If I'm sitting in the US, they all pass and Era4 fails. It's really depending upon the lens and the location you're looking through. Exactly right. Yeah.

Dave Vellante

>> Great. Well, Karl, fantastic conversation. I really appreciate your time. For me and Amit, really awesome to have you here. Thank you.

Karl Havard

>> Thanks for having me.

Dave Vellante

>> Okay. And thank you for watching Sovereign AI: From Strategy to Reality, made possible by our friends at Hewlett Packard Enterprise. For Amit Govrin, I'm Dave Vellante. Thanks for watching.