AI's rapid growth is fuelling a trillion-dollar wave of data centre investment and a race to secure cheap, clean and abundant energy. As developers chase hydro, nuclear and geothermal sources, a new energy infrastructure is taking shape.
How will this race for power redefine AI's next phase and what new business approaches will emerge to sustain it?
At Davos, leaders from government, chips, data centers, and energy argued that AI’s compute boom is colliding with “19th and 20th century technologies,” especially aging grids and slow permitting. Sweden’s Deputy Prime Minister Ebba Busch framed the policy mandate as “speed up or die,” urging technology-neutral rules, faster approvals, and public-private partnerships to restore competitiveness and security. Schneider Electric’s Olivier Blum said the breakthrough isn’t more sensors but making ubiquitous connectivity usable: AI can finally turn abundant operational data into “energy intelligence” that reduces peak demand across buildings and factories.
Arm CEO Rene Haas highlighted a major efficiency lever: shifting AI from cloud to edge devices, which could curb the unsustainable default that “virtually every single piece of AI processing takes place in the cloud.” Nscale’s Joshua Payne argued that AI delivers the highest “economic output per electron,” making siting near surplus clean power (e.g., Norway’s hydropower oversupply) strategically rational—but Europe needs faster grid approvals and more risk capital, since “70 to 80%” of AI startup spend is compute.
Bloom Energy’s KR Sridhar contended the grid is mismatched to digital loads—“the grid speaks English, the chips speak Japanese”—and predicted “bring your own power” models using onsite, dispatchable, low-emissions generation. The panel’s bottom line: AI’s energy burden is real, but it can also become the tool that modernizes the energy system itself.
Welcome, ladies and gentlemen, to our session today on racing for compute and its end game. That's right. You're going to hear a deep and thoughtful look into the future from those who are helping make it. I'm delighted to be with you. I'm Vijay Vaitheeswaran. I'm the global energy and climate innovation editor of The Economist, and it's my great honor to be the immoderate moderator for our feisty panel today, including interventions from the floor. So I encourage you all to start thinking about what you might even ask our expert series panelists, who I'm going to introduce in a moment. I want to start by mentioning this is a livestream, and so I encourage all of you to use your social media to continue to discuss this topic online, and also to mention to you there is some fantastic analytical work that the forum has done. I encourage you to think about a couple of the reports that you can find at the QR code that will be on your screens as well. For those of you watching on the live screen, I recommend to you some of the net zero work as well. And even beyond, the AI growth we know is the phenomena of our age. Perhaps the most powerful, impactful, expensive, and ultimately, perhaps most important transformation since the Industrial Revolution itself. But it raises many questions, and we're finding out there are some challenges to getting to the, the the sunlit uplands, as Winston Churchill called them. Along the way, along the journey. It turns out that though we may be thinking about 21st century intelligence and superintelligence, we're dealing with 19th and 20th century technologies. For example, with the grid, with electrification, the average age of the grid in the United States and Europe is decades old, in France is perhaps approaching 50 years of age, and the distribution systems need to be upgraded. And so we're finding that there are physical constraints. Capital may be less of a constraint, but directing it to the right places and for the novel solutions, perhaps ones that may not be the the easiest to reach for, or the most obvious ones may not be getting the capital that that they might need to scale quickly. So we'll talk about that. And another one of the areas we'll talk about are what are some of the innovative technologies and business models that are emerging that could provide an alternative path to the sort of reflexive drill, baby drill, build, baby build mindset that tends to associate energy. And I've covered energy for more than two decades. And so the initial response to any challenges we need to build out lots of big new power plants, maybe. But there are enabling factors like the grid. There are alternative new technologies like clean technologies. We'll talk about carbon free technologies as well as infrastructure that can be upgraded, intelligent, supplied through AI itself, serving as a virtuous cycle to manage the grid, enhance efficiency. So there's a holistic set of technologies and business models and new ways of thinking that can actually come out of what is currently a crisis. We may well find opportunity to renovate and build the infrastructure the 21st century. Hence, we talk about the race to compute and heading towards that end game, my, panel, I will introduce them, in to my left, of course, to your right, we have Ebba Busch, who's the deputy Prime minister and Minister for energy, Business and Industry of Sweden. We have, next we have Olivier, Olivier Blum, chief executive officer of Schneider Electric, from France and also on the International Business Council. We have, let me see here. Do I have right. Rene Haas, CEO of Ahm United Kingdom, known to many of you chip aficionados in the room, we have Josh Payne, founder and CEO of N-scale, also here from the United Kingdom, and KR Schreeder, well known to those of you at Davos for the last 20 years as a startup founder, I still think of your company as a startup, even though you are now scaled up. And I know we'll hear more about it. KR but it was a celebrated startup from the first wave of clean tech 20 years ago, and really making your impact in the current wave of climate and energy tech now with AI. So we'll hear about that, I'm sure, from KR Schreeder of Bloom Energy from the United States. So give them a nice round of applause to welcome them to the stage. I thought we would start with you, Deputy Prime Minister, if we may. The road ahead. While there are tremendous transformative opportunities, there are challenges. And some of them do have to do with how we think about governance, how we think about market structure, sometimes boring, fiddly things. We generally organized things like energy to be safe and stable and not overly innovative. We wanted to just keep the lights on. Oftentimes, utilities are rewarded not for taking risks or investing in trying new technologies, but on the contrary. And so we've had also flat growth in electricity in Europe and America for a decade or more. And suddenly we're finding a period of tremendous, tremendous growth in these developed markets. And of course, in developing countries, we have not only the prospect of data centers and AI, but air conditioning, electrification, economic growth is even bigger drivers as we're in this age of electricity in which we're we're rapidly accelerating into a super cycle. From a point of view, as a policymaker, how do you look at this and what do you prioritize? How do you get the rules right?
In many ways, this is such a headache era for for policymakers because this is asking, but but it's also, I mean, the best opportunity ever in truly dark and challenging times. There's so much hope connected to this. But, for me, this is asking for policymakers, to become truly technology neutral. And we have shown one again and again, and not least in the area of, in the era of, area of energy that we love to be very ideologically driven and sometimes putting even our, our faith in our own political rhetorics instead of seeing that, well, that will still be out trumped by, by the basic laws of physics. So how to become truly technology neutral and especially in terms of AI, how do you regulate, and, and make a policy for something that is shape shifting faster than a Pokemon is one thing today it's going to be a completely different thing. By the time the Swedish, the Swedes hop around the Swedish midsummer pole singing Little Frog song. And that's just in six months time, hopefully. And then the Swedes will get the sun back.
And planning a reconvening of this session up in Sweden. Exactly by which point we would have solved all these problems.
Hopefully. But that's the one on on technology neutrality, because most democracies are not built for speed, built for stability in the European Union, for sure is built for stability. The second one is then tempo and and the third one is because look at Sweden, we're cutting some of our permitting process by half. That's, that's a great success. But it's still way too many years for you to get connected to, to the grid system. And the third and last part that I'm obsessed with is who do you partner up with nowadays? Because, I think one of the huge success stories behind Sweden's growth journey from 100 years ago, being one of the poorest countries in Europe, having 1 million inhabitants migrate to the US and now being one of Europe's wealthiest countries and most interconnected first movers. Also on on an individual level, I mean, that is because we partner up between public and private sector. We've actually managed to solve problems together, but we've also done it with other countries. And that's becoming even more.
We'll talk with some, business people. That's who we have to your left in a moment. So the partnership point will surely get taken up. Let me press you just a little bit on the idea of speed. It was very impressive, the statistic you cited about accelerating your permitting across Europe. I know the general figure is it takes multiple years to get new projects sited. Some of the concerns companies have is that there might be, you know, the first to file rather than the first to actually deploy. There might be projects that aren't really substantial. They're clogging up the queue. There's traditional slow slowness permitting processes. Could AI accelerate that? So a lot of questions are brewing on both sides of the Atlantic, indeed around the world. And how to speed up this process. Can you tell us what you did to be able to cut that queue, give us a little reveal so that others can learn. But as well, more generally, this idea of increasing the metabolism, the pace of change of government to keep up with the pace of AI or technology or society or business. How realistic is this?
Well, I would say, first of all, you have to be able to zoom out and see it speed up or die. I mean, it, I mean, look at the growth gap between Europe and India today and the EU and and US, the growth gap that we are seeing between the EU and India now will soon be the growth gap between EU and US and India. If we don't see that our own bureaucracy has now become our own worst enemy. So either you start seeing this as truly a matter of competitiveness, speed and second, in these times also a matter of security. So I truly mean it when I say it speed up or die. And I think that is what changed things, at least in Sweden. And the other one is daring, not as a policymaker, not to micromanage, to say that this is the overarching goal and we steer towards that. And then you set timeframes for when you are to reach that goal. I'm not saying we've done this enough. We still have a long way to go, but whenever we've applied that throughout the line, that's when we've actually managed to cut time by so much, because then everyone stops looking at each other and different levels of bureaucracy, saying, that's your responsibility. That's my responsibility. You start saying that, okay, well, this is the the overarching steering, and it can't be, then you have to put competitiveness and security highest up.
Okay. Well, those those are some good principles to lay out. I'm sure the practitioners will have experience on the hard end of the stick. But let's let's hear from them what the reality is. Olivier, let's come to you. Now, your company's global, you're deeply involved with energy technologies, energy efficiency. You understand, grid systems have been to some of your facilities in Italy, for example, where you're involved with the cooling systems that go into these data centers and waterless and even even better recovery systems. So you're deeply enmeshed in that business of thinking about emissions, about energy input and efficiency. What have you learned from the early days of the AI revolution that, gives you encouragement as to what we need to do more and maybe any course correction you would recommend so that we can accelerate to that end game?
Yeah, sure. So what I've learned, first of all, it's a reality. I still hear a lot of people asking a lot of questions about AI. Is it a reality? Not a reality, but it's a reality. And you said it. It's a question of how fast you want to embrace. So that's part number one. Point number two. The good news is, you know, we've worked and think about where we have been in the past ten years since the Paris Agreement. Paris agreement has led on the foundation that we will need energy in the future, but we have to make energy more efficient if we want to achieve our net zero goal by 2050. What I like, and by the way, what I like about the World Economic Forum, we move from a discussion which was only on supply to a discussion which was on supply and demand. So that has led on the foundation, really to make the AI revolution happen, because of course, the new type of data center that we are building in the US or the large hyperscalers, and what we are working with Joss, of course, they consume more energy. Now, the good news is AI needs more compute, compute more energy. But for the first time, probably in the history, we have this solution to make energy more intelligent. If you look at the presentation Schneider we've made ten years ago following the Paris Agreement, we said the future will be more energy efficient if you are able to lay down electrification, because this is a cleaner source of energy, if you are able to lay down on top of it, automation and digitalization. But the reality, what you can get out of digitalization, you can get it through AI. And when I look at all the solutions we've developed at Schneider Electric for the past ten years, they were great, but they were still too complicated for our customers. They were still too expensive. They were still very difficult to communicate among the different players.
Olivier.
If you go to AI, you have a solution in front of you, which is really, really enabling the AI revolution itself.
Well, as I was challenging the Deputy Prime Minister, I'm going to challenge you as well. I've been a business journalist a long time at The Economist. I can remember 20 years ago as being shown digital twins, how they're going to change the world by all the leading companies. You could imagine their names, their present outside this hall. What's different now? And along the way, we've had the Internet of Things will change everything. And and of course, these are incremental advances. They've all had their impact. But we haven't seen step changes. Right. These have been deployed and they've kept lots of consultants busy in the meanwhile. Right. But beyond that, is there something that's transformative about what AI can be applied, whether it's ML or language models or small models that you're seeing in energy systems and the way we use energy to enable the good things we want, like economic growth, like factory electrification, like AI. Can you give us some something more concrete other than intelligence?
It's good with a moderator that's hard to impress.
A you know, the good news, everything is in your question. The biggest difference compared to 20 years ago today, every single device is connected at the level of the device, at the level of the control. They are connected. What does it mean? You have access to data. The biggest difference? Today you are able to capture data from the system. You are able to federate those data. You can bring intelligence. Schneider is not an AI company. Now with the language model, bring to Schneider Electric the way to manage those data to make sure we make them more intelligent for our customers. We are experts in infrastructure. We understand how things works. We understand how data center works. We know how to extract the data. But then when you combine AI on top of that and all the large language model, we can combine our own intelligence of the asset of the physical world with all those data and the intelligence of large language model, then you can bring really energy intelligence to our customer. That's the biggest difference. We are not able to do that only five years ago.
That's that's very good. And it's it's a good answer because we did end up with so much data. But making sense of it is what turns it into intelligence. And now we're getting the tools to do that. You know, the I want to come to our chip maker next about what is the potential for, dramatic reductions in or increases in efficiency. Now, where can we find solutions, in terms of energy and infrastructure inputs into AI. But before that, I just want to follow on from from the comments we had just now, there was a another area which has become politically very sensitive, which is water consumption, another one of the areas that doesn't get as much attention as it should. But here too, there are misperceptions and there's best of class behavior and there's laggard behavior. Right. Some of you who are keenly, sighted will notice that I'm wearing boots, high boots. The last time I wore these boots to Davos, but the last time I wore them was in Abilene, Texas, because I had been invited to get a tour of Stargate at the world's largest AI complex, led by Sam Altman on that example. And they told me, if you don't wear high boots because of rattlesnakes, we won't take you on the tour. So I immediately went and bought, you know, as strong and high a boot as I could get. And now these sands are mixing with the snows of Davos. I imagine there are rattlesnakes here of a different sort. I'm keeping my eyes open for this. But the reason I mentioned this is when I toured the night before. I like to talk to the locals and they have their happiness about having jobs, but they complain about this and that. The rents are going up. Fair enough. Jobs are coming. Yes, but they said. But they're going to take our water. This was the number one complaint. All these data center guys. AI is going to take our water. And I was told from an old timer in Texas, whiskey's for drinking water is for fighting. Yeah. And they were ready to fight for their water. But what I saw when I went on the inside of the tour of their the first of their nine data centers that they've already built up was actually Nvidia chips, water cooled, immersed, like the kind of technology I saw at your facility in Italy. So I was quite aware of what I was looking at. And this facility entire, in the aggregate gigawatt scale, will consume about as much as an American one American household, 1 or 2. And Microsoft has just made an announcement a couple of days ago. They're going to be net contributing in terms of water to the community. So if you choose sustainability from the beginning, if you choose to think about efficiency again, that end game where you want to be, as Wayne Gretzky put it, he skates to where the puck is going to be, not where it is today. Then it is possible to achieve some of these other objectives. So with that in mind, we looked at chips. We looked to the efficiency and the demands that we know that they're going to rise. But what can you give us in terms of hope about how to think about designing for the future, designing with sustainability in mind?
Sure. Yeah. Thank you for the question. So so 15 seconds on on arm. We are a provider of the most ubiquitous CPU ever invented in mankind. The CPU is the digital brain of every single device. I can assure you that everybody in this audience has not only one ARM processor on them, but probably dozens. ARM is the CPU that powers wearables, smartphones, cars, PCs, and data centers. We have a unique view of the world, and today we are the chief CPU in all of Nvidia's Grace Blackwell and Vera Rubin technologies. The secretary made a very interesting comment about how difficult it is for governments to sort of skate ahead of the puck here, even for chip makers, it is hard. But I think the opportunity that has been untapped is running AI at the edge. And what is what is what is the edge mean? The edge is your smartwatch, your smartphone, your PC, your security camera. Today, virtually every single piece of AI processing takes place in the cloud. Every single one that is in of itself is not sustainable. For the reasons that we have talked about. The game is not yet started relative to running artificial intelligence on the edge devices. And you might ask, well, why? Why is that? I mean, it's a known problem. Actually, not so much if you think about, Gemini. Gemini Nano, which introduced about a year ago that was put into a smartphone, the Galaxy S25, the chip that was designed for the Galaxy S25, was designed two years prior. The processor that was designed to run that AI workload. A year before we were shooting at the puck in 2025, with a 2022 design impossible to run that at the edge, I believe there are going to be breakthroughs in the next number of years where you're going to see a distributed computing model that will make this far more efficient, and we've seen this movie before. If you go back in time to mainframes or people who would go to university and all the computing would have to take place, you'd have to be in the college campus at the workstation and then PCs, smartphones, etc., etc. AI is moving so quickly and the sheer amount of data required, it's going to require a level of innovation that we haven't yet seen. Advanced memory technologies, packaging materials. How do you put all that AI processing into something that fits in your hand and your hand doesn't melt? But I'm very optimistic we're going to solve that because we've seen that movie before. And when that happens, I think you'll start to see a big change relative to the size of these data centers and every single AI workload having to run in the cloud.
Sure. There's for those who are aficionados, there'll be the immediate challenge. There's a Jevons Paradox, there'll be some rebound effect. But I think we've seen with the internet and multiple other examples that that's a great race. And we can win that race with continued acceleration, innovation and making efficiency of use a priority, which sometimes it's not. Governments have talked about and the IEA has had a formal efficiency target for energy that has consistently not been met. Right. So it tends to happen bottom up rather than.
Been in semiconductors for for 40 plus years. Right. We're seeing a big memory shortage today in terms of running these models. Shortages are the necessity of invention. And when you see the kind of problems that we're facing, the technology companies tend to rally around fixing these problems. And again, I think we've seen this movie before. I don't want to say it's an easy one to solve, because the scale of data that we talk about with AI is unprecedented, but the industry is quite focused on it and an arm. We're very, very deeply involved in it.
Good. You mentioned scale. It's a nice segue to our next panelist, Josh, at scale. You're you're thinking a lot about this. The, maybe for some of those in the audience who don't know what you do, maybe you can tell them in a sentence what your company does. But as a developer of these centers, help us take take us through this question of, hyperscale versus maybe some of the, differently located places, not just in Virginia or Ireland, but rather, located where the energy is or with some of the new bring your own power concepts that are taking off. How do you think about a way to develop the the amount of energy that is needed in the kind of quantum, but also the cleanliness that people demand without tipping into a political crisis of affordability, which is a top issue around the world today.
Thank you Vijay. So pleasure to meet you all. So I'm the founder of Anyscale. We're a fully vertically integrated AI cloud platform. In essence, what we do is we both build and own the data center. We're building some of the largest clusters on Earth in three separate locations the north, the north of Norway, Portugal and West Texas. And then we're delivering the cloud software to deliver an end to end service. One of the challenges for true scale is access to energy. And I think I want to first touch on a philosophical point, which I think is important. This is a very polarizing topic when thinking about access to energy. Nation states are worried about, you know, pricing out the consumer or simply just having enough contiguous power to serve these industries. If you were to break down, like the makeup of a nation's GDP, most heavy industrial businesses or, heavy, large scale energy use industries can be summarized in one word. It's one sentence. It's really turning energy into value. Doesn't matter whether it's steel, whether it's oil or otherwise. It's important to note that AI infrastructure is the the largest economic output per electron of any industry by an order of magnitude. And so when thinking about the best use of your surplus energy, take the north of Norway as an example. The north of Norway specifically Mno4, is the only constrained grid grid environment in Europe. It has an abundance of hydropower 4.6 terawatts of oversupply that is currently underutilized. So when thinking about the best location on Earth, or specifically in Europe, to build large scale AI infrastructure, both for the benefit of the nation of Norway but also for the use of the entire continent. Without question, the best location to do so is in the north of Norway. So I would encourage nation states to begin to think about it from from this perspective now, in terms of what we can do to really, really improve the outcome of of landing AI infrastructure on the ground to spin the AI flywheel. I think it's a couple of things. First, we need to really incentivize and reform the grid approval process to allow for the the tempo to change. As Ebba and I discussed back in the back room to ensure that we're getting streamlined approvals to stand up large scale, contiguous infrastructure to deliver this, these AI services. The second point is, is just large scale investment in these services. I think there's Europe has a problem in both its fragmentation, but also its lack of access to risk capital for these kind of investments. And so I think a confluence of both these things should allow us to together solve for the scaling challenges of AI across the continent.
So we've talked already to some degree about the pace of change and accelerating that and some of the reforms that government can make. But on the access to Capital Point, I'm surprised because we have heard there's plenty of capital and we know the world is awash in capital. The world is not capital starved. AI companies are not capital starved. And yet you use the word key, operative word Europe. Right. And we were familiar with the Draghi report. We're familiar with some of the challenges in Europe. But as an economic bloc, it's a bigger trading bloc than North America. We know the value here is enormous. The potential is here. So can you explain a little bit more about what it is? What is the capital constraint for the kind of things that you'd like to see that you just mentioned in Europe, because you've got this massive project in Norway and others as well? Is it private equity? Is it the bank of Big Tech that's going to finance this? Where will the capital come from? To overcome the constraints of maybe traditional venture capital or risk capital models here?
Yeah. So I think it's really two points. One, one is venture capital for investment in startups. I think, you know, when thinking about the AI flywheel, really what it is is it's a couple of things. First, you need risk capital investing in really innovative ideas. I think Silicon Valley is just a perfect example of just dense risk capital and dense innovation as a result of access to that risk capital. I think Europe has somewhat of a scarcity for this. And so, you know, reform on on certain, you know, tax benefits for, for investment in AI startups, I think would be one way to really spin the AI flywheel. But the second and most important thing is just lending AI infrastructure on the ground to allow startups to consume this resource. If you think about the overall use of proceeds for an AI startup, in large part, 70 to 80% of that is compute costs, because they need to consume the resource to build their product and integrate it to.
It's a perishable product, right? You can't wait four years for a GE turbine or Siemens. You need to get the speed to power, right?
Exactly. It's all about.
Chips in two years are worth half their value or less. Yeah.
It's all about time to train. And so speed of deployment, speed of access to capital. These are all very important factors in order to spin.
Okay. Well we'll pick that up. That's a nice segue to Kara and we'll come back to you. Deputy Prime Minister. Let's give our man at the end. Sorry to keep you waiting there. I know you're chomping at the bit, on several of these issues, among other things, novel technologies, financing models. I know now you have a rich uncle. I think one of the world's largest private equity funds has made a $5 billion deal with your company to help, maybe accelerate the pace of fuel sales. Can you tell us what is the connection between this magical technology conceived decades ago? In fact, you're a man with a NASA background really thought of for space tech. Mars. How is it connected to the future on Earth with AI? And what can we see at the acceleration pace? Now that you've got a significant bankroller, help us connect it to the conversation.
You know, maybe I just, reset our conversation in the following way. The grid, which was the greatest innovation of last century, was built for the mechanical age. We've been putting Band-Aids on it after Band-Aids to somehow make it adapt to the digital age. Chips don't consume that kind of power. You know? A blogger writing about bloom had a great analogy, so I'm going to steal it. You know, the grid speaks English, the chips speak Japanese, and you spend $150 billion in $1 trillion investment on just language translators. This is high voltage power coming from a grid to bringing it to low voltage changing AC to DC. That's why each one of you carry all these chargers. And in every airport and every security, you have to check that through. It's nonsense. Right? We we have not adapted to the digital age. So in 2001, 25 years ago, we decided that we're going to create a solid state energy conversion device using a molecule that's 100% reliable, 24 over seven, and that's what the world is going to need. You just straight line the Moore's Law and you plot it out. Forget CPUs, GPUs. We are almost there today with everything going on in GPUs and the grid simply cannot support it. Why? Here's the simple reasons why. You know, just today we released our Data Center Power report. We do this every six months. 150 participants participate. The size of an average data center is morphing towards the gigawatt class. Let's put this in perspective. Large aluminum smelters are 300MW. Large petrochemical refineries are hundreds of megawatts. They don't use the grid. They produce their power, a gigawatt of power coming through a local highway called the grid that supports everybody else, and increasing the highway throughput just for this one consumer is bonkers, right? And then you go into inference, which I completely agree with you. If this whole training data centers look like large amounts of power, wait till you see how much inference is going to consume. So this podium that we are sitting in, where a Reuben plus will be ten megawatts of power, that is 10,000 European homes worth of power that needs to come there because of latency. It needs to be very close to certain end users. For many, many applications that are the most lucrative, right? You can't allow latency. How are you going? You can't even solve the transmission problem. How are you going to solve the distribution problem in the last mile.
So you want to leapfrog that.
Leapfrog that challenge the only way the world goes. So the impetus of AI, which really benefits in so many ways, the planet is going to be distributed power. The grid is going to be there. The grid is a big flywheel. We need it. It's and and we need these. When you put this together, what happens. That power is going to be clean. Not because of mandate. Because your children and your grandchildren breathe the air coming out of that.
Absolutely. I was I was stunned. I went to visit one of your facilities in California, in the heart of Silicon Valley, in a residential commercial area. We were able to have a client, one of the world's largest data center companies, at that kind of scale that you just described with your bank of fuel cells there, because the local emissions meet Californian regulations, right. You can install them. You couldn't put a gas plant or.
No noise, no water use.
But why has this been niche until now? It sounds magical. It sounds fantastic.
Because the world did not, as long as the grid was available, that we built a long time ago. Right? And the hyperscalers could simply ride on it, not having to pay for it. Right? Why would they do anything else?
But those days of free riding are over, right? That's right. We've seen the US government announcements in the last few days. It's not exactly. Companies are waking up to this.
Bring your own power.
Bring your own power. Great. Okay. Now I'm going to come to the audience in a moment. Please get your questions together. And we have microphone runners who will come to you. But I promised you a follow on Deputy Prime Minister. So you get a quick follow on or a.
I also want to ask, does that mean that you will solve my problem for just an increasing demand for more grids? Because the problem is, a lot of countries are doing what Sweden tried to do for a few years. Just add more electricity and we're not going to see AI needs electricity. End of story, not energy, electricity. And but but the problem is, if you do as Sweden did for at least ten years, just adding more electricity production or installed production capacity to the system, not also looking at, does that mean that we can increase demand? Can we actually consume more, which we have not? We added almost the equivalent net, of ten large scale conventional nuclear power reactors, but all of it was wind power not being able to be stored. And that means that we have almost had a flat increase of consumption in the last ten years. And this is the reason why we're now, apart from still loving renewables, also now becoming a doing a nuclear renaissance in Sweden, because we need that dispatchable power. That is when and where it is needed. We need that baseload. And that's why I'm.
I think I think you went right there. At the end of the day, dispatchable power is what matters. It is the same as having plumbed water in your home rather than depending on rain water. The wind only blows when it wants to blow. You need that electricity all the time, so you need that firmed up power. So AI is going to help as you you know, as you pointed out, of bringing many, many disparate sources together. But you're going to need both. You're going to need renewables to put more, more capacity in the grid. But you know, you're going to need firming up that power and that firming up the power is going to happen with the molecule for the foreseeable future.
Let's see if we have any questions from the floor. I see a lady with her hand up here. Let's get a microphone. Just some ground rules. Please identify yourself and make it a question. Not a long winded speech. Nobody likes that.
Hi, I'm Chris and I'm from Carnegie Mellon University.
A bit closer, closer.
Oh, I hear myself now. We had a lecture series last year at Carnegie Mellon University, and someone from Microsoft came and was talking about edge computing versus cloud computing for AI, and he talked about the memory problem, and he asked the university solve my problem for me. And so that made me ask a question to you, especially about energy. How much collaboration do you do with universities like public, private, public university collaboration to solve some of your problems? And will that help with the race?
Great. Great question. Who's got something to say? Yeah.
I can start. I used to be a university professor. That's where I started. So we have strong collaboration in multiple ways. You know, like students come in and intern with us, summer internships, we sponsor, university research. Many college campuses are running on our fuel cells, and the students get to actively participate in learning this new technology. We are open to any other forms of collaboration.
Yeah.
Jump in. Yeah. One thing I could also add to that is we at ARM have a great collaboration with Carnegie Mellon. And I was with Farnam, last summer. You have fantastic robotics program. Robotics is a an AI application that is not going to need the cloud nearly as much as some of these other areas. When you think about latency, speed, the efficiency. So you guys have been doing a great job. And ARM with Keio University has a joint program with Carnegie Mellon. So thank you for everything you guys do there.
Is there another question from the floor? While you gather your thoughts, I've got a question for, our dream team here on stage. If you were to envision, the, the idealized or maybe the most successful data center, if we were to meet here ten years from now on the same stage and discuss, what are some of the characteristics of the model of data center AI data center that succeeded? Would it look like what I saw in Stargate, or if I come to visit you in Stargate Norway? Mega-sized. You know, the way that we're envisioning things now, or would it be out in space or under the ocean or. You know, what are we thinking about? If I give you ten years, I'm limiting it. I'm bounding it to that. Not 100, not infinite money in the real world, what's significant pathways of advanced progress? Will we see? Would it be, as some people suggest, that actually our model or theory of AGI is wrong, and after a couple of years it's going to be a bust and we'll go to maybe some, energy efficient, model frugally made model. China is a champion of that model. Maybe India in future, could we actually see a paradigm shift to a distributed model that doesn't actually follow this curve, that this is just a phase of growth? I'm curious. I welcome any provocative comments from anyone that wants to weigh in.
Yes. I think whatever happens at the model layer, what we know for sure is that inference demand is going to increase by at least 1000 x, by simple virtue of implementation of the technology into every product, every job and otherwise. And then we're looking at ten years. It's kind of difficult to think about the density of a data center, given how dense we already are getting on a year on year basis. But immediately I go towards, you know, the model design that we have for our Stargate Norway project, which, you know, is both closed loop cooling. So we're not using a lot of water. It's a single closed loop. So it's we have a water tank and we're not sucking water from the local ecosystem or the local water reservoir. We're designing very efficiently with completely renewable energy. We have an industry leading PUE of 1.15, using some of the most abundant hydropower resources in the region. And then I suppose the final point, which we're not doing at Stargate Norway, but I certainly see the market moving to, is exactly the point you raised onsite generation. As we see large, the cluster sizes getting larger and larger, it's critical that we don't begin to impact the grid because it is legacy and it requires a significant upgrade. And so onsite generation, either through the molecule you referenced or also, you know, small modular reactors, I think there's very likely to be a nuclear renaissance in the early 2030s, which I would fully expect to consume the market. If I look at ten years.
Let me add to that.
Yeah, sure.
I think we can complete each other's sentences here. So, so look, I think there's a saying you can only predict the future if you're inventing it. So, part of what we want to do, and we are fairly confident we will get done in the next five years. A large data center built extremely close to a gas supply, with no fugitive emissions of methane coming out during extraction. Number one.
Well, that require carbon capture.
No, no, this is extraction of the natural gas. Okay, okay. That getting converted at high 60s efficiency, electrical efficiency, which is unheard of. Okay. And it is 800 volt DC going directly in because we don't need to convert that to AC and convert all languages. Right. Directly going in. And then the excess heat coming from our system doing evaporative cooling, doing all the cooling, the tailpipe coming out of our system because we don't burn the fuel is just carbon dioxide and water. We knock out the water. We not only provide water for the cooling, we provide excess water. We pump the CO2 back right where that molecule came from in the ground. That soil that was holding for millennia. A methane molecule allows a carbon dioxide molecule more. So it'll hang on to that more than it held on to the CO2. So it's net zero. This is the vision of the future. This will happen. We will make it happen.
Great. I'm glad you landed on it, sir. We're going to come to the Prime Minister because again, I harken back to the excellent net zero work that's being done at the WAF and with all the partners that have contributed to it. It really is a bold vision and increasingly not such a crazy one. You know, you might have seen some years ago this was a group of dreamers. But because I've been out on the front lines, we're seeing companies stepping up to this as best practice as to have in order to have social license to operate. Like that fella I was drinking with who was telling me about the whiskey and the water in Texas, people want to make sure that grandma's utility rates aren't going up or their, aquifers aren't getting dry. And this is well within the reach of tech companies. Not only that, it's a relatively small investment compared to the value at risk in the chips, or the value of potential economic growth and revenues that they're looking at. So it's it's a problem that can be solved by these companies at relatively achievable cost. And within the technology that are that are within reach, it seems to me, please.
Yeah. And my shareholders are my citizens. If, if you.
Make the most important ones.
Of all. Exactly. And so my vision and my ask is that it's possible that it has become less mysterious what a data center do or what AI actually can contribute with, because I think, now we have a conflict of interest. Also, who should get the power, should access to the grid and electricity go to the factory where you'll be able to see, oh, they're providing 2000 job opportunities for our local community, or it's going to be a gray box. And why are we providing so much power for something that makes cat memes and, you know, so that that is that is my big ask of of of the private sector is how do we make this less mysterious and, and increase public legitimacy by telling people and showing people what use this actually gives. And speaking to the point of capital, we're seeing now a major shift for for Stockholm in many ways, becoming sort of the capital of capital. And one of the reasons for this is, of course, very low debt. We we have a, you know, complete control of our finances, and now we're increasing public spending. So we're doing a lot of heavy investments. We've seen a very hot market for, for, for IPOs and for, for the bonds in in Sweden. Happy to see that. But it's also because we're making money work. Almost 70% of all households in Sweden have their own private savings. And they're on the stock market that those private savings are working on the stock market and our pension funds are working on the stock market. And this is what at least I know, the European Union could make use of this. We would access so much more capital if we were to make sure that all of that, those pension money that we have and private savings were accessed to actually get to work within the European.
Union, capital of capital. I like it exactly.
I think there's a lot of room for innovation beyond large language models. Stringing words together and coming up with an answer is actually not how humans think. It's not actually how we solve problems. There's a huge amount of innovation that needs to take place there that will take place there. My brain consumes about 15W, probably not as efficiently as these gentlemen, but it's 15W and I can do a lot with 15W. We don't quite understand how those 15W has context, why certain triggers cause certain functions to fire. But we will learn that over time and again back to what we will need data centers. Because on the flip side, if you think about the amount of enterprises today that actually use true AI in terms of their workflows, it's very close to zero. Quite frankly, it's it's someone asking questions in the chat, GPT and getting something back. So I do think that the token demand is going to explode because we're nowhere. But on the flip side, I do think that large language models will evolve into something that will be far more efficient.
Olivier.
I'd like to follow up on that point, because it's a very interesting discussion that we are having on data center. And of course, it's true. We should challenge the consumption of data center. But let's look at the other part of the equation. Those data centers, to your point, are building are giving the technology that will help us to solve a large part of the equation. You know, the problem we have today with electricity consumption is not the number of kilowatts is is the peak consumption that you have. You cannot ask the data center. I cannot ask this guy to shut down the data center during the day. That does not work. But you can do it with a factory. You can do it with buildings. You can do it with your home. If you are able to leverage that technology to solve the energy dilemma that we have in the rest of the world. And you know what data center consumed today? A fraction of electricity in the world. Actually, the bigger challenge that we have in front of us is how we reduce energy on the other part. And we can do it with AI. What about your home? You have an electrical panel in your home. Is it connected? Do you have something that helps you to manage your energy consumption? Imagine the day where your electric panel is connected, where I can extract data. I can apply AI and I reduce 2,030% of your energy consumption without you knowing it. Then you solve a big part of the problem.
For sure.
We're getting in that direction. We're almost out of time, but go on.
So I think Josh made a point that I think we all need to really understand a little more clearly. The greatest thing that humans have ever manufactured with electricity is intelligence, highest value. So the question is not can you offer the question is can you afford not to? Okay, if you want to progress in the world, can you afford not to? And the amount of capital it's unbelievable. So I'm you know, I'm from Silicon Valley within a ten mile radius. There are four companies that invest more than $1 billion every single day, 365 days of the year in just CapEx, in just CapEx, not labor, not anything else, just CapEx. Never in the history of humankind has this much money gone into anything, $1 billion a day just for companies. Think of the potential.
I think that's a powerful moment to draw our conversation to a close. The we can see from this conversation that. The transformation of the energy system. And we've heard some important insights about not betting on individual technologies, about moving faster, the pace of scale, particularly in government, maybe a novel forms of capital, maybe thinking about the energy systems themselves more holistically, not just supply oriented, but thinking more broadly about the potential for efficiencies, about novel technologies, about the edge, whether it's in computing or in energy, energy at the edge. So I think they've given us a tremendous amount to think about and to take away with us and apply in our own lives as we think about what comes ahead in our AI transformation of our companies, of our families, of our own lives. Will you join me in giving a round of applause to our experts?
Thank you. Thank you very much. And again, I encourage our.
Audience online, everyone, to please use social media and enjoy your Davos. Thank you.