PassW0rd – 12 February 2025 (The Cost of the Future)

Speaker A: This program is brought to you by Resonance FM. If you like what you hear, please support our work by making a donation at resonancefm.com/donate.

Speaker B: Hello, happy new year and welcome to our first password of 2025. The program brought to you by Resonance FM that explains the impact of the technology now underpinning our lives. In this program, data centers, their pivotal role in AI development, and what you need to know about their cost in terms of energy and the environment, and even more importantly, in geopolitical terms. This is the result of over a year and a half of investigation by Future Intelligence, the Password program’s parent organization., and it has revealed some stunning findings, including concerns that we will see the power needed by data centers in the US and the UK jump by between 4 and 6 times over the next 5 years, competition for water in particular areas, and very real concerns about the geographical location of our data. Last year, they were the great unknowns of the high-tech world. But this year, because of AI, data centers are at the top of the political agenda, with both US President Donald Trump and UK Prime Minister Sakhir Starmer laying out ambitious plans to build high-speed data networks. So far, already this year, the UK Premier has announced an investment of over £20 billion in creating a UK AI data center superhighway. Though the figures released by the UK government at its International Investment Summit last October point to even more money being earmarked for AI infrastructure, with Blackstone committing to a massive £10 billion AI data centre in Blyth in Northumberland and Amazon Web Services pledging £8 billion on AI development. In the US, Donald Trump has announced the staggering $500 billion Stargate project. While the Chinese have thrown an AI spanner in the works with the announcement of its Deepseek tool, which, with its alleged budget of £4.8 million, has thrown the economic model for AI up in the air by slashing development costs. And it hasn’t stopped there. On Sunday, the 10th of February, President Macron announced France would be spending €109 billion on data centers AI over the next 4 years. And 2 days later, at the end of the AI Action Summit in Paris, the EU’s Ursula von der Leyen announced the AI race is far from over in a speech committing further tens of billions of AI investment following the private €150 billion EU AI Champions Initiative announced the day before. And added that an Invest AI initiative would top up these investments by a further $50 billion, making it the largest public-private partnership in the world. So it’s incredibly appropriate that we devote this program to these data centers and developments that are essential to every social media message, every internet search, every YouTube broadcast, every Ernest podcast, every video call. In fact, everything we do online. And now, as we have found, each data centre will be the echo of these starter pistols firing in the AI arms race. Bland, anonymous, and boring, data centres are something we are happy to leave permanently in the background of our bright internet-lit lives, quietly humming away on the outskirts of our towns as we get on with the serious business of living on our phones and laptops.

Speaker C: Jobs.

Speaker B: Yet without those data centers, our lives would grind to a halt and leave us staring blankly at our empty screens. But what we should also know is that they also represent a competitive threat to us, not just because of AI, but because of their demands for water and energy that will be massively increased by AI usage. Here’s Professor Andrew Chien, the William Eckhart Distinguished Service Professor of Computer Science at the University of Chicago, a former vice president of research for the US chip giant Intel, and an expert on computer power use, on the issue.

Speaker D: As recently as 20 years ago, when we were at the turn of the millennium, computing was just a tiny, tiny part of electrical power consumption in the world. And it was that for a lot of reasons. One, because computing wasn’t used for everything like it is today, but also it was because we’d been riding this wonderful technology curve around Moore’s Law and Dennard scaling that had allowed us to have the technology become much more efficient. So even though we use more and more computing, the power use didn’t increase that much. That’s why at the turn of the millennium we had 10-megawatt data centers And today we’re talking about gigawatts, right? So the question would be, what changed? Well, two things. One is the exponential growth finally caught up with us. It got to be really large. And then the second thing is that we’ve seen a slowing of those technological advances of Moore’s Law. And Dennard scaling ended about 2005. So now it’s catching up and we’re starting to see data centers as really, really large power consumers.

Speaker B: And that power consumption, I was remotely attending a conference that was going on in America for, I think it was the National Academy of Sciences. They were saying that that power use is going to be incredible. I mean, you talk about Moore’s Law, people were saying that there’s going to be a sixfold increase in energy demands over the next 6 years.

Speaker D: Yeah, that’s right. The projections for, say, the United States in particular, are that the power consumption of data centers will grow from a maybe 2% to about 8 or 9% by 2030. So that’s 5 years away now. And that increase will be 80 gigawatts of average consumption. Interestingly, for various reasons, the United States or North America actually has about half of the world’s data centers. So you can think of the global number as being 2 times that, generally speaking. And to make it really sort of poignant and to the point, that’s pretty scary to begin with. But just think about what comes after that if that continues. It’s the next 5 years that will really make an impact.

Speaker B: So that is what, because of artificial intelligence, or shall we say large language models, because of the way that the amount of processing that they currently demand?

Speaker D: Yeah, it’s, it’s a, it’s a good, good question. So it’s popular in the press to focus on the newest thing and AI and large language models. There’s no question they’re a big driver of this. But it’s also true that the cloud itself had already been a big driver of this. And we would probably be in talking about data centers being a few percent, right? Maybe not 8%, but maybe 4% or something like that, 4 or 5% in 2030, even if we didn’t have this huge AI driver. But there’s no question AI is like, put this growth on steroids. And that’s because it’s a fundamentally different way of delivering services. Right? It’s made what were mostly information-intensive services on the internet, serving web search or some other kind of thing like that, IT services, into a compute-intensive service, which is what this large language model is. It’s like a giant sparse matrix computation. They do it every time you put 3 words into ChatGPT.

Speaker B: Which is one of the odd things, isn’t it? I mean, people aren’t aware, for instance, that every time they put 3 words into ChatGPT or a search engine, that a certain amount of water is needed to cool it.

Speaker D: Ah, well, there’s electricity as we’ve talked about, and there’s water. Water’s used in, in two ways in most data centers nowadays. For GPUs and the like, it’s popular to use water to do closed-loop cooling where you pump cold water over the GPUs and then warm water comes out and then you vent the heat and you recirculate the water. That’s not so bad. But the thing that’s bad is that in order to increase the energy efficiency of the data centers, Many data centers have been doing what’s called evaporative cooling, which is just what you do on a hot day in the summer when you put a cold rag on your head. You get the endothermic reaction of water evaporating to carry away the heat. So it turns out that’s a really energy-efficient way to remove heat from the data center. So these data centers literally spray water into the air in order to get rid of their heat. The problem with that is for environmental purposes, that effectively destroys the water from the environment. ’cause it goes off into the air and it flies away in the atmosphere. So people talk about that as water consumption, and that has become a great concern for data centers. It’s become very pressing, a concern over the last year. But there’s, there’s reason for hope. A number of the large hyperscalers have committed to having their future data centers not use water in that way. But not everyone has committed to do that. And even the most progressive companies that have committed to do that for their future data centers haven’t committed to retrofitting their old data centers, for which there’s quite a bit already, to stop use of this kind of evaporative technique. So that’s a big concern. Obviously, it’s more of a concern in places where water’s in short supply, in the desert, in the Southwest of the United States, for example. Perhaps where you’re sitting, it’s not so much of a concern. I don’t know, the Thames is a pretty large river. Where I sit in Chicago, we have the Great Lakes, which is the largest supply of fresh water, I think, in the world. But even here, people are concerned about water use.

Speaker B: Chicago University’s Professor Andrew Cheon, who we will be hearing more from later. The picture in the US has been terrifying scientists, as they made clear in a meeting last November at the National Academies of Science, Engineering, and Medicine. Where they pointed out that data centers in Virginia were now competing with the state’s inhabitants for clean water. Something that could very soon be occurring in the UK because of the concentration of data centers around London, particularly near the UK’s data center capital, Woking. It’s a concentration that Dr. Helen Monroe of the data infrastructure company Pulsant pointed out.

Speaker E: In a recent report that Tech UK published, there was a figure that cited around 80% of data centre capacity is centred in that Greater London region with SLA. We’ve got a few sites down there ourselves, but in terms of supporting kind of regions across different cities, you know, we’ve got quite a few different locations. One of the interesting aspects, as well as the requirement for latency to serve various different sort of smart applications and applications that might kind of demand a data centre that’s physically close to where that latency is being demanded, is how the data centre might interact with the local available energy. So one of the really interesting things that’s happening in the energy markets within the UK is that the different regions have different levels of emissions intensity. So for instance, Scotland, we’ve got lots of wind power, and quite often there’s— when the wind’s blowing, we can’t use all the electricity within that Scottish region, and we can’t transmit it elsewhere within the UK either. Constraints on that transmission network. So at that point in time, you’re effectively kind of— any extra power that you’re using is not having an impact in terms of demand for gas generation, whereas a lot of the regions in other parts of the UK still have the marginal demand that’s not being taken up by— that’s that’s not supported by renewable energy is being taken up by traditional gas generation mostly. So there’s some kind of really interesting dynamics between the different characteristics of power grids across the UK. And it does raise some interesting questions about if you’re planning an IT workload, should you think about that in terms of where you’re going to actually put it.

Speaker B: Well, also there’s another point too, isn’t there, which is it will probably go some way towards your carbon footprint, and it possibly could be something that you could say, yeah, I’m offsetting this because I’m using renewable energy.

Speaker E: Yeah, the way I see it, the renewable energy question is one side of the energy coin. I think there’s two sides of the coin. Firstly is what are data center organizations doing with their purchasing power? Are they doing all they can to buy energy in a good way? But that’s a separate question from what are they actually using, because, you know, data centers are all— I say in the UK, they’re all on the grid. So when data centers use energy, they’re demanding energy on that particular grid region and using that grid infrastructure. So if you think about the marginal impact of, say, increasing or reducing power consumption, that’s going to sort of increase and reduce demand on whatever generation source is providing the most flexible output. And typically that doesn’t want to be gas because Wind energy, unless there’s a surplus, we can’t really control how much of it there is. Same with solar. Nuclear is— it doesn’t flex up and down very responsibly. So, you know, the gas takes up the slack in a lot of the regions.

Speaker B: Which is one of the reasons why everybody’s suddenly started to become interested in hydrogen, because hydrogen could take up slack in a similar way, even though that capability isn’t yet there. Although I understand from people that I’ve been talking to, I went to Leipzig, which is meant to be the green hydrogen capital of Europe. There people are doing lots of research into this. I mean, one of the big problems in all of this is that people aren’t aware, are they, that there’s a cost in moving data.

Speaker E: What do you mean by a cost in moving data?

Speaker B: Well, obviously, if you’re processing locally, it’s not as expensive as if you’re going to move data to process it centrally. It was one of the issues that they were talking about in America was that if you’re going to move data, physically move data, obviously it means that you use energy and you use cycles, etc., etc. So if you’re going to ask a question in a I mean, I think that there was an interesting figure. I haven’t got it to my fingertips at the moment, but it was the cost in terms of water for a Google search. The amount of water that is used in the Google search is significant. So that’s what I mean.

Speaker E: The question of water is an interesting one. I don’t have answers about that. What I have seen on a lot of the analysis is that the water consumption figures that have been looked at include both water that might be used in a data center’s cooling system and water that’s used in cooling electricity generation. So, you know, a gas-fired power plant, for example. And what we’re seeing with it— I know that the environment agencies do anything but of work around this is trying to— some data centers use water as part of their cooling systems, some data centers don’t. It’s not quite clear how many data centers within the UK actually do go through a lot of water, but it’s something that operators are sort of moving towards a bit more of a consciousness over, and there’s quite a few data center operators that are starting to disclose a bit more information about their water use and, uh, and the metric of efficiency.

Speaker B: Dr. Helen Monroe of the data infrastructure company Pulsant, underlining the need for cooling water in data centers and the fact that AI will mean that it will compete for water in urban areas with people. It’s a competition that makes it tempting to site data centers in remote areas, something that will push data centres into conflict with wildlife and the environment. And building data centres in a particular area also doesn’t generate income in that area, which would make them more attractive. While trying to push the problem overseas is also not a solution, as Chicago University’s Professor Xian, a speaker at the National Academies meeting in November on the issue, points out.

Speaker D: I think that data centers are in competition with people and of course with wildlife, you know, with the environment and biodiversity and all those things that are concerns. I think people do good studies. Freshwater supply and freshwater stress is actually pretty widespread in the world these days. And you don’t even need to go into how climate change might be making that more difficult. But say, for example, in the western United States and in the Southwest, I think anything, any industrial use that goes in and consumes water is in competition with wildlife and people and so on. There are whole nations like in India where much of the country is in water stress all of the time. So I’m sure they’re looking at that, looking at it with the view that fundamentally, how do we develop this information technology economy that we have to have and then deal with these challenges? You know, China is something we often talk about and China, has huge water stress in many parts of the Chinese geographic area, right? The water is not well distributed, as they say, right? So I think it’s a common problem. And I’m sure, you know, there’s parts of Europe that have the same issues. I’m just not as familiar.

Speaker B: It’s evident, isn’t it? I mean, all you have to do is watch a wildlife program about Africa and see, or India, and see that they have a wet and a dry season. So obviously water is going to be in shortage in certain places at certain times of the year, which therefore will create competition. However, One of the things that people were talking about at the National Academy conference was that Virginia has issues of water supply because of the amount of data centers in Virginia.

Speaker D: Yeah, actually, so two things worth calling out. One is in the power grid in Virginia, the Dominion Energy area mostly, the data centers have now grown up to be almost 25% of the power grid.

Speaker B: Added.

Speaker D: That’s an amazingly large number. And if you take 25% and you compound it with 20% growth a year, you know, it doesn’t take you that long to get to 50. So those are amazing numbers. But, you know, the concern of the, the local Virginia community points out the following problem with data centers. So you put a data center in a community, the benefits of the data center don’t accrue to that community. It’s part of this global internet. It’s serving this generative AI request, who knows? I mean, people, yeah, people in Virginia, but people in California and people in Canada and people in the UK, they’re using these services. So there’s this misalignment with the environmental costs, which of course are local, and the attributed benefits. And, and that creates a bunch of interesting tensions. And of course, the local community’s view is we’re not getting the benefits and we’re suffering the costs., and you, this big company, are making off with the profits. So how do we balance this ledger, if you will, so that everyone gets at least what they deserve, or at least something?

Speaker B: We made a program about energy just recently, and one of the things that people were talking about was possibly using wave power. There was some discussion about using tidal power, although apparently tidal power can only be used in certain places where it’s actually going to be— you’re able to harness the power of the tide. That said, then, that would almost suggest that you would say, okay, well, where we can harness the power of the tide, for example, why don’t we site loads and loads of data centers there? Because there we’ve got sustainable energy. This was again another point that came up in the conference where people were saying, why are all of these data centers in Virginia? Why can’t we put them in Texas?

Speaker D: That’s a great question. And you probably know from my personal research, we’ve been trying to study where data centers should be placed and how they can take advantage of excess renewable energy. So that’s not exactly the requirement that you put the data centers next to the renewable energy. But if you think about transmission and congestion, where you will quickly conclude that, well, you’d like to be closer, and maybe even the ultimate closest, which is right on top of it. So I think you’re starting to see that, because what happens is when you have a lot of renewable energy in locations, the power prices costs tend to be low and there can be a lot of supply. So you’ve been hearing about for decades people building data centers in the Northwest of the United States, in Washington State, in Oregon, where they have these massive hydroelectric dams, right? Old-fashioned renewables, if you will. And you’re seeing it in places where, you know, Microsoft has built data centers out in Iowa. China is building them in Gansu, right? You know, where they have massive renewables. So you are starting to see some of that. So the question is, why aren’t they all in places like that? Well, it turns out that the internet actually does care about location a little bit. And the way it cares about location is that there’s latency in the time of flight of, well, photons in fibers, but information across the world is limited by some fraction of the speed of light. So if you have to access from London, the data centers in Virginia, then it’s going to be some time of flight. If it’s Japan, you know, it’s another distance and so on. So what happened over time is that people have built data centers in clusters. And those clusters are centered around large bundles of fibers that, say, transit the Atlantic. In the case of the United States, that’s very close to a lot of government infrastructure that’s super important and so on. So as you get these clusters like in Northern Virginia, as we were talking about, then there becomes a critical mass effect. Connect. So it’s not just you and me accessing that data center. Now more and more of the Internet is data centers talking to each other. So, you know, your virtual machine in Amazon talking to somebody else’s virtual machine in Google talking to somebody else’s virtual machine in Microsoft. And then maybe they’re talking to Facebook. And let’s see if I can touch on all of the players here. So there’s these synergies about being close by. And by some estimates I’ve seen that 50% of the world’s Internet traffic flows around in Northern Virginia. And the way I like to explain this, to just crystallize it, is you might think of the internet as not having a center of mass. But if it has a center of mass, it’s in Northern Virginia. So that kind of thinking is actually what drives this pressure to put all of that stuff in particular places.

Speaker B: But there’s another issue too, isn’t there? And again, it was one that was touched on in the conference, which is something that, again, people don’t think about. But it costs money to move data. It costs energy to move data. And people just because of, I suppose, our lack of familiarity with the way that this technology has evolved, we don’t think about that.

Speaker D: Yeah, that’s right. And one of the interesting projections that was done a number of years ago, sort of, I think, right around 2019 or 2020, looked at how data center power would grow and how Internet, that is the actual networking power, would grow. And it’s kind of amazing, actually, at that time they were projecting that the actual transmission of information through these wide area links and so on, and the super fast 5G wireless that we all live on and so on these days, would actually be comparable to the data centers. Now with AI, generative AI, I think it’s probably not quite going to be comparable, but it could be half or something like that. So there’s a gigantic amount of energy. And the thing that’s hard for us to understand is because the energy per bit is so small, right, is that even though the technologies are quite efficient, the quantities of data that are being transmitted are staggering. Like, you know, we’re doing a video call, right? This is a huge amount of data with a gigantic amount of resolution. And everyone does this every day for free in the world. So now you got maybe a billion people doing this kind of stuff. So they’re moving a huge amount of data. They’re streaming videos, right? They’re watching movies. So there’s a very, very large amount of data being moved, gigabytes, you know, maybe even terabytes per person per day. And that’s what accounts for the very, very large amount of energy that the networks are using.

Speaker B: So as Professor Xian says, at the moment, we really have not thought about the closely associated issues of AI and electrical power. Essentially, it’s obvious, and it should be noted that in last October’s UK Investment Summit, over £26 billion of the £63 billion announced was from energy companies who were stating that they would increase electricity production. One of the silver linings for the UK, because as regular listeners to Password will know from past programs on energy and data centers, The UK is in a unique position to harness power from the wind and the sea due to its geographical location, something that we should grasp as an opportunity to develop new business, according to Chris Hocknell, head of the wave power company Eight Versa.

Speaker C: If we get it right, there are a lot of opportunities. I’m not sure we are always thinking of it in the correct manner, but yeah, there’s, there’s certainly a lot of opportunity for this transition. It can create jobs all over the place. One of the problems we do have is actually we, because we don’t really make a great deal here. So much of the green tech, for example, is obviously imported. And that’s a bit of a missed opportunity because I suppose if we, if we go big on heat pumps or any other kind of technology, green tech, because most of that is manufactured and most of the jobs sit elsewhere, that’s to other countries’ benefits. But if we’re a bit smarter and tactical about things, I do believe we can, you know, we could be talking hundreds of thousands of jobs, new jobs. In various different sectors, just the enhancement and the improvement of the grid infrastructure that’s needed to obviously change the energy system, that alone creates 20,000 permanent jobs. And, uh, yeah, that’s not to be sniffed at.

Speaker B: Somebody that we’re going to be interviewing tells me that some people are talking about plans to start sailing ships going around the coast of the UK, which would rejuvenate all of these old ports and harbours that we used to have. And it sounds quite romantic.

Speaker C: Yeah, I suppose it does. I mean, that’s one of the key things, I suppose, if we start— I mean, I think this does come with some security issues, but we’re going to move a lot of electricity generation infrastructure from on land in very localised, high density kind of generation points throughout the country to less dense, much spread out areas. Away from the land and into the, into the sea with, you know, for example, the huge offshore wind farm that’s planned at Dogger Bank. That is absolutely enormous. And that’s obviously, as I said, there are some— I have a great deal of security concerns about that. But I mean, that whole East Coast will— well, you’ll need people going, thousands of people coming and going every day for maintenance and all kinds of surveys and things like that. And if, yeah, if done properly, You could think of a very, I think, a very effective way to rejuvenate some of these communities that have really been chewed out and spat out from obviously changing habits and tourism and deindustrialization, etc. Hopefully a lot of those seaside towns, as we know, have got a lot of problems. It’d be fantastic to think that we could create a lot of high-tech, or at least highly paid technician jobs there. That would be an incredibly exciting prospect.

Speaker B: Yeah, it would mean that places like Clacton would suddenly have a complete rejuvenated economy, and it might even make them want to change their MP. One of the people that we’ve been interviewing, this is a fascinating comment, was that he was talking about all of these islands that are in the Pacific, for example. And, you know, hitherto they would have had problems with conventional energy. And this guy was an expert on wave power. And he was saying, well, they are surrounded by this ocean of energy. It’s just that they haven’t been looking at it in the right way. That’s the big problem that we’ve had, isn’t it? We’ve been a bit sort of energy myopic.

Speaker C: Yeah, I say so. I think we’ve certainly of the— in the complexions and the ideologies within governments for decades now seem to have quite shied away from the idea of industrial strategies and I suppose advanced technical technically, I suppose, based strategies to this problem. It’s a real shame. And, you know, I don’t think it’s a coincidence that many other countries have made a lot of progress in and around, you know, latching on to these industries and creating sectors and employment, etc., as we touched on before. I’ve been hearing about wave and tidal power in the UK, its potential, for decades now, and we just It says, oh, you know, the UK has got the greatest resources, or one of the greatest resources and opportunities there, but very little happens. And often when you find— this was really the case with the previous government— then anytime we found some, I suppose, public-private partnership notion or idea, would be it was small modular reactors or something like that, the government would put in some pitiful amount, £40, £50 million, when they were— and they were expecting the private sector to put in £3 or £4 billion. And it’s just not really been taken seriously, that lack of investment and that kind of agnosticism, I suppose, to technologies and industrial strategies, something that’s, I think, is quite sad. It’s ground my gears for quite a while because we’re kind of trapped in a bit of a stasis while other countries and industries move on. I mean, all of the wind turbines are made somewhere else. All of the PV panels are made somewhere else. All the heat pumps are made somewhere else. We do very, very little in this. And while we’ve got very grand ambitions, You know, we don’t have much in the way of domestic, domestic capacity and opportunity. And, and some of that we do have, I suppose, innovation from an academic perspective. We are obviously still doing quite well in the UK, but usually as we, as we know, once it reaches a certain point, it goes off to the US to get commercialized or perhaps somewhere else in East Asia. And then it really leaves. So, uh, yeah, I think the government’s also got a lot to blame in this. They really have not seen a strategic opportunity and nowhere near, near any investment has gone into it. At all in the last kind of 10, 15 years.

Speaker B: Chris Hocknell of the wave power company Aitversa echoing the comments from many of those interviewed for our green energy program last September. People like the Scottish entrepreneur Russell Dalgleish, who all pointed out that for renewable energy to work, it has to make business sense, particularly because even making AI use less energy due to new innovations like DeepSeq will not mean that we use less of it. And as many people have stressed, AI development has not really started in many parts of the world. Here’s Professor Chien once again on why our AI use will only increase.

Speaker D: Sometimes increasing the efficiency doesn’t actually reduce consumption. So it basically increases consumption because it enables it to be easier and easier to use. So he was studying coal, right, and steam engines, but it applies equally to computing. So the fact that each video conference gets cheaper means that they can be given away for free, which means that I can use 10 a day instead of 1 a day or something like that. So I think that computing has been the, you know, poster child for what’s called take-back in the environmental community or Jevons paradox that every quest to make it more efficient causes more use to happen. Now, I’m not against efficiency, right? I don’t mean to say that the efficiency is not worthwhile because the use, I think, is enriching our lives, right? So it makes things possible, it makes, you know, gives us joy, gives us commerce, all those things. But I don’t think people who think efficiency is going to solve this problem really understand economics. Impacts in human behavior. I’ve begun to work actually not on efficiency over the last 10 years. I’ve come to the conclusion that sustainability has to be about holistically trying to mitigate the negative impacts of the consumption, because you cannot slow the increase of the consumption. And autonomous cars are going to make that even worse. We’re going to have more cars.

Speaker B: I can see that as the postulated future progression of the technology will be, that we will be in 3D environments and increasingly realistic 3D environments with even greater data demands, and as you’re saying, that the efficiency will just lead to more consumption.

Speaker D: It looks that way. I don’t know about the 3D environments, but it looks like consumption grows steadily. And, you know, it’s not just us. Those of us living in the developed world have to be a little bit humble about this. We’re already living a high-consumption lifestyle. There’s billions of people that aspire to this kind of high-consumption lifestyle. And we have no right to prevent them from, from growing into this, this kind of rich life that we have. So we have to prepare to find ways to sustainably support this growth in consumption that appears to be inevitable.

Speaker B: But one of the problems for those people, and again, it’s something that we picked up on in previous radio programs, is that there is this huge data divide in places like Africa. As you say, they won’t have the data center infrastructure to be able to get that life that you’re talking about unless they actually go over into the servers where the servers are, which are 50%, as you say, in the US, but a substantial amount in the UK and Europe as well.

Speaker D: I would separate two things, right? So I would say without doubt, there’s going to be a huge, hugely larger number of data centers and data in Africa over the next 10, 20, 30 So that number is growing, right? Are they going to catch up to the numbers that are growing in the developed world? Well, that’s hard, right? In the US and the UK, they’re growing exponentially, right? So how do you catch up? But I think there’s no question that there’s secular growth across all of these, and that’s the environmental problem.

Speaker B: And data centers do not just hold issues for the environment. Underlying the movement for data centers in Europe and other parts of the world of the world are geopolitical concerns. Having your nation’s data outside of your national borders is increasingly being seen as a national security issue. And currently, over 50% of the world’s data centers are in the US. And data from Europe arrives there through undersea cables that come up in Virginia, the place where data centers are now competing with people for water. It’s also a state that, due to the emergence of those undersea cables, has also been traditionally attractive to the powerful US intelligence agencies, the CIA and the NSA, which both have their headquarters nearby.

Speaker D: Well, it’s funny because today is, uh, January 6th in the United States, so that’s a particularly notable day for us, but Yeah, I think there is an interest in control over data. I think that is the continuing trend of historical assertion of sovereignty by nations. There’s no question that the data is probably as important, if not more important, than physical sovereignty over people, spaces, enforcement of law, and so on. So that’s one of the big societal challenges to work it out, because the dream of the internet pioneers, which was information flowing freely everywhere, wherever it wants, and the famous information rules that were published by this gentleman at Harvard, they just don’t work with those dimensions of society. So this is one of the big struggles we face. And perhaps it’s not surprising that half of the data centers in the United States are in North America, because in some sense the data can move more freely and might be arguably viewed as more valuable, can be exploited for more value in the United States because we have so little regulation around that, right? Now, I’m not saying that’s the way it should be. I’m just saying that as an observation about where we are.

Speaker B: But there’s also another observation that somebody could make. You say that those data centers, for example, that are located in North Virginia have been there because of the fiber optics in the ground. That’s also where cables come across from the Atlantic. So those are places where that data can be analyzed. It can be, as you say, it can be picked out for its value by the large big tech companies. And you want to do that, as you say, for citing reasons as closely as possible, because it’s economics, isn’t it? It’s economics that is actually driving that, driving that cluster.

Speaker D: Yeah, I think that’s right. It’s economics. And, you know, that kind of economic structure goes back to the insurance exchanges in London, the stock exchange in New York, and so on. These are information mountains, if you will, that have led to economic activity and value for— and these things are not new human trends, right? It’s not just Northern Virginia, right? There’s an analog of this on the other side where there’s a cluster of data centers in Ireland, because Ireland is where many of these cables are coming in and mediating Everyone knows about the Great Firewalls in China. Those are the moral equivalent, right, of this peripheral boundary, right? So I think this is just the evolution of the structure of how information is shared and stored and mediated in the world being expressed in the internet.

Speaker B: Professor Andrew Chen, the William Eckhart Distinguished Service Professor of Computer Science at the University of Chicago. A former vice president of research for the US chip giant Intel and an expert on computer power use whose help in the making of this program was invaluable. If you’d like to hear the full interview with Professor Qian, please log on to the Future Intelligence website, www.futureintelligence.co.uk, where you will find it as our latest Peter Warren Interview Podcast. As Professor Tian has pointed out, making AI more efficient will only increase our use of it, but the need to do so is now imperative, according to Sandeep Channa, the chief technology officer at data cloud provider CSI Group and an expert on AI use.

Speaker F: Generative AI, you know, surging sort of GPU shipments in the industry. Data centers are really starting to struggle and they’re having to scale at a phenomenal rate. And let’s be honest, we knew that data centers couldn’t cope as they are. We know that the mean temperatures have been pushed up in DCs over the years. So less power is consumed. And actually, you’ve got the big players like the NVIDIAS of this world that start to sort of roll out next-gen AI-enabled chipsets. And all this kind of stuff that’s happening. The challenge then goes back to the same thing, right? And there’s a very famous, there’s an OCV case study, right? Where actually we can understand heavy use of compute really at the beginning while a model is being set up or trained, an AI model of some description. But actually, once that’s done, there is nothing to say, let’s move it onto cheaper compute. Let’s bring the scale back because now the model’s clean, the data’s cleansed, everything’s there, it works correct, let’s move it. Away. Nothing happens like that. And the industry is creating a real problem for itself. I think we’ll keep going. And at some point, they’ll realize that there’s a real problem. We’re running out of power. And now we’ll have to scale the whole thing back. But yeah, it’s a problem that’s not going away that easily from what I can see.

Speaker B: But I think the industry has been creating a real problem for itself right from the word go, because it’s basically been tailoring stuff to a problem that it wasn’t designed for in the first place. So computing has always sort of, oh, we can use it for this, we will fit it in this. I mean, you look at the internet, no security inherent within the internet. There is no security inherent within computing right from the word go because people didn’t think that people would do bad things.

Speaker F: We’re starting to hear that energy companies are actually saying things are starting to creak, you know, where people are pushing generative AI, but the resource workloads are falling. Are absolutely frightening. And look, I know the chipset manufacturers are trying to bring out much more efficient chipsets, but I think there’s a massive gap here. And a lot of people who we talk to in the industry, whether they’re creating AI platforms or whether they’re consumers, nobody talks about the compute. And nobody talks about how much energy is going to get wasted. Hang on, does this even fit into your company as this ESG policy over here? You’ve not married the two up, that you’re just compounding the problem on this side. It may not live in your data center. It might be in a public hyperscaler. Are you just pushing the problem somewhere else? None of that is, is again being looked at.

Speaker B: Yeah, but the problem seems to be we started off with the CPU, then people wanted more and more GPUs because they wanted to have these pretty pictures and these sort of graphics. Now, We’ve got both of those and people are talking about an AI PU. So you’ve got these 3 chips and the thing just seems to be growing organically rather than anybody sitting down and saying, hang on, I’m going to start this from scratch and I’m going to build something that is made for purpose. I could see NVIDIA. I know it’s bigger than Apple at the moment, but I could see NVIDIA, somebody coming along with some solution and NVIDIA just disappearing like that because people will say, yeah, why on earth were we using all of these GPUs?

Speaker F: I agree. And look, energy companies are starting to cotton on to this. People are saying, you know, you’re talking in the next 5, 10 years, a tenfold increase in data. I’ve heard some countries are starting to actually limit and say, no, we can’t have so many new data centers or so much power being consumed by data centers because it’s a real problem. As we know, right, you know, there’s things going around across the globe, there’s wars going on, that power is premium. You know, everyone knows costs are escalating. You know, us internally as an organization, our power costs across all our global footprints have just soared, and all we’re doing is making this problem worse.

Speaker B: And Sandeep Channa’s words are not falling on deaf ears. On the 21st of January, Energy Secretary Ed Miliband, in giving evidence to the House of Lords Environment and Climate Change and Science and Technology Committees, not only confirmed that government expected to see electricity use soar over the next 4 years, he added that it expected it to double from today’s figures by 2050, a 100% increase driven by technology. It is these concerns, and those expressed by Professor Qian, that are driving many involved in the industry to call for very joined-up thinking about AI, electrical power, and cooling. It’s something that Shaun Redmond, the head of Silicon Catalyst and one of the government’s advisers on computer chip development, says has to happen.

Speaker G: We need, as you say, a 22nd century network infrastructure that’s going to take us to delivering on— as AI compute grows at 10x per year, the amount of data moved around the network is going to move with it. We have to be able to have infrastructure that deals with that. Otherwise, it becomes the bottleneck. And in fact, it is, you know, switching is the bottleneck for all data centers. And network switching is recognized as a fundamental blocker of progress. Some of what we’re talking about here hasn’t even been invented yet, so needs to be. So, you know, what’s available? Could UK infrastructure invest in that and improve it?

Speaker A: Yes.

Speaker G: But if it takes another 5 to 10 years, we’ll be delivering orders of magnitude improvements in technology. Which is needed in the UK.

Speaker B: What’s really interesting about what you’re saying is that AI is enormously inefficient, isn’t it? What we’re doing, everybody is presenting it as this sort of sentience. It’s not sentience. It’s just huge amounts of responses being plastered across something until something seems to make sense at the top. Of it. And so it’s probabilistic with the determinism of a human being sitting on top saying, oh yeah, that’s it. And you’ve done that faster than me. But as a process, it’s terribly inefficient. And so what you’re saying is that we have got this terribly inefficient system. We’re using huge amounts of energy to make that work at the moment.. And also in terms of the distribution of that information, when you’ve got more and more and more people asking for more and more and more solutions to a particular issue and they’re asking for all of that compute, you’re just accelerating this, this enormously inefficient process. So do we need better chips in strategic places using the current model? And if so, where do we need those chips? Do we need them in sort of localized servers or localized service centers, or do we need to? Because evidently we need to solve those problems that you’ve outlined so eloquently already about, you know, what we do with this. I mean, the weird thing about this is it’s been presented in the media as though it’s a done deal and we’ve solved everything. We haven’t.

Speaker G: This is literally the start. You’re absolutely right there, Peter. You know, this really is literally just the start, which makes it exciting, but also somewhat terrifying as well. Because if, if we go down the same path we’re treading, then 15 to 20% of the world’s electricity consumption will be data center requirements. That’s not sustainable. We don’t have enough energy supply to do that today. And so if we’re talking about infrastructure, if we keep going down the same path and relentlessly repeating what we’re currently doing, then we’ll need to seriously think about power station infrastructure in the UK because we wouldn’t be able to sustain that amount of data center compute. So we’ve got to move it out of the data center. It’s got to move to the edge, primarily because, you know, moving decisions up and down to where it actually is needed doesn’t make a lot of sense anyway. So it slows everything down and uses a lot of energy through the data center switching. And so those architectures at the edge are really important. And I think even improving the energy efficiency of them, if you could make them near zero energy, what a wonderful world we would be living in, because that really transforms application spaces. Applications start to arise that we just wouldn’t even conceive of today. It’s a bit like the transformation, you know, went to the mobile phone and then suddenly all of it became this compute engine. And we live off that world of application-driven computing in our hands today. So near-zero energy compute at the edge, all sorts of wonderful new applications would emerge, which will come down to transforming healthcare in society, transforming our environmental challenges in cities, dealing with crime. Then we start to talk about transformational things. If you can get down to near zero energy compute, which I think is a wonderful ambition to have, and it’s great that we’re seeing that in the UK because you don’t quite— you don’t see it elsewhere. But this absolute drive for performance is the undying sort of driving factor that comes out of the US all the time. And yet what we see out of our research base is, well, let’s solve the energy consumption problem of this.

Speaker B: Sean Redmond, managing partner of Silicon Catalyst and a government advisor on its industrial digital strategy, on the need for clear thinking over AI. Development, a clear thinking that must be focused on due to the sudden race towards AI. On Tuesday of last week, speaking at the Paris AI Action Summit, U.S. Vice President J.D. Vance made a speech setting out the need for positivity around AI development. Both the U.S. and the UK also declined to sign an agreement regulating the technology. It’s an attitude the UK Prime Minister has set out as a central plank of policy. To deploy technology and create the next revolution, not only must regulation be light, opposition to infrastructure developments must also be minimized. Saying no to pylons does not appear to be an option yet. Barely two weeks ago, at the presentation of the Future of Work Institute’s three-year report into technology implementation, the organisation stressed the need to include county councils and local authorities in infrastructure deployment. Though, Professor Bart van Ark also pointed out that it was inevitable that the AI revolution will leave some people and communities in the UK behind. Just what the impact will be is difficult to gauge because, as J.D. Vance pointed out in his Paris speech, we are at the beginning of a profound change in our existence. This change in government attitudes around the globe has prompted some to point out that in the heady early days of AI, it was trumpeted that there was a need to consult every sector of society. Society. It’s a refocus that already appears to indicate that AI may not operate for the good of all, according to Elke Schwartz, Queen Mary University of London’s Professor of Political Theory.

Speaker A: Speaking on the day of the start of this week’s AI Action Summit, Professor Schwartz said: I am hoping to see from this action summit a greater balance or rebalancing of the scales between AI industry and public interest. And some of the work streams and some of the workshops and events and participants that have been drawn from the non-AI sector, let’s say, and, and from individuals who are concerned with the public interest is quite promising in that respect. But as ever, the challenges remain that for an event that centers around artificial intelligence, that industry interests and industry narratives and the general drive towards more artificial intelligence will prevail over perhaps a more cautious approach to say certain things cannot be addressed with artificial intelligence. It’s a good, it’s a good and an interesting plurality of participants and speakers. So I’m hopeful that this will be a less one-track industry-determined kind of summit.

Speaker B: But we’ll have to see. There is a little bit of an issue though here, isn’t there? In the heady early days of AI, it was suggested that everybody, all sectors of society should participate. We’re not quite seeing that. We’re just seeing industry and academia and senior politicians, aren’t we?

Speaker A: Yes, it seems to be the case. And what I’ve, what I have observed over the last, let’s say, 5 years or so is that government discourse seems to very much have aligned with industry discourse itself. So that the drive to unrestricted almost innovation seems to be foregrounded by government voices as well, and not just industry. So things become a little bit more, yeah, unidirectional, let’s say, and a little less open for dissent, for example, for where we draw absolutely hard lines in terms of where we don’t want AI to be involved. So I would welcome, especially with the speed of rollout of various products or ideas, you know, usually the products are not necessarily always that sophisticated in the first place, but with the rollout of various ideas and future-oriented kind of prospects of the fantastic things that we will be able to achieve as societies with AI, to kind of bring more of a realistic, less hyped refocusing on what does society need, where the actual hard problems, which are always social problems and not technical problems, and how can we solve them.

Speaker B: It appears from the outcome of the summit that Professor Schwartz’s concerns were justified. To hear the full interview with Professor Schwartz, go to the Future Intelligence website, www.futureintelligence.co.uk. You’ve been listening to Password on Resonance FM about the world’s dizzying future. Password was written and presented by me, Pete Warren, and produced by Blue Buffery, and is a Future Intelligence production.

Speaker A: Thanks for listening and goodbye. This program has been brought to you by Resonance FM. If you like what you heard, please support our work by making a donation at resonancefm.com/donate.