Is AI Compute A Public Utility? | Paradigms

Description

If compute is to AI what energy is to an aluminum plant, what happens when we start treating it like a utility? As 2025 draws to a close, tech firms have already poured over $400 billion into AI infrastructure globally – and AI continues to reshape our world with each passing second.

While AI tools and products become woven into government services, businesses, and our daily routines, we stop to ask: shouldn't the providers of compute infrastructure be considered public utilities? And if so, what opportunities arise?

In this episode, we revisit our conversation from May, which remains more relevant than ever, and dive into the current debate on AI compute and the risks of rapid scaling without proper regulation. Listen along as we explore potential solutions against keeping production capacity in private hands – from market liberalization (think energy markets) to private investment over state subsidies, state loans, and hybrid models.

How can we prioritize fairness in the AI compute market and what is lost when the potential profits of the world's richest compute companies far outweigh the reputational risk of rolling back their sustainability plans?

Link to Max's LinkedIn post: https://ltmv.io/max-post

Sources: CNN Business

Paradigms is a podcast produced by Leitmotiv – an independent think tank focused on operationalizing values in our digital future.

Spotify: https://open.spotify.com/episode/6SIqDnLt5wAl6jZRboqgSF?si=2e29d270de83472d
Apple Podcasts: https://podcasts.apple.com/us/podcast/is-ai-compute-a-public-utility/id1842815462?i=1000743083552
YouTube: https://youtu.be/ADPMxjw-4xA

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Transcription

Speaker #0
This is the next chap.
Speaker #1
Welcome to Paradigms, a podcast by Lightmotive, the independent digital policy think tank that exists to operationalize values in our digital future.
Speaker #2
In each episode, we unpack the latest developments and debates in our digital world. Together, we try to make sense of what is happening to our digital future. who's shaping it and what we can do to change its path.
Speaker #1
My name is Susanna Ludwikova and I'm here to ask the typical questions.
Speaker #2
And my name is Max Schulze and I'm here to give simple answers. At Labmotiv, I think about the policies that shape our digital future and how to operationalize our societal values within them.
Speaker #1
And my job is to bring our thinking to your ears, eyes and minds. In our podcast, we invite you to think with us as we unpack the building blocks of our digital world and work towards a better future that amplifies the best of our core values and our humanity. Today, I would like to talk about what you wrote recently. And it's not too long of a piece. So... I'll be quick. In your short opinion piece, this one you posted two weeks ago on your LinkedIn profile. And it's the one where you compare AI computes to a public utility. And you point out how artificial intelligence is essentially embedded into governmental services and businesses everywhere. And then you also ask the question, aren't the providers of compute infrastructure public utilities? And I believe what you're highlighting here, you can correct me if I'm wrong, is the growing... ubiquity of AI in our everyday lives and AI growing into a whole infrastructure as well as the concerns about lacking regulation and while we put aside the obvious yet still true claims that there is no such thing as free only the air you breathe I believe there's more nuance to it and maybe more ways we can think about what's fair and useful and especially when in compute power equals basically performance and performance equals profit. So, Max, in your words, could you explain and talk a little bit about why today we should start considering AI computes should become a public utility? And maybe to even first explain, what does AI compute even consist of?
Speaker #2
Yeah, I think it's a complex topic that I would try to unpack a bit to make it simple. I think my, who knows, for people who know me well, they know my favorite analogies, of course, of electricity. and When we think about the last industrial revolution, what was the key ingredient to that was energy in some shape or form. And a lot of that energy came to the industry through electricity, for example, with steam generators. And so for that industrial revolution, the key ingredient was access to electricity and energy in some form. As that evolved, electricity... Good networks were everywhere and they also became part of all of our lives. So, of course, everybody now has electricity at home. It's really embedded into our lives. And it's, I think, realistic to say that nothing would work in our society today without electricity. And so in that sense, electricity relatively quickly became a utility. Utility being defined as something that is critical to society and the absence of it. would have devastating effects on society. The same is true for water. Electricity is a good one always. And there's other utilities, of course. And because it's so critical to society, it can be considered a critical infrastructure for us. And in that case, of course, we want a lot of safeguards in place and regulation to make sure it's always available. And that's what a public utility to a certain degree is. It's a state-authorized monopoly, often owned also by the state, with very, very heavy regulation that makes sure that at any given time we have electricity. And now if, and I think we're still in the process of that, if we add AI functionalities to the products we use in our daily lives, one could say that's maybe not so critical, but if we add it also to every product, every business, in our society as well as government services. even maybe to a certain degree, adding AI into the energy system and electricity system or water utilities. So gluing it onto everything, one could argue that the ingredient of that AI functionality, which is computation, is as critical to us as electricity is to us. And so to make compute not such a complicated concept, I would just consider that the fuel of AI functionality that we use in all our products and our society. I hope that makes sense. First check to users at that.
Speaker #1
Sure. But can you maybe give one example of AI compute just so that we can imagine it?
Speaker #2
Yeah, I think the most simplistic form is there are ways, of course, now to run large language models. and other models on your local computer. There's open source models that you can run on your laptop. And then the laptop essentially turns electricity or for either from the battery or from the grid into computation that that model can use. And that computation is nowadays often coming from what are called GPUs, which are graphical processing units, which used to be primarily used for gaming and rendering of 3D. animations and things like this. And these specialized chips turn electricity into very high performing mathematical computing operations that are needed to power the models that we now consider AI.
Speaker #1
Got it. And with chips, are we also talking about companies such as Nvidia?
Speaker #2
Yes, of course. They are very prominent, though there's alternatives as well and a lot of standard computing chips like from intel or from apple's own m series that is made by arm they contain gpus on inside the the main chip um so when i say you can run models locally it's because you already have non-nvidia gpus also in in most laptops for example these days right and maybe then tying it to the utility um question so First of all, is it a utility? I think so, especially if we want to deploy AI everywhere. I think then it has to become a utility in that sense. Now, the real problem that I have and why I keep raising this question is that, of course, the companies providing that compute at a larger scale, so not your laptop, but cloud infrastructure companies and IT infrastructure providers, these companies then play de facto So... the role of the public utility producing the compute for us. And having that production capacity in private hands may pose a risk. To stick with my electricity analogy, we actually were in this situation before because a lot of energy companies in Europe used to be privately owned. And at their peak, there's example of a German energy company that was... At some point, one of the largest in the world called E.ON, at their peak, they made 150 billion euros a year revenue. And if a private company gets to that level of revenue and profitability and things like this, it's very, very hard to regulate them. And it's very hard to put brakes on them because, of course, they can deploy vast amounts of money to lobbying and blocking regulation. employ so many people that they can very simply threaten the government to let go of people, which would have bad outcomes for politics. And so they become so powerful that it's almost impossible to regulate them, which is why in Europe we had something called the market civilization, where we basically forced these companies to split up into smaller chunks. you could say. So they were still allowed to produce the electricity. You could say they were still allowed to produce compute. But for example, they had to trade, they had to sell that compute, not directly to customers anymore, but they had to, or the electricity in that case, but to go through to an open marketplace or an exchange to be precise. And that European Energy Exchange, it's called, there's many of them actually. They allow a basically very transparent pricing and trading of electricity. which in turn brings prices down. But also it makes sure that, first of all, we have full transparency how much electricity there is in the system. And we can see if we don't have enough, then the prices go up. We have very clear signals. And, and that's probably the most important thing, anyone can trade on that exchange, which also means that anyone who has a generator to make electricity at all. for example, a solar panel on the roof, can now participate as well. So you don't need to be a gigantic company to be able to be in this market. Anyone can be trading in that market. And that is very good for utility because it makes sure that we always have enough of it. The production of it is balanced across many different actors and the market is open and transparent so that anyone can... participate and it can be regulated.
Speaker #0
The world gets bigger.
Speaker #2
Yeah, to put that maybe a bit in perspective, the current, let's call them compute utilities, cloud infrastructure companies and IT infrastructure companies. I said already before, at its peak, E.ON made 150 billion euros of revenue per year. Today, we still have a company in Europe. EDF in France, which is state-owned, which still makes about 150 billion euros a year revenue owned by the French state. And these were considered so big that they were already hard to regulate. And to compare that to the current compute utilities, just one company alone, Microsoft, makes 250 billion euros revenue per year. So these compute utilities are already orders of magnitude bigger than what we had in the energy system, which should really trigger the alarm bell for us to say, well, now is the moment to apply the same methodology we applied to electricity to ensure an open market and transparency and the ability to regulate and ensure that all citizens have access to the compute they might need for their AI functionalities. We should apply the same framework we applied for energy. to compute.
Speaker #1
Right. I think this very well connects to the event that you mentioned also in the post. You participated in a political AI and tech summit in Brussels. And I think there you mentioned that a leading economist from the think tank encouraged private investment into the compute infrastructure rather than state subsidies. So in the post, you mentioned that you agree with this, that you prefer privately financed compute infrastructure. over state subsidies and why so?
Speaker #2
Well, I think that state subsidies can often lead to a distortion of the market. So, I mean, the plan is basically to say we'll hand out a lot of subsidies to European compute utilities and hope that they will then be more competitive or that they can overbuild. capacity, I think it will lead to only one thing is that we'll end up with way too much compute and way too many data centers because the subsidies would be pushed into a market that is not transparent. So we don't know how much compute we actually need. We don't know how much supply there really is right now. We don't know if we need more of it right now. And we are just subsidizing for the sake of growing something. So I would rather, I prefer private investors because private investors in that sense can determine if it makes sense to invest, if there is demand, if it's sensible to do so now. So because they don't, of course, want to lose the money that they're investing. So they're a little bit more cautious and a little bit more driven by market forces. However, I want to be really careful here because in the current compute utility system, that's not really feasible to invest. There's many reasons for this. We wrote a few papers on this because the market, again, there's no exchange. So if I now decide, oh, I want to become a compute utility, I want to invest in data centers and equipment to provide compute to customers. I basically create that capacity and they can't trade it with anyone because most of the customers are locked in to the utilities of the existing providers. So even though I might want to, as a private investor, invest and participate in this market, I just can't. So another reason why I'm saying no state subsidies is that I would rather see the state fix the market. So applying the same framework of market liberalization that we did in the energy system, create that exchange so that we have transparency. How much compute do we need? Is it too expensive right now? Do we need to invest in supply? And then after the transparency is given, private investors together maybe with the state can then invest into more infrastructure, into creating more compute. But even in that case, I would prefer the state to give out loans rather than subsidies because the loans could potentially earn interest. potentially be paid back at some point. So the taxpayer is not just handing out money, because that's at the end of the day, subsidies come from our pockets, handing out money for something we don't know if we need it. But with the loans, we say, well, if an investor thinks that it's worth building Compute now, and we have market signals and an exchange that tells us, yes, now is a good time, then we can facilitate that investment with a loan. which we then get back after five to 10 years. That's something I'm very much in favor of. But just blank check subsidies into infrastructure that we don't know we need, I think is very difficult.
Speaker #1
But if a private investment is a go for you, you've mentioned that in the post, you mentioned that Amazon sort of immediately offered to lead this investment in the discussion. And you labeled this a low risk investment, or at least. one with a positive expected return. So could you explain this a little bit more and also the tensions that emerge here?
Speaker #2
No, of course. I think what I see problematic here is that on the one side, the economist calls for private investment, which again, I agree with. But on the flip side, only very few private investors can invest in this market. And those are the ones who already have a chokehold on the market. And those are often the big cloud providers. They have locked in customers. They have the world's largest sales forces to sell more compute to customers. They even release the models, the AI models. They push us to use them. So they basically can create their own demand. So it makes sense for them to basically say, yeah, we can invest more. We can always invest more because we know we can create more demand. and that's a little bit that that's basically like private investment from a handful of people who already know that they own the market and that's to me a bit um we have to be very very careful with that because it's it's a little bit the analogy that is probably the most practical here is basically basically somebody who who already has electricity contracts with every company in europe um saying that yeah of course would be happy to build more power plants because again they already have all the contracts so for them it makes sense and at the same time then asking government to basically make sure that these companies use more electricity which is the equivalent of them asking governments to push companies to use more ai so that's really problematic especially given that these companies are now so big and so powerful that we can barely regulate them So instead of letting three, four or five companies invest even more and get even bigger in the space, we should make sure that more private investors can access this market. But that really requires us to completely change how the compute market works today.
Speaker #1
Yeah, and also as an opposing alternative or as a possible solution to this. One of your commenters, Marleen, sort of explained how it works in the Netherlands, where the Dutch utility companies are rarely fully owned by the government anymore. But still often they form sort of a hybrid model. So this model is partly public and partly operating independently. And in this case of public utilities, she said it works for energy and water companies that are publicly owned, but they operate efficiently and they are still market oriented. And that such a model would also fit in the case of compute. So public guarantees, but private efficiency. What are your thoughts on this?
Speaker #2
I very much agree. I think... Why I always look to those comparisons, and I really appreciate that comment as well, is I'm just trying to say this is not a new problem. If we just treat compute as a utility, which I think we can at this point all agree that it is, it's critical to society, it's critical to the functioning of our economy, it's even probably critical to our national security at this point. So if we just say it is a utility, then we can apply everything we've learned from energy and water. And there's a lot of hybrid models, partly state-owned, completely privately owned. We have all the regulations already developed. We know how to do it. Let's just apply our lessons learned to compute. It is really just, it's not so different. But I think a lot of people want us to believe that it's somehow different. It really is not.
Speaker #1
And finally, something you also already referred to. but today many Tech hyperscalers such as Microsoft and Google are preparing to use nuclear energy to power their AI data centers. And they partner with public utilities to fund and operate the plants. And since the tech industry operates much faster than the power industry, which is usually government run, and you could also say slower to react. What are the growing challenges here?
Speaker #2
I think it's interesting that companies... like I said, that have twice as much revenue as the largest utility in France, which owns how many nuclear power plants? I don't know from my head, but I think it's more than 40 or 50. These companies need to lean on the state or lean on the utilities to get their job done. I think that's a bit ironic, in my opinion. I think the biggest risk and We already see that with Musk's supercomputer that he built is that until the nuclear is available, which is questionable if it should be available to begin with, most of these companies will fall back to using gas generators. Because that is the cheapest, fastest way to get more energy very quickly. And we see that Musk is powering his, I think, 50 or 60 megawatt data center completely with gas generators. We've seen similar trends in Ireland and it's very easy to point out that probably, especially given that most other sectors are decarbonizing, this would mean that the compute space is basically carbonizing its energy supply. I think what's even more interesting is that especially the big tech companies were the first to really run the narrative of renewable energy. They're the largest buyers in the world, but it's pretty clear now that it's not enough. And if it's not enough, they are willing to go either nuclear or gas to just fill their needs right now. Because the money they can make, I think, is much greater than the reputational risk of rolling back their sustainability plans. And I think that's a bit disappointing. It's maybe the simplest word.
Speaker #0
amazing wow you were jackass no i'm definitely disappointing i agree well thank you very much for talking to me today and see you next time thank you zana for the great questions

About Paradigms

Unpacking the Leitmotiv of our digital future – where we are now, where we're going, and who is setting the direction? Each week, Zuzana and Max discuss the latest developments and public debates in the digital sphere to figure out our future and ask: Is this where we, as a society, want to go?

This podcast is part of our mission as an independent think-tank focused on operationalizing values in our digital future. Together with you, we want to imagine a Leitmotiv (’vision’) for a digital world that is grounded in societal values and mobilize our power to shape it. Follow us on LinkedIn, Bluesky or Instagram to stay on the pulse of digital policy or the latest tech talk and learn everything about how we’re building a values-centered digital future.

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Description

Link to Max's LinkedIn post: https://ltmv.io/max-post

Sources: CNN Business

Paradigms is a podcast produced by Leitmotiv – an independent think tank focused on operationalizing values in our digital future.

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Transcription

Speaker #0
This is the next chap.
Speaker #1
Welcome to Paradigms, a podcast by Lightmotive, the independent digital policy think tank that exists to operationalize values in our digital future.
Speaker #2
In each episode, we unpack the latest developments and debates in our digital world. Together, we try to make sense of what is happening to our digital future. who's shaping it and what we can do to change its path.
Speaker #1
My name is Susanna Ludwikova and I'm here to ask the typical questions.
Speaker #2
And my name is Max Schulze and I'm here to give simple answers. At Labmotiv, I think about the policies that shape our digital future and how to operationalize our societal values within them.
Speaker #1
And my job is to bring our thinking to your ears, eyes and minds. In our podcast, we invite you to think with us as we unpack the building blocks of our digital world and work towards a better future that amplifies the best of our core values and our humanity. Today, I would like to talk about what you wrote recently. And it's not too long of a piece. So... I'll be quick. In your short opinion piece, this one you posted two weeks ago on your LinkedIn profile. And it's the one where you compare AI computes to a public utility. And you point out how artificial intelligence is essentially embedded into governmental services and businesses everywhere. And then you also ask the question, aren't the providers of compute infrastructure public utilities? And I believe what you're highlighting here, you can correct me if I'm wrong, is the growing... ubiquity of AI in our everyday lives and AI growing into a whole infrastructure as well as the concerns about lacking regulation and while we put aside the obvious yet still true claims that there is no such thing as free only the air you breathe I believe there's more nuance to it and maybe more ways we can think about what's fair and useful and especially when in compute power equals basically performance and performance equals profit. So, Max, in your words, could you explain and talk a little bit about why today we should start considering AI computes should become a public utility? And maybe to even first explain, what does AI compute even consist of?
Speaker #2
Yeah, I think it's a complex topic that I would try to unpack a bit to make it simple. I think my, who knows, for people who know me well, they know my favorite analogies, of course, of electricity. and When we think about the last industrial revolution, what was the key ingredient to that was energy in some shape or form. And a lot of that energy came to the industry through electricity, for example, with steam generators. And so for that industrial revolution, the key ingredient was access to electricity and energy in some form. As that evolved, electricity... Good networks were everywhere and they also became part of all of our lives. So, of course, everybody now has electricity at home. It's really embedded into our lives. And it's, I think, realistic to say that nothing would work in our society today without electricity. And so in that sense, electricity relatively quickly became a utility. Utility being defined as something that is critical to society and the absence of it. would have devastating effects on society. The same is true for water. Electricity is a good one always. And there's other utilities, of course. And because it's so critical to society, it can be considered a critical infrastructure for us. And in that case, of course, we want a lot of safeguards in place and regulation to make sure it's always available. And that's what a public utility to a certain degree is. It's a state-authorized monopoly, often owned also by the state, with very, very heavy regulation that makes sure that at any given time we have electricity. And now if, and I think we're still in the process of that, if we add AI functionalities to the products we use in our daily lives, one could say that's maybe not so critical, but if we add it also to every product, every business, in our society as well as government services. even maybe to a certain degree, adding AI into the energy system and electricity system or water utilities. So gluing it onto everything, one could argue that the ingredient of that AI functionality, which is computation, is as critical to us as electricity is to us. And so to make compute not such a complicated concept, I would just consider that the fuel of AI functionality that we use in all our products and our society. I hope that makes sense. First check to users at that.
Speaker #1
Sure. But can you maybe give one example of AI compute just so that we can imagine it?
Speaker #2
Yeah, I think the most simplistic form is there are ways, of course, now to run large language models. and other models on your local computer. There's open source models that you can run on your laptop. And then the laptop essentially turns electricity or for either from the battery or from the grid into computation that that model can use. And that computation is nowadays often coming from what are called GPUs, which are graphical processing units, which used to be primarily used for gaming and rendering of 3D. animations and things like this. And these specialized chips turn electricity into very high performing mathematical computing operations that are needed to power the models that we now consider AI.
Speaker #1
Got it. And with chips, are we also talking about companies such as Nvidia?
Speaker #2
Yes, of course. They are very prominent, though there's alternatives as well and a lot of standard computing chips like from intel or from apple's own m series that is made by arm they contain gpus on inside the the main chip um so when i say you can run models locally it's because you already have non-nvidia gpus also in in most laptops for example these days right and maybe then tying it to the utility um question so First of all, is it a utility? I think so, especially if we want to deploy AI everywhere. I think then it has to become a utility in that sense. Now, the real problem that I have and why I keep raising this question is that, of course, the companies providing that compute at a larger scale, so not your laptop, but cloud infrastructure companies and IT infrastructure providers, these companies then play de facto So... the role of the public utility producing the compute for us. And having that production capacity in private hands may pose a risk. To stick with my electricity analogy, we actually were in this situation before because a lot of energy companies in Europe used to be privately owned. And at their peak, there's example of a German energy company that was... At some point, one of the largest in the world called E.ON, at their peak, they made 150 billion euros a year revenue. And if a private company gets to that level of revenue and profitability and things like this, it's very, very hard to regulate them. And it's very hard to put brakes on them because, of course, they can deploy vast amounts of money to lobbying and blocking regulation. employ so many people that they can very simply threaten the government to let go of people, which would have bad outcomes for politics. And so they become so powerful that it's almost impossible to regulate them, which is why in Europe we had something called the market civilization, where we basically forced these companies to split up into smaller chunks. you could say. So they were still allowed to produce the electricity. You could say they were still allowed to produce compute. But for example, they had to trade, they had to sell that compute, not directly to customers anymore, but they had to, or the electricity in that case, but to go through to an open marketplace or an exchange to be precise. And that European Energy Exchange, it's called, there's many of them actually. They allow a basically very transparent pricing and trading of electricity. which in turn brings prices down. But also it makes sure that, first of all, we have full transparency how much electricity there is in the system. And we can see if we don't have enough, then the prices go up. We have very clear signals. And, and that's probably the most important thing, anyone can trade on that exchange, which also means that anyone who has a generator to make electricity at all. for example, a solar panel on the roof, can now participate as well. So you don't need to be a gigantic company to be able to be in this market. Anyone can be trading in that market. And that is very good for utility because it makes sure that we always have enough of it. The production of it is balanced across many different actors and the market is open and transparent so that anyone can... participate and it can be regulated.
Speaker #0
The world gets bigger.
Speaker #2
Yeah, to put that maybe a bit in perspective, the current, let's call them compute utilities, cloud infrastructure companies and IT infrastructure companies. I said already before, at its peak, E.ON made 150 billion euros of revenue per year. Today, we still have a company in Europe. EDF in France, which is state-owned, which still makes about 150 billion euros a year revenue owned by the French state. And these were considered so big that they were already hard to regulate. And to compare that to the current compute utilities, just one company alone, Microsoft, makes 250 billion euros revenue per year. So these compute utilities are already orders of magnitude bigger than what we had in the energy system, which should really trigger the alarm bell for us to say, well, now is the moment to apply the same methodology we applied to electricity to ensure an open market and transparency and the ability to regulate and ensure that all citizens have access to the compute they might need for their AI functionalities. We should apply the same framework we applied for energy. to compute.
Speaker #1
Right. I think this very well connects to the event that you mentioned also in the post. You participated in a political AI and tech summit in Brussels. And I think there you mentioned that a leading economist from the think tank encouraged private investment into the compute infrastructure rather than state subsidies. So in the post, you mentioned that you agree with this, that you prefer privately financed compute infrastructure. over state subsidies and why so?
Speaker #2
Well, I think that state subsidies can often lead to a distortion of the market. So, I mean, the plan is basically to say we'll hand out a lot of subsidies to European compute utilities and hope that they will then be more competitive or that they can overbuild. capacity, I think it will lead to only one thing is that we'll end up with way too much compute and way too many data centers because the subsidies would be pushed into a market that is not transparent. So we don't know how much compute we actually need. We don't know how much supply there really is right now. We don't know if we need more of it right now. And we are just subsidizing for the sake of growing something. So I would rather, I prefer private investors because private investors in that sense can determine if it makes sense to invest, if there is demand, if it's sensible to do so now. So because they don't, of course, want to lose the money that they're investing. So they're a little bit more cautious and a little bit more driven by market forces. However, I want to be really careful here because in the current compute utility system, that's not really feasible to invest. There's many reasons for this. We wrote a few papers on this because the market, again, there's no exchange. So if I now decide, oh, I want to become a compute utility, I want to invest in data centers and equipment to provide compute to customers. I basically create that capacity and they can't trade it with anyone because most of the customers are locked in to the utilities of the existing providers. So even though I might want to, as a private investor, invest and participate in this market, I just can't. So another reason why I'm saying no state subsidies is that I would rather see the state fix the market. So applying the same framework of market liberalization that we did in the energy system, create that exchange so that we have transparency. How much compute do we need? Is it too expensive right now? Do we need to invest in supply? And then after the transparency is given, private investors together maybe with the state can then invest into more infrastructure, into creating more compute. But even in that case, I would prefer the state to give out loans rather than subsidies because the loans could potentially earn interest. potentially be paid back at some point. So the taxpayer is not just handing out money, because that's at the end of the day, subsidies come from our pockets, handing out money for something we don't know if we need it. But with the loans, we say, well, if an investor thinks that it's worth building Compute now, and we have market signals and an exchange that tells us, yes, now is a good time, then we can facilitate that investment with a loan. which we then get back after five to 10 years. That's something I'm very much in favor of. But just blank check subsidies into infrastructure that we don't know we need, I think is very difficult.
Speaker #1
But if a private investment is a go for you, you've mentioned that in the post, you mentioned that Amazon sort of immediately offered to lead this investment in the discussion. And you labeled this a low risk investment, or at least. one with a positive expected return. So could you explain this a little bit more and also the tensions that emerge here?
Speaker #2
No, of course. I think what I see problematic here is that on the one side, the economist calls for private investment, which again, I agree with. But on the flip side, only very few private investors can invest in this market. And those are the ones who already have a chokehold on the market. And those are often the big cloud providers. They have locked in customers. They have the world's largest sales forces to sell more compute to customers. They even release the models, the AI models. They push us to use them. So they basically can create their own demand. So it makes sense for them to basically say, yeah, we can invest more. We can always invest more because we know we can create more demand. and that's a little bit that that's basically like private investment from a handful of people who already know that they own the market and that's to me a bit um we have to be very very careful with that because it's it's a little bit the analogy that is probably the most practical here is basically basically somebody who who already has electricity contracts with every company in europe um saying that yeah of course would be happy to build more power plants because again they already have all the contracts so for them it makes sense and at the same time then asking government to basically make sure that these companies use more electricity which is the equivalent of them asking governments to push companies to use more ai so that's really problematic especially given that these companies are now so big and so powerful that we can barely regulate them So instead of letting three, four or five companies invest even more and get even bigger in the space, we should make sure that more private investors can access this market. But that really requires us to completely change how the compute market works today.
Speaker #1
Yeah, and also as an opposing alternative or as a possible solution to this. One of your commenters, Marleen, sort of explained how it works in the Netherlands, where the Dutch utility companies are rarely fully owned by the government anymore. But still often they form sort of a hybrid model. So this model is partly public and partly operating independently. And in this case of public utilities, she said it works for energy and water companies that are publicly owned, but they operate efficiently and they are still market oriented. And that such a model would also fit in the case of compute. So public guarantees, but private efficiency. What are your thoughts on this?
Speaker #2
I very much agree. I think... Why I always look to those comparisons, and I really appreciate that comment as well, is I'm just trying to say this is not a new problem. If we just treat compute as a utility, which I think we can at this point all agree that it is, it's critical to society, it's critical to the functioning of our economy, it's even probably critical to our national security at this point. So if we just say it is a utility, then we can apply everything we've learned from energy and water. And there's a lot of hybrid models, partly state-owned, completely privately owned. We have all the regulations already developed. We know how to do it. Let's just apply our lessons learned to compute. It is really just, it's not so different. But I think a lot of people want us to believe that it's somehow different. It really is not.
Speaker #1
And finally, something you also already referred to. but today many Tech hyperscalers such as Microsoft and Google are preparing to use nuclear energy to power their AI data centers. And they partner with public utilities to fund and operate the plants. And since the tech industry operates much faster than the power industry, which is usually government run, and you could also say slower to react. What are the growing challenges here?
Speaker #2
I think it's interesting that companies... like I said, that have twice as much revenue as the largest utility in France, which owns how many nuclear power plants? I don't know from my head, but I think it's more than 40 or 50. These companies need to lean on the state or lean on the utilities to get their job done. I think that's a bit ironic, in my opinion. I think the biggest risk and We already see that with Musk's supercomputer that he built is that until the nuclear is available, which is questionable if it should be available to begin with, most of these companies will fall back to using gas generators. Because that is the cheapest, fastest way to get more energy very quickly. And we see that Musk is powering his, I think, 50 or 60 megawatt data center completely with gas generators. We've seen similar trends in Ireland and it's very easy to point out that probably, especially given that most other sectors are decarbonizing, this would mean that the compute space is basically carbonizing its energy supply. I think what's even more interesting is that especially the big tech companies were the first to really run the narrative of renewable energy. They're the largest buyers in the world, but it's pretty clear now that it's not enough. And if it's not enough, they are willing to go either nuclear or gas to just fill their needs right now. Because the money they can make, I think, is much greater than the reputational risk of rolling back their sustainability plans. And I think that's a bit disappointing. It's maybe the simplest word.
Speaker #0
amazing wow you were jackass no i'm definitely disappointing i agree well thank you very much for talking to me today and see you next time thank you zana for the great questions

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Embed

You may also like

Description

Link to Max's LinkedIn post: https://ltmv.io/max-post

Sources: CNN Business

Paradigms is a podcast produced by Leitmotiv – an independent think tank focused on operationalizing values in our digital future.

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Transcription

Speaker #0
This is the next chap.
Speaker #1
Welcome to Paradigms, a podcast by Lightmotive, the independent digital policy think tank that exists to operationalize values in our digital future.
Speaker #2
In each episode, we unpack the latest developments and debates in our digital world. Together, we try to make sense of what is happening to our digital future. who's shaping it and what we can do to change its path.
Speaker #1
My name is Susanna Ludwikova and I'm here to ask the typical questions.
Speaker #2
And my name is Max Schulze and I'm here to give simple answers. At Labmotiv, I think about the policies that shape our digital future and how to operationalize our societal values within them.
Speaker #1
And my job is to bring our thinking to your ears, eyes and minds. In our podcast, we invite you to think with us as we unpack the building blocks of our digital world and work towards a better future that amplifies the best of our core values and our humanity. Today, I would like to talk about what you wrote recently. And it's not too long of a piece. So... I'll be quick. In your short opinion piece, this one you posted two weeks ago on your LinkedIn profile. And it's the one where you compare AI computes to a public utility. And you point out how artificial intelligence is essentially embedded into governmental services and businesses everywhere. And then you also ask the question, aren't the providers of compute infrastructure public utilities? And I believe what you're highlighting here, you can correct me if I'm wrong, is the growing... ubiquity of AI in our everyday lives and AI growing into a whole infrastructure as well as the concerns about lacking regulation and while we put aside the obvious yet still true claims that there is no such thing as free only the air you breathe I believe there's more nuance to it and maybe more ways we can think about what's fair and useful and especially when in compute power equals basically performance and performance equals profit. So, Max, in your words, could you explain and talk a little bit about why today we should start considering AI computes should become a public utility? And maybe to even first explain, what does AI compute even consist of?
Speaker #2
Yeah, I think it's a complex topic that I would try to unpack a bit to make it simple. I think my, who knows, for people who know me well, they know my favorite analogies, of course, of electricity. and When we think about the last industrial revolution, what was the key ingredient to that was energy in some shape or form. And a lot of that energy came to the industry through electricity, for example, with steam generators. And so for that industrial revolution, the key ingredient was access to electricity and energy in some form. As that evolved, electricity... Good networks were everywhere and they also became part of all of our lives. So, of course, everybody now has electricity at home. It's really embedded into our lives. And it's, I think, realistic to say that nothing would work in our society today without electricity. And so in that sense, electricity relatively quickly became a utility. Utility being defined as something that is critical to society and the absence of it. would have devastating effects on society. The same is true for water. Electricity is a good one always. And there's other utilities, of course. And because it's so critical to society, it can be considered a critical infrastructure for us. And in that case, of course, we want a lot of safeguards in place and regulation to make sure it's always available. And that's what a public utility to a certain degree is. It's a state-authorized monopoly, often owned also by the state, with very, very heavy regulation that makes sure that at any given time we have electricity. And now if, and I think we're still in the process of that, if we add AI functionalities to the products we use in our daily lives, one could say that's maybe not so critical, but if we add it also to every product, every business, in our society as well as government services. even maybe to a certain degree, adding AI into the energy system and electricity system or water utilities. So gluing it onto everything, one could argue that the ingredient of that AI functionality, which is computation, is as critical to us as electricity is to us. And so to make compute not such a complicated concept, I would just consider that the fuel of AI functionality that we use in all our products and our society. I hope that makes sense. First check to users at that.
Speaker #1
Sure. But can you maybe give one example of AI compute just so that we can imagine it?
Speaker #2
Yeah, I think the most simplistic form is there are ways, of course, now to run large language models. and other models on your local computer. There's open source models that you can run on your laptop. And then the laptop essentially turns electricity or for either from the battery or from the grid into computation that that model can use. And that computation is nowadays often coming from what are called GPUs, which are graphical processing units, which used to be primarily used for gaming and rendering of 3D. animations and things like this. And these specialized chips turn electricity into very high performing mathematical computing operations that are needed to power the models that we now consider AI.
Speaker #1
Got it. And with chips, are we also talking about companies such as Nvidia?
Speaker #2
Yes, of course. They are very prominent, though there's alternatives as well and a lot of standard computing chips like from intel or from apple's own m series that is made by arm they contain gpus on inside the the main chip um so when i say you can run models locally it's because you already have non-nvidia gpus also in in most laptops for example these days right and maybe then tying it to the utility um question so First of all, is it a utility? I think so, especially if we want to deploy AI everywhere. I think then it has to become a utility in that sense. Now, the real problem that I have and why I keep raising this question is that, of course, the companies providing that compute at a larger scale, so not your laptop, but cloud infrastructure companies and IT infrastructure providers, these companies then play de facto So... the role of the public utility producing the compute for us. And having that production capacity in private hands may pose a risk. To stick with my electricity analogy, we actually were in this situation before because a lot of energy companies in Europe used to be privately owned. And at their peak, there's example of a German energy company that was... At some point, one of the largest in the world called E.ON, at their peak, they made 150 billion euros a year revenue. And if a private company gets to that level of revenue and profitability and things like this, it's very, very hard to regulate them. And it's very hard to put brakes on them because, of course, they can deploy vast amounts of money to lobbying and blocking regulation. employ so many people that they can very simply threaten the government to let go of people, which would have bad outcomes for politics. And so they become so powerful that it's almost impossible to regulate them, which is why in Europe we had something called the market civilization, where we basically forced these companies to split up into smaller chunks. you could say. So they were still allowed to produce the electricity. You could say they were still allowed to produce compute. But for example, they had to trade, they had to sell that compute, not directly to customers anymore, but they had to, or the electricity in that case, but to go through to an open marketplace or an exchange to be precise. And that European Energy Exchange, it's called, there's many of them actually. They allow a basically very transparent pricing and trading of electricity. which in turn brings prices down. But also it makes sure that, first of all, we have full transparency how much electricity there is in the system. And we can see if we don't have enough, then the prices go up. We have very clear signals. And, and that's probably the most important thing, anyone can trade on that exchange, which also means that anyone who has a generator to make electricity at all. for example, a solar panel on the roof, can now participate as well. So you don't need to be a gigantic company to be able to be in this market. Anyone can be trading in that market. And that is very good for utility because it makes sure that we always have enough of it. The production of it is balanced across many different actors and the market is open and transparent so that anyone can... participate and it can be regulated.
Speaker #0
The world gets bigger.
Speaker #2
Yeah, to put that maybe a bit in perspective, the current, let's call them compute utilities, cloud infrastructure companies and IT infrastructure companies. I said already before, at its peak, E.ON made 150 billion euros of revenue per year. Today, we still have a company in Europe. EDF in France, which is state-owned, which still makes about 150 billion euros a year revenue owned by the French state. And these were considered so big that they were already hard to regulate. And to compare that to the current compute utilities, just one company alone, Microsoft, makes 250 billion euros revenue per year. So these compute utilities are already orders of magnitude bigger than what we had in the energy system, which should really trigger the alarm bell for us to say, well, now is the moment to apply the same methodology we applied to electricity to ensure an open market and transparency and the ability to regulate and ensure that all citizens have access to the compute they might need for their AI functionalities. We should apply the same framework we applied for energy. to compute.
Speaker #1
Right. I think this very well connects to the event that you mentioned also in the post. You participated in a political AI and tech summit in Brussels. And I think there you mentioned that a leading economist from the think tank encouraged private investment into the compute infrastructure rather than state subsidies. So in the post, you mentioned that you agree with this, that you prefer privately financed compute infrastructure. over state subsidies and why so?
Speaker #2
Well, I think that state subsidies can often lead to a distortion of the market. So, I mean, the plan is basically to say we'll hand out a lot of subsidies to European compute utilities and hope that they will then be more competitive or that they can overbuild. capacity, I think it will lead to only one thing is that we'll end up with way too much compute and way too many data centers because the subsidies would be pushed into a market that is not transparent. So we don't know how much compute we actually need. We don't know how much supply there really is right now. We don't know if we need more of it right now. And we are just subsidizing for the sake of growing something. So I would rather, I prefer private investors because private investors in that sense can determine if it makes sense to invest, if there is demand, if it's sensible to do so now. So because they don't, of course, want to lose the money that they're investing. So they're a little bit more cautious and a little bit more driven by market forces. However, I want to be really careful here because in the current compute utility system, that's not really feasible to invest. There's many reasons for this. We wrote a few papers on this because the market, again, there's no exchange. So if I now decide, oh, I want to become a compute utility, I want to invest in data centers and equipment to provide compute to customers. I basically create that capacity and they can't trade it with anyone because most of the customers are locked in to the utilities of the existing providers. So even though I might want to, as a private investor, invest and participate in this market, I just can't. So another reason why I'm saying no state subsidies is that I would rather see the state fix the market. So applying the same framework of market liberalization that we did in the energy system, create that exchange so that we have transparency. How much compute do we need? Is it too expensive right now? Do we need to invest in supply? And then after the transparency is given, private investors together maybe with the state can then invest into more infrastructure, into creating more compute. But even in that case, I would prefer the state to give out loans rather than subsidies because the loans could potentially earn interest. potentially be paid back at some point. So the taxpayer is not just handing out money, because that's at the end of the day, subsidies come from our pockets, handing out money for something we don't know if we need it. But with the loans, we say, well, if an investor thinks that it's worth building Compute now, and we have market signals and an exchange that tells us, yes, now is a good time, then we can facilitate that investment with a loan. which we then get back after five to 10 years. That's something I'm very much in favor of. But just blank check subsidies into infrastructure that we don't know we need, I think is very difficult.
Speaker #1
But if a private investment is a go for you, you've mentioned that in the post, you mentioned that Amazon sort of immediately offered to lead this investment in the discussion. And you labeled this a low risk investment, or at least. one with a positive expected return. So could you explain this a little bit more and also the tensions that emerge here?
Speaker #2
No, of course. I think what I see problematic here is that on the one side, the economist calls for private investment, which again, I agree with. But on the flip side, only very few private investors can invest in this market. And those are the ones who already have a chokehold on the market. And those are often the big cloud providers. They have locked in customers. They have the world's largest sales forces to sell more compute to customers. They even release the models, the AI models. They push us to use them. So they basically can create their own demand. So it makes sense for them to basically say, yeah, we can invest more. We can always invest more because we know we can create more demand. and that's a little bit that that's basically like private investment from a handful of people who already know that they own the market and that's to me a bit um we have to be very very careful with that because it's it's a little bit the analogy that is probably the most practical here is basically basically somebody who who already has electricity contracts with every company in europe um saying that yeah of course would be happy to build more power plants because again they already have all the contracts so for them it makes sense and at the same time then asking government to basically make sure that these companies use more electricity which is the equivalent of them asking governments to push companies to use more ai so that's really problematic especially given that these companies are now so big and so powerful that we can barely regulate them So instead of letting three, four or five companies invest even more and get even bigger in the space, we should make sure that more private investors can access this market. But that really requires us to completely change how the compute market works today.
Speaker #1
Yeah, and also as an opposing alternative or as a possible solution to this. One of your commenters, Marleen, sort of explained how it works in the Netherlands, where the Dutch utility companies are rarely fully owned by the government anymore. But still often they form sort of a hybrid model. So this model is partly public and partly operating independently. And in this case of public utilities, she said it works for energy and water companies that are publicly owned, but they operate efficiently and they are still market oriented. And that such a model would also fit in the case of compute. So public guarantees, but private efficiency. What are your thoughts on this?
Speaker #2
I very much agree. I think... Why I always look to those comparisons, and I really appreciate that comment as well, is I'm just trying to say this is not a new problem. If we just treat compute as a utility, which I think we can at this point all agree that it is, it's critical to society, it's critical to the functioning of our economy, it's even probably critical to our national security at this point. So if we just say it is a utility, then we can apply everything we've learned from energy and water. And there's a lot of hybrid models, partly state-owned, completely privately owned. We have all the regulations already developed. We know how to do it. Let's just apply our lessons learned to compute. It is really just, it's not so different. But I think a lot of people want us to believe that it's somehow different. It really is not.
Speaker #1
And finally, something you also already referred to. but today many Tech hyperscalers such as Microsoft and Google are preparing to use nuclear energy to power their AI data centers. And they partner with public utilities to fund and operate the plants. And since the tech industry operates much faster than the power industry, which is usually government run, and you could also say slower to react. What are the growing challenges here?
Speaker #2
I think it's interesting that companies... like I said, that have twice as much revenue as the largest utility in France, which owns how many nuclear power plants? I don't know from my head, but I think it's more than 40 or 50. These companies need to lean on the state or lean on the utilities to get their job done. I think that's a bit ironic, in my opinion. I think the biggest risk and We already see that with Musk's supercomputer that he built is that until the nuclear is available, which is questionable if it should be available to begin with, most of these companies will fall back to using gas generators. Because that is the cheapest, fastest way to get more energy very quickly. And we see that Musk is powering his, I think, 50 or 60 megawatt data center completely with gas generators. We've seen similar trends in Ireland and it's very easy to point out that probably, especially given that most other sectors are decarbonizing, this would mean that the compute space is basically carbonizing its energy supply. I think what's even more interesting is that especially the big tech companies were the first to really run the narrative of renewable energy. They're the largest buyers in the world, but it's pretty clear now that it's not enough. And if it's not enough, they are willing to go either nuclear or gas to just fill their needs right now. Because the money they can make, I think, is much greater than the reputational risk of rolling back their sustainability plans. And I think that's a bit disappointing. It's maybe the simplest word.
Speaker #0
amazing wow you were jackass no i'm definitely disappointing i agree well thank you very much for talking to me today and see you next time thank you zana for the great questions

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Description

Link to Max's LinkedIn post: https://ltmv.io/max-post

Sources: CNN Business

Paradigms is a podcast produced by Leitmotiv – an independent think tank focused on operationalizing values in our digital future.

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Transcription

Speaker #0
This is the next chap.
Speaker #1
Welcome to Paradigms, a podcast by Lightmotive, the independent digital policy think tank that exists to operationalize values in our digital future.
Speaker #2
In each episode, we unpack the latest developments and debates in our digital world. Together, we try to make sense of what is happening to our digital future. who's shaping it and what we can do to change its path.
Speaker #1
My name is Susanna Ludwikova and I'm here to ask the typical questions.
Speaker #2
And my name is Max Schulze and I'm here to give simple answers. At Labmotiv, I think about the policies that shape our digital future and how to operationalize our societal values within them.
Speaker #1
And my job is to bring our thinking to your ears, eyes and minds. In our podcast, we invite you to think with us as we unpack the building blocks of our digital world and work towards a better future that amplifies the best of our core values and our humanity. Today, I would like to talk about what you wrote recently. And it's not too long of a piece. So... I'll be quick. In your short opinion piece, this one you posted two weeks ago on your LinkedIn profile. And it's the one where you compare AI computes to a public utility. And you point out how artificial intelligence is essentially embedded into governmental services and businesses everywhere. And then you also ask the question, aren't the providers of compute infrastructure public utilities? And I believe what you're highlighting here, you can correct me if I'm wrong, is the growing... ubiquity of AI in our everyday lives and AI growing into a whole infrastructure as well as the concerns about lacking regulation and while we put aside the obvious yet still true claims that there is no such thing as free only the air you breathe I believe there's more nuance to it and maybe more ways we can think about what's fair and useful and especially when in compute power equals basically performance and performance equals profit. So, Max, in your words, could you explain and talk a little bit about why today we should start considering AI computes should become a public utility? And maybe to even first explain, what does AI compute even consist of?
Speaker #2
Yeah, I think it's a complex topic that I would try to unpack a bit to make it simple. I think my, who knows, for people who know me well, they know my favorite analogies, of course, of electricity. and When we think about the last industrial revolution, what was the key ingredient to that was energy in some shape or form. And a lot of that energy came to the industry through electricity, for example, with steam generators. And so for that industrial revolution, the key ingredient was access to electricity and energy in some form. As that evolved, electricity... Good networks were everywhere and they also became part of all of our lives. So, of course, everybody now has electricity at home. It's really embedded into our lives. And it's, I think, realistic to say that nothing would work in our society today without electricity. And so in that sense, electricity relatively quickly became a utility. Utility being defined as something that is critical to society and the absence of it. would have devastating effects on society. The same is true for water. Electricity is a good one always. And there's other utilities, of course. And because it's so critical to society, it can be considered a critical infrastructure for us. And in that case, of course, we want a lot of safeguards in place and regulation to make sure it's always available. And that's what a public utility to a certain degree is. It's a state-authorized monopoly, often owned also by the state, with very, very heavy regulation that makes sure that at any given time we have electricity. And now if, and I think we're still in the process of that, if we add AI functionalities to the products we use in our daily lives, one could say that's maybe not so critical, but if we add it also to every product, every business, in our society as well as government services. even maybe to a certain degree, adding AI into the energy system and electricity system or water utilities. So gluing it onto everything, one could argue that the ingredient of that AI functionality, which is computation, is as critical to us as electricity is to us. And so to make compute not such a complicated concept, I would just consider that the fuel of AI functionality that we use in all our products and our society. I hope that makes sense. First check to users at that.
Speaker #1
Sure. But can you maybe give one example of AI compute just so that we can imagine it?
Speaker #2
Yeah, I think the most simplistic form is there are ways, of course, now to run large language models. and other models on your local computer. There's open source models that you can run on your laptop. And then the laptop essentially turns electricity or for either from the battery or from the grid into computation that that model can use. And that computation is nowadays often coming from what are called GPUs, which are graphical processing units, which used to be primarily used for gaming and rendering of 3D. animations and things like this. And these specialized chips turn electricity into very high performing mathematical computing operations that are needed to power the models that we now consider AI.
Speaker #1
Got it. And with chips, are we also talking about companies such as Nvidia?
Speaker #2
Yes, of course. They are very prominent, though there's alternatives as well and a lot of standard computing chips like from intel or from apple's own m series that is made by arm they contain gpus on inside the the main chip um so when i say you can run models locally it's because you already have non-nvidia gpus also in in most laptops for example these days right and maybe then tying it to the utility um question so First of all, is it a utility? I think so, especially if we want to deploy AI everywhere. I think then it has to become a utility in that sense. Now, the real problem that I have and why I keep raising this question is that, of course, the companies providing that compute at a larger scale, so not your laptop, but cloud infrastructure companies and IT infrastructure providers, these companies then play de facto So... the role of the public utility producing the compute for us. And having that production capacity in private hands may pose a risk. To stick with my electricity analogy, we actually were in this situation before because a lot of energy companies in Europe used to be privately owned. And at their peak, there's example of a German energy company that was... At some point, one of the largest in the world called E.ON, at their peak, they made 150 billion euros a year revenue. And if a private company gets to that level of revenue and profitability and things like this, it's very, very hard to regulate them. And it's very hard to put brakes on them because, of course, they can deploy vast amounts of money to lobbying and blocking regulation. employ so many people that they can very simply threaten the government to let go of people, which would have bad outcomes for politics. And so they become so powerful that it's almost impossible to regulate them, which is why in Europe we had something called the market civilization, where we basically forced these companies to split up into smaller chunks. you could say. So they were still allowed to produce the electricity. You could say they were still allowed to produce compute. But for example, they had to trade, they had to sell that compute, not directly to customers anymore, but they had to, or the electricity in that case, but to go through to an open marketplace or an exchange to be precise. And that European Energy Exchange, it's called, there's many of them actually. They allow a basically very transparent pricing and trading of electricity. which in turn brings prices down. But also it makes sure that, first of all, we have full transparency how much electricity there is in the system. And we can see if we don't have enough, then the prices go up. We have very clear signals. And, and that's probably the most important thing, anyone can trade on that exchange, which also means that anyone who has a generator to make electricity at all. for example, a solar panel on the roof, can now participate as well. So you don't need to be a gigantic company to be able to be in this market. Anyone can be trading in that market. And that is very good for utility because it makes sure that we always have enough of it. The production of it is balanced across many different actors and the market is open and transparent so that anyone can... participate and it can be regulated.
Speaker #0
The world gets bigger.
Speaker #2
Yeah, to put that maybe a bit in perspective, the current, let's call them compute utilities, cloud infrastructure companies and IT infrastructure companies. I said already before, at its peak, E.ON made 150 billion euros of revenue per year. Today, we still have a company in Europe. EDF in France, which is state-owned, which still makes about 150 billion euros a year revenue owned by the French state. And these were considered so big that they were already hard to regulate. And to compare that to the current compute utilities, just one company alone, Microsoft, makes 250 billion euros revenue per year. So these compute utilities are already orders of magnitude bigger than what we had in the energy system, which should really trigger the alarm bell for us to say, well, now is the moment to apply the same methodology we applied to electricity to ensure an open market and transparency and the ability to regulate and ensure that all citizens have access to the compute they might need for their AI functionalities. We should apply the same framework we applied for energy. to compute.
Speaker #1
Right. I think this very well connects to the event that you mentioned also in the post. You participated in a political AI and tech summit in Brussels. And I think there you mentioned that a leading economist from the think tank encouraged private investment into the compute infrastructure rather than state subsidies. So in the post, you mentioned that you agree with this, that you prefer privately financed compute infrastructure. over state subsidies and why so?
Speaker #2
Well, I think that state subsidies can often lead to a distortion of the market. So, I mean, the plan is basically to say we'll hand out a lot of subsidies to European compute utilities and hope that they will then be more competitive or that they can overbuild. capacity, I think it will lead to only one thing is that we'll end up with way too much compute and way too many data centers because the subsidies would be pushed into a market that is not transparent. So we don't know how much compute we actually need. We don't know how much supply there really is right now. We don't know if we need more of it right now. And we are just subsidizing for the sake of growing something. So I would rather, I prefer private investors because private investors in that sense can determine if it makes sense to invest, if there is demand, if it's sensible to do so now. So because they don't, of course, want to lose the money that they're investing. So they're a little bit more cautious and a little bit more driven by market forces. However, I want to be really careful here because in the current compute utility system, that's not really feasible to invest. There's many reasons for this. We wrote a few papers on this because the market, again, there's no exchange. So if I now decide, oh, I want to become a compute utility, I want to invest in data centers and equipment to provide compute to customers. I basically create that capacity and they can't trade it with anyone because most of the customers are locked in to the utilities of the existing providers. So even though I might want to, as a private investor, invest and participate in this market, I just can't. So another reason why I'm saying no state subsidies is that I would rather see the state fix the market. So applying the same framework of market liberalization that we did in the energy system, create that exchange so that we have transparency. How much compute do we need? Is it too expensive right now? Do we need to invest in supply? And then after the transparency is given, private investors together maybe with the state can then invest into more infrastructure, into creating more compute. But even in that case, I would prefer the state to give out loans rather than subsidies because the loans could potentially earn interest. potentially be paid back at some point. So the taxpayer is not just handing out money, because that's at the end of the day, subsidies come from our pockets, handing out money for something we don't know if we need it. But with the loans, we say, well, if an investor thinks that it's worth building Compute now, and we have market signals and an exchange that tells us, yes, now is a good time, then we can facilitate that investment with a loan. which we then get back after five to 10 years. That's something I'm very much in favor of. But just blank check subsidies into infrastructure that we don't know we need, I think is very difficult.
Speaker #1
But if a private investment is a go for you, you've mentioned that in the post, you mentioned that Amazon sort of immediately offered to lead this investment in the discussion. And you labeled this a low risk investment, or at least. one with a positive expected return. So could you explain this a little bit more and also the tensions that emerge here?
Speaker #2
No, of course. I think what I see problematic here is that on the one side, the economist calls for private investment, which again, I agree with. But on the flip side, only very few private investors can invest in this market. And those are the ones who already have a chokehold on the market. And those are often the big cloud providers. They have locked in customers. They have the world's largest sales forces to sell more compute to customers. They even release the models, the AI models. They push us to use them. So they basically can create their own demand. So it makes sense for them to basically say, yeah, we can invest more. We can always invest more because we know we can create more demand. and that's a little bit that that's basically like private investment from a handful of people who already know that they own the market and that's to me a bit um we have to be very very careful with that because it's it's a little bit the analogy that is probably the most practical here is basically basically somebody who who already has electricity contracts with every company in europe um saying that yeah of course would be happy to build more power plants because again they already have all the contracts so for them it makes sense and at the same time then asking government to basically make sure that these companies use more electricity which is the equivalent of them asking governments to push companies to use more ai so that's really problematic especially given that these companies are now so big and so powerful that we can barely regulate them So instead of letting three, four or five companies invest even more and get even bigger in the space, we should make sure that more private investors can access this market. But that really requires us to completely change how the compute market works today.
Speaker #1
Yeah, and also as an opposing alternative or as a possible solution to this. One of your commenters, Marleen, sort of explained how it works in the Netherlands, where the Dutch utility companies are rarely fully owned by the government anymore. But still often they form sort of a hybrid model. So this model is partly public and partly operating independently. And in this case of public utilities, she said it works for energy and water companies that are publicly owned, but they operate efficiently and they are still market oriented. And that such a model would also fit in the case of compute. So public guarantees, but private efficiency. What are your thoughts on this?
Speaker #2
I very much agree. I think... Why I always look to those comparisons, and I really appreciate that comment as well, is I'm just trying to say this is not a new problem. If we just treat compute as a utility, which I think we can at this point all agree that it is, it's critical to society, it's critical to the functioning of our economy, it's even probably critical to our national security at this point. So if we just say it is a utility, then we can apply everything we've learned from energy and water. And there's a lot of hybrid models, partly state-owned, completely privately owned. We have all the regulations already developed. We know how to do it. Let's just apply our lessons learned to compute. It is really just, it's not so different. But I think a lot of people want us to believe that it's somehow different. It really is not.
Speaker #1
And finally, something you also already referred to. but today many Tech hyperscalers such as Microsoft and Google are preparing to use nuclear energy to power their AI data centers. And they partner with public utilities to fund and operate the plants. And since the tech industry operates much faster than the power industry, which is usually government run, and you could also say slower to react. What are the growing challenges here?
Speaker #2
I think it's interesting that companies... like I said, that have twice as much revenue as the largest utility in France, which owns how many nuclear power plants? I don't know from my head, but I think it's more than 40 or 50. These companies need to lean on the state or lean on the utilities to get their job done. I think that's a bit ironic, in my opinion. I think the biggest risk and We already see that with Musk's supercomputer that he built is that until the nuclear is available, which is questionable if it should be available to begin with, most of these companies will fall back to using gas generators. Because that is the cheapest, fastest way to get more energy very quickly. And we see that Musk is powering his, I think, 50 or 60 megawatt data center completely with gas generators. We've seen similar trends in Ireland and it's very easy to point out that probably, especially given that most other sectors are decarbonizing, this would mean that the compute space is basically carbonizing its energy supply. I think what's even more interesting is that especially the big tech companies were the first to really run the narrative of renewable energy. They're the largest buyers in the world, but it's pretty clear now that it's not enough. And if it's not enough, they are willing to go either nuclear or gas to just fill their needs right now. Because the money they can make, I think, is much greater than the reputational risk of rolling back their sustainability plans. And I think that's a bit disappointing. It's maybe the simplest word.
Speaker #0
amazing wow you were jackass no i'm definitely disappointing i agree well thank you very much for talking to me today and see you next time thank you zana for the great questions

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

About Paradigms

Apple Podcasts: https://ltmv.io/podcast-apple

Spotify: https://ltmv.io/podcast-spotify

YouTube: https://ltmv.io/podcast-yt

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Embed