How Gi21 Capital Sees The Next Wave Of AI Infrastructure Growth

[00:00:00] Big thank you to Denodo for supporting the Tech Talks network and making these conversations possible. Because when your lake house stores the data, the real challenge is getting that data where it needs to go and faster. And your lake house stores the data, but Denodo helps deliver it faster.

[00:00:20] So with real-time access, built-in governance and a business-ready data marketplace, Denodo can help your teams unlock insights without costly duplication. And you can learn more by simply visiting denodo.com. What does it actually take to build the infrastructure that powers all those AI experiences that we take for granted?

[00:00:44] Well, as businesses race to deploy larger models, autonomous systems and data-heavy workloads, the conversation often stays focused on software, GPUs and hype. If we take a look under the hood, behind every AI breakthrough sits an enormous physical infrastructure challenge.

[00:01:05] A challenge that involves power grids, cooling systems, land acquisition, redundancy planning and billions in investment. And my guest today understands that world better than most. He started coding as a teenager, founded VS Hosting at just 17 years old, eventually scaling it into one of Central Europe's leading cloud infrastructure providers.

[00:01:29] And along the way, he has designed and built data centres from scratch that went on to achieve something almost unheard of in this industry. I'm talking about verified 100% uptime. And doing so for more than a decade. And the experiences that he has had has shaped, and those experiences have shaped how he thinks about resilience, redundancy and the hidden engineering decisions

[00:01:55] that determine whether modern infrastructure survives or buckles under pressure. Now, through GI21 Capital, he's investing across AI, robotics and deep tech, all while building one of Europe's largest AI data centre platforms. His projects include liquid-cooled, high-density AI facilities in the Nordics that are designed to support the next generation of compute demand.

[00:02:25] So today, we will talk about what most people never see behind the AI boom. Because my guest will explain why the rules of AI infrastructure are changing so quickly. And how rack densities have exploded from 10 kilowatts to more than 150 kilowatts per cabinet. And why traditional air cooling systems just can't keep pace anymore. And we'll also talk about that growing battle for power, land and grid access across Europe.

[00:02:54] And why powered land could become one of the most valuable assets in tech infrastructure over the next decade. And of course, on the flip side of this, I want to learn more about the bigger picture around sustainability, energy demand, digital sovereignty, and why Europe's AI ambitions depend on solving infrastructure bottlenecks that very few outside the industry fully appreciate. So there are plenty of headlines about AI changing the world.

[00:03:24] But today's conversation shines a light on the physical systems that are quietly making all that possible. But enough for me. Let me introduce you to my guest now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? Hi, my name is Imran Nino and I'm an infrastructure guy. So basically for the last 20 years, since I was 17, I'm in hosting business, cloud business.

[00:03:51] So I'm investor, to put it this way, in infrastructure, data centers, hosting company, cloud companies. So investor slash commercial pilot to make it more interesting. Love it. And as you said, you are an infrastructure guy and very modest too, because you've built and operated data centers at a level most people only get to read about, including achieving verified 100% uptime over more than a decade.

[00:04:19] And for anybody listening outside of tech, that is some feat on its own. But for people listening that are working in infrastructure today, what would you say are the hard lessons about reliability that so many operators still fail to appreciate until something goes wrong, of course? So based on certification design, 100% is not possible. So it's a bit of luck, honestly, and coincidence.

[00:04:45] But when we were designing our data center in Prague, which was 12 years ago already, and we had 100% availability, that's true. You know, it was my very first project. And I was forced to sign a promissory note to the bank to get funding. So I was very nervous. And therefore, we decided to invest more into redundancy, to equipment. So instead of having N plus one UPS system or N plus one cooling units,

[00:05:14] we decided to have two times N plus two, for example, or cooling the same. So basically, we added a few extra units for redundancy, meaning higher capex, of course, much higher reliability towards our customers. So that was basically the approach. And we do hear constant conversations, especially online, around AI demand, how it's driving a new infrastructure race.

[00:05:40] But many people still underestimate what AI workloads actually require behind the scenes. So how different is building infrastructure for AI compared to traditional enterprise cloud environments? I suspect it's a very different picture. So from outside view, it's basically the same. So it's simplified building with lots of energy and cooled air.

[00:06:06] But if we go deeper, we will realize that just a few examples, a rack or single cabinet for servers in the data center, it used to be for 5, 10, sometimes 12, 15 kilowatts. Nowadays, it's more than 150 kilowatts per rack. The utilization of the same space increased significantly. And all complications basically come along with it.

[00:06:35] In one building, you can easily have 100 plus megawatts of power grid connection, meaning that you are not able to cool it by air. You need to do so-called direct-to-chip cooling. That's a different structure in terms of energy, different capex profile, of course, risk profile. So the density of power is much higher than it was in previous generation data centers. So that's the biggest difference.

[00:07:04] On the other hand, the typical enterprise-level data centers tend to be more relying on certification and high redundancy. AI data centers for interference, mostly for hyperscalers, sometimes they don't even have DC generators because they are redundant on the geographic level. So if one data center fails, another one in different countries takes over. And the reason, especially in Europe,

[00:07:33] is that if you're installing a huge amount of diesel generators, especially, to be more specific, it's more than 300 heat input power of generators, which is approximately 100 megawatt data center, you need to have special environmental impact assessment study, which takes two years just to obtain it. And also a capex profile. So some data centers are redone that level of redundancy. And so this is also somehow different.

[00:08:04] And you're developing large-scale AI data center projects across Europe, including liquid-cooled high-density facilities in the Nordics. But I'm curious, behind the scenes there, what design decisions became essential? Once you start building specifically for AI at this kind of scale? So a minute ago, I described two models, like enterprise level of data centers and AI level data centers. We are basically combining both. So we are, in terms of density, it's an AI data center,

[00:08:34] but with high level of redundancy and reliability. That's just the concept. We are designing its AI platform across Europe, probably one of the biggest ones. So we are aiming to have almost 750 megawatts of power. One of the first projects is being permitted in Finland, actually. It's going to be a 100 megawatt data center. And so we are just reacting on a huge demand, which is already on the market. And the free capacities on the most demanding locations,

[00:09:03] flat-D locations, which is Frankfurt, London, Amsterdam, Paris, and Dublin, only 6% is free capacity, which is the lowest number ever. And so there are simply increasing demands and not enough capacity. So we are just expecting that this will get even worse. And therefore, we decided to start constructing our platform. And across Europe, there are so many conversations

[00:09:31] around digital sovereignty and AI competitiveness. Still many organizations, they rely heavily on external cloud providers and overseas compute capacity. From your perspective here, what will it realistically take for Europe to close its AI infrastructure gap? There's a lot of talk around it at the moment, a lot of ambition, but what's needed here? So it's a topic. So on your level,

[00:09:57] there are many discussions about data center, digitalization, AI. And the biggest issue is that there are many speculators. So this is becoming a hot topic. So plenty of new construction companies, operators, everybody wants to be in this industry. And sovereignty is a topic for sure because Europe is lagging behind US and China.

[00:10:26] And so I would say it's, you know, from a strategical point of view, it's important to have critical data in Europe. On the other hand, hyperscalers, I would say they are very active in terms of constructing new data centers in Europe. So they are locating for the data, all of them in Europe. We can see project, basically neighbor project in Finland, you know, basically all of them are there. So it is topic, but it's a more political topic than commercial topic

[00:10:56] because we are aiming to get, you know, commercial clients. But if you read articles from European Union, you know, authorities, they have different priorities done and commercial clients have. But let's see what happens in upcoming years, right? Because Anthropic puts first 1.5 gigabyte of data center into the space. So let's see how this discussion about sovereign and actually move forward in future. Love it. And one topic that rarely gets discussed publicly

[00:11:24] is the challenges that come with securing power, land, and grid access for modern data centers. Can you walk me through what actually happens behind the scenes? We're trying to bring a major AI facility online today. I would imagine there's a lot of opposition and NIMBYs, people saying not in my backyard, et cetera. But what are the big challenges there? There's a magic term, which is called powered lands. Once you have powered lands, you win. To have this powered lands,

[00:11:55] you need to secure grids. And so, yeah, I'm actually right now in Prague and we have nuclear power plants. And the power of this power plant, nuclear plant, is 1 gigabyte. And actually, one of our data centers needs 10% of this, which is a huge number. So just imagine, and it's just 100 megawatts. By the way, we are aiming to get 100 megawatts per location because of this very long permitting process in Europe,

[00:12:23] which should be gone already, in my opinion. But there are other projects which are aiming for 500 megawatts. So just to get such power to the lands takes enormous amount of investments for the grid company and lots of permitting. So it's lots of planning. And so it's just simply too much energy, which needs to be to the land. So that's the issue.

[00:12:52] Problem is also capacity. The tier one location, the flat D locations, they don't have enough capacity for big-scale data centers anymore. So now the hot topics are tier two locations like Poland, Spain, Italy, Nordic, especially because of their energy prices and separate energy markets. But the issue is basically overall everywhere. It's always an issue to get hundreds of megawatts to the land. And I recently found out that there are many

[00:13:21] reserve capacities for speculators. they are just trying to flip a land without any actual project. So legislation is being changed in Nordic, Poland, UK as well, that if you don't have permit in place, you're not being served by the public grid because you don't have anything. So they are now prioritizing a project based on the actual permitting and designing process. Wow. There's some massive figures that you dropped there.

[00:13:51] I almost felt like saying like Doc Brown in Back to the Future, Great Scott. There's so many big figures there. But I mean, on a personal level, you're someone that's gone from entrepreneur to investor, backing companies across AI, robotics, and deep tech. When you look at the next decade of infrastructure, where do you think the biggest opportunities will emerge for founders and innovators and people listening there? What should we be looking for next, do you think? I think the biggest, you know, big thing is robotics.

[00:14:21] Yeah. Just take a look what's happening in China. So it's a different world. So, you know, social care. So when you get old and you want to keep your dignity and you have some, you know, level of comfort, you will just get robots from a social security company or whatever and it will take care of everything. So it will talk to you, you know, it's got everything, buy stuff, you know, all the stuff, organize everything. So the robotics, I think,

[00:14:51] combined with AI with real world models is something which is the real change. It will replace many of industries which we know. So lots of people will be out of the job. They will have to do something else, most likely. It will be a lot of discussions for hours. But probably if you will go for surgery to hospital in upcoming, you know, 10 years, robots will do better job than human. So this is the biggest thing. So therefore, in G21,

[00:15:20] we are supporting entrepreneurs in early stage with their startups. And yeah, but honestly, infrastructure is here for decades and it will be here for another decades in future. So I'm betting on AI infrastructure regardless if it's bubble or not. Definitely it is bubble. Some of the companies are overvalued. It's crazy. But the AI resource need is here and we are not utilizing not even for 1-2%.

[00:15:49] It's its capability. Therefore, I think that the real need for resources, data centers, everything will be huge in the next five years. Yeah, I completely agree with you. And one topic we must bring up as well, of course, is the fact that sustainability and AI infrastructure, they're increasingly colliding in the public debate around these things. So how do you respond to concerns around energy consumption, water usage and long-term environmental costs of supporting

[00:16:18] large-scale AI compute growth? I'm sure it's a question you get a lot, but how do you see this playing out? You know, it's always a question how you produce the energy. It's simply needed. So if we would shut down all factories, lights, everything, it would be great, you know, back to the trees, but the meat is simply here. We need to find a way how to ecologically produce power, either nuclear, wind, whatever. Our data centers, and most of the data centers actually are committed

[00:16:47] to use green energy, I mean, renewable energy already. So for example, in my original company, which I founded 20 years ago, we are already for many years using only 100% renewable energy. And the same applies for our new data center platform. Nordic area, they have plenty of wind turbines, you know, solar panels, they have, you know, excess of this energy. The issue is different. I don't think that it would be issued to make the energy, actually.

[00:17:15] The issue is how to plan, because originally, you know, many years ago, without AI. So data centers used to be very predictable power usage on one hand, and on the other hand, you had very predictable power source, either, you know, water or coal, power plant, whatever. So it was quite easy to plan the grid capacity, but right now, especially with wind and solar panels, you never know how much energy will produce next week. You can predict

[00:17:45] like 24 hours, but not more very precisely. And on the other hand, GP load is also different in data centers. If you start computing something, it also, you know, it's also very variable. And sometimes, those requirements, they just don't match. So therefore, there are like bouncing batteries in place. So technology somehow anticipating this new situation. But on general level, I wouldn't say that the real impact of data centers is big. So of course,

[00:18:15] it's a huge energy usage. But on the other hand, for example, we are looking away how to use the waste heat from the data centers, 100 megawatts, so you can heat up the whole city, basically. And honestly, what I actually do like politically-wise, if the ecology goes in the same way like economic profile, like, so if you, for example, this waste, heat, re-usage, if you save money, save costs, make new revenue stream, and it also, it's ecological, it's win-win for everybody, right?

[00:18:44] So I don't think it's an issue, quite honest. And finally, if business or tech leaders are listening today and they want to prepare for that next phase of AI infrastructure, the evolution that's coming their way, are there any mindset shifts or maybe even investment priorities that they should already be thinking about now? Any advice there? So I would say AI data centers are, in general level, quite safe investments, something which is sitting between real estate and infrastructure.

[00:19:13] There is nothing much which can be, you know, invented than direct-to-chip cooling. So we are approaching some, let's say, kind of limits how much density we can put into one server. On the other hand, there's another part of AI infrastructure which is the servers itself, GPUs, and I'm quite cautious here because it's very CapEx heavy and there's a huge competition market AMD, NVIDIA, Intel, and others

[00:19:42] and you never know in one, two years which new chip will be released on the market and so if we are investing into data centers, we are more investing into the building and equipment than into the actual servers. It's more on our customers because, you know, payback time two, three years it's quite risky and at the same time it's very dynamic it's a very dynamic industry. So I would be very cautious investing to drive hardware because it will change for sure. It will be more efficient, more power,

[00:20:12] you never know. So this is very, very dynamic. Exciting times ahead and for anyone listening that would like to carry on this conversation, maybe find you or your team online, find out more about anything we talked about today. Where would you like me to point everyone listening? Where can they find more information? Good question. So they can find more information on gi21capital.com so I'm trying to publish more information about what's going on on my LinkedIn profile. So I'm, again, trying to be, you know, more up-to-date

[00:20:42] publishing what's going on in this industry and, yeah. Awesome. Well, we covered a lot in a short amount of time there from what that shift to AI infrastructure demands from operators and builders today and solving Europe's critical shortage of AI compute capacity, what that requires, so many big takeaways. So I will include all the links that you mentioned there. So for anyone listening that want to find out more information, go to techtalksnetwork.com. You'll find a blog post associated with this episode. Click on any of those links

[00:21:11] and you'll be able to find out more information and even connect with my guest today. But a big thank you for sitting down with me in your car, no less, as well, in searing heat to speak with me. I really appreciate your time. Thank you. Thank you very much. I think when people talk about AI, the focus usually lands on the models, the apps, or the companies racing for market dominance. But conversations like this are a reminder that none of it exists without infrastructure, power, cooling,

[00:21:41] redundancy, land, connectivity, and engineers that are willing to bet big on what that future demand will look like years before anybody sees it. And there's some big stats in there too. I mean, a single AI data center requiring the output of a major power plant sounds almost surreal. Until you realize it's quickly becoming the new reality of compute. And at the same time, there's that balancing out between sustainability, commercial land,

[00:22:09] and political priorities across Europe. So if today's conversation sparked your curiosity about the future of AI infrastructure, data centers, and the hidden systems that are shaping tomorrow's economy, I'll add links to everything. So you can find a blog post over at techtalksnetwork.com. You'll find everything that you need to carry on this conversation. And over to you. Are you and your business prepared for the scale of infrastructure that AI will require

[00:22:39] over the next decade? Or are you possibly underestimating the challenge completely? Let me know. But that's it for today. I'll return again tomorrow with another guest. But thank you for stopping by today. And hopefully, I'll get a chance to speak with you again real soon. Bye for now.