Silicon Empires w/ Nick Srnicek

This episode of The Dig is brought to you by our listeners, who support us at Petriand.com, and by Haymarket Books, which has loads of great left-wing titles, perfect for dig listeners like you. One that you might like is Cold War on Five Contents, a global history of empire and espionage by Alfred W. McCoy, transforming battlegrounds in Africa, Asia, and Latin America into veritable hellscapes. The surrogate wars of the Cold War era left behind a legacy of collective trauma and social conflict that persisted into the present. In the ambitious work, Alfred W. McCoy uses a bottom-up outside-in approach to offer an unexpected new perspective on the longest, most consequential conflict in modern history. McCoy renders an intimate portrait of both em battled covert operatives and committed anti-war protesters, thus humanizing the history of the Cold War. Find Cold War on Five Contents at Haymarketbooks.org, where all hardcover books are 10% off every day. Welcome to The Dig, a podcast from Jacobin Magazine. My name is Daniel Denver, and I'm broadcasting from Providence, Rhode Island. Today, I ask my guest Nick Cernick to explain how AI came to occupy the epicenter of everything about our unshitified moment, our tech oligarch-dominated economy, our fascist authoritarian maga government, the most fundamental aspects of both our nomic individual consciousnesses and shredded social fabric. But this interview is above all else about how AI is also, critically, the full crum of contests and conflicts, shaping the new world order, with control over semiconductor and AI supply chains and technology, a key site in the struggle for economic and military dominance. There is so much to discuss. Nick, among many other things, is the author of the new book Silicon Empires, The Fight for the Future of AI. Before we get this podcast started, first, much love and solidarity to everyone in the streets and Minneapolis. I've got an interview coming soon on the deep, long-running work that's made Minnesota one of the best organized places in this country. And then, last thing, a tiny fraction of regular listeners, probably under 5%, make this whole thing possible with contributions at patreon.com/thedig. If we pay well episodes, the number of contributions we got would be a lot higher, and we would make a ton more money. But our total number of listeners would be much lower too, and we're not here to make money. We are here to provide rigorous, non-dogmatic political education for a global left that is doing such urgent and important work everywhere. And so we don't pay well anything. We want everyone to listen to everything. If you are a regular listener, please become one of the people who make this show possible. That's p-a-t-r-e-o-n.com/thedig. Please contribute now. Okay, here's Nick Surnick, senior lecturer in digital economy at King's College London. His books include Silicon Empires, After Work, a history of the home in the Fight for Free Time, co-authored with Helen Hester, Platform Capitalism, and Inventing the Future, Post Capitalism and a World Without Work, co-authored with Alex Williams. Nick Surnick. Welcome to the Dig. Thank you for having me. I want to start by asking a somewhat abstract, giant question before getting into a lot of particulars. What does it mean that AI has so quickly come to dominate our economy, foreign policy, politics, culture, energy markets, land use, etc., etc.? What then does that have to do with the powerful embarkation of big tech oligarchic economic power and fascist authoritarian political power under Donald Trump? All of that amid a culture drenched in hideous AI slop and chat GPT-generated mass literacy. What does it mean for AI to be at the inshidified center of this moment of grotesque violence and equality, domination, war, ecological crisis, and generalized insanity? Starting off nice and simple, I see. I think at the heart of it is the promise, and there's, you know, we can have debates about this, but I think the promise that AI is a transformative technology that's at least on par with earlier transformative technologies like the steam engine or electricity or things like that. On the back of that promise, there's hopes for controlling the future of the economy if you're a big tech company and controlling the future of the world if you're a major geopolitical power. On the back of that promise, you end up wrapping in every other element of the economy. As I'm studying this field and trying to learn about it, you just start, and you end up in places that you never expected to go. I know far more about gas turbines today than I ever thought I would need to know. You know, there's just, it starts stretching into every part of the economy and every part of our lives, but it is, I think, fundamentally that promise that it is a transformative technology. In addition, I think there's the key element that AI, unlike a lot of earlier transformative technologies, it gives early movers a massive advantage potentially, whether that be in the economic sphere, whether that be in the geopolitical sphere. There's a real sense in a lot of imaginations about the future that if you can get there first, you will massively benefit from it. So that, I think, is really what's driving a lot of this stuff at the moment, the intense focus on artificial intelligence. As of October, 80% of US stock gains were due to AI companies alongside just an astounding 40% of US GDP growth, according to the FT. And it increasingly seems like the whole thing might pop, but we'll get into that potential popping and its implications near the bottom of the interview. Let's start by assessing how we got to where we are now and where we are now. The AI boom, it's been overwhelmingly US and then China based. And you write that it's been comprised of different sorts of companies trying to make money in different sorts of ways. There are companies developing AI models like open AI, meta and thropic and deep seek. The big cloud companies like Amazon, Microsoft Oracle that are developing infrastructure, this explosive boom in sprawling data centers to provide the growing demand for compute the so-called hyperscalers. Then we have the companies making and selling the hardware, the chips, most importantly, TSMC and Nvidia. But you write that the layer that is quote, the most uncertain and influx is diffusion, the making and selling of practical AI applications, which is where the big revenue generation that the entire industry's profitability, it's what that profitability hypothetically depends upon. That's where it would happen in the application diffusion layer. Some companies are trying to do it all by being conglomerates, Google, Huawei, all of these companies you write have different sorts of strategies to capture the profits of an industry that's heavily reliant on huge amounts of perhaps naively optimistic investment. Before we drill down into a lot more detail, can you lay out this in big picture form, the geography of this suddenly enormous industry? Yeah, so there's a couple of different ways to look at it. One is the different layers of the AI stack as you point out. So you can think from effectively the bottom, which is the hardware, the chips and videos, GPUs, you can move on up to the infrastructure layer, which is the cloud computing companies, which are putting at this point hundreds of thousands and potentially millions of chips together into massive, what are called compute clusters. And then on top of that layer, you have the models themselves, which are being built. These massive amounts of parameters going into these into these neural networks to produce what we know of as an AI model today. And then on top of that, at the very upper layer, you've got apps, which is how we actually access these models. So most of us are not accessing these models through an API, for instance. Instead, what we're doing is we're accessing it through chat GPT as an app, or deep seek as an app or whatever the case might be. Now, that's one way to think about this infrastructure, this landscape of AI. And I think it's an important one to keep in mind. The other way, though, and the way I think is really crucially important, is to see AI, generative AI in particular, as a general-purpose technology. And so, from economics, the idea of a general-purpose technology is that it's a technology which is hugely significant, changes productivity in a variety of different industries across the economy, accelerates economic growth, leads to disruptions of the existing business ecosystem, overthrows and compents, and then also has social and political impacts as well. Now, if you take AI to be a GPT, I think what's really important for the sort of story that I want to tell in my book is this distinction between the general technology itself and the specific applications that are made of it. So, the semiconductor, for instance, is a general-purpose technology, but the application of it to your fridge, to your phone, to your laptop, to a tablet, each of these things is a very different application of that general technology. Same thing holds for AI. You've got these sort of general models, and then they're sort of being applied to the legal space, or the biotech space, or the education space, all these different applications being made of it. And what I sort of note in my book is that historically, research shows that the people who actually make the GPT, the general-purpose technology, they don't tend to gain most of the benefits from that innovation. And in fact, most of the value goes to those applications. So, this has always been a problem in the past for the producers of general-purpose technology, and one of the common themes in the economics literature is precisely that there's so few incentives for companies to produce these GPTs that in fact the government has to step in to provide incentives, because it's socially beneficial to have these technologies. But individually, it has a company that's not beneficial. With AI, it seems slightly different, and it may turn out to be historically unique. But I think the companies are at the moment trying to deal with that distinction between the general and the specific, and trying to capture as much value from the specific as is possible. And this is sort of leading them, each of the major AI companies, in different directions to try and capture that value. And so a historical precedent there in terms of a general-purpose technology would be something like the invention of steam power. Yeah, exactly. So, steam power, you might think, well, steam engines in a locomotive or one thing, steam engines in a factory or another thing. Another key element of a general-purpose technology is that they tend to. they have the capacity for innovation within their sort of technological lineage. So, the steam engine can get more powerful, it can get more efficient, it can get smaller. We see similar things with AI models today. They're becoming more capable, they're able to do new things. So, this, you know, the fact that it can be applied to all these different sectors of the economy and the fact that it can be improved over time, all of that starts to. it causes this sort of virtuous cycle where these technologies end up having massive economic impacts potentially. So, the various companies developing AI models, which is one layer of this broader ecosystem, we're putting aside for now, Nvidia and TSMC and the hyper-scaler cloud companies. The companies developing AI models, open AI andthropic deep-seek. These are all pursuing. are all pursuing what's called the frontier strategy, which is attempting to build the highest performing model. So, as to then, we're in a huge share of the market and ensuing profits. Meanwhile, Google and Huawei are extending themselves into these vertically integrated, vertically integrated conglomerates that do everything in-house from cloud computing to AI models to applications. Meanwhile, meta and alibaba have open models, which don't present a profit opportunity directly, but do allow the company to benefit from the development of their own AI model that they can deploy for their products instead of paying for another company's model. And what's more, these open models can be easily modified and so, encourage outside developers to tinker with those models and then those innovations can in turn be rolled back into the company so that they can improve their own models. How do these various model-building companies hope to make money from the AI that they are driving so much investment into training? Because, as you just said, there's no reason to believe that leading on innovation will lead to monopoly profits, which, as an amateur observer, seems to be the operating assumption behind a lot of this investment. In fact, as you said, the research on general-purpose technologies, like the Steemen General Electricity, that research suggests that it's the diffusion and industry-specific application of the technology that will deliver the big profits. What are these different model-building companies trying to do? And are they operating on a misplaced assumption about where the profits will be made? So, I think the ultimate end goal of a company like OpenAI and Anthropic is very, very clear. They want to build something like artificial general intelligence, which very loosely defining it as basically an AI, which is capable of doing most human labor, especially cognitive human labor. It's sort of been, that definition has been narrowed down over the past few years. But increasingly, it's all about, can you effectively act as a remote worker? Do you have an AGI that could do that sort of stuff? The promise is that, actually, if these models become capable enough, that they sort of transcend that general and specific division of a general-purpose technology, that in fact, they become so general, that they can apply to any specific instance that you might need. And we see sort of inklings of this already. You can go to a chatbot and you can ask it to create any random image that you want, but you can also ask it for to read a legal document and tell you if there's any issues with it. You can ask it for health advice. It's already got some of these elements where it's able to be used in these specific applications. And so the goal of these companies is to build these models in such a way that they become more and more capable and more and more able to do this sort of thing without having to without having to have the sort of adoptions and fine-tuning and the frictions of diffusion that tend to go along with this stuff. So that's their promise. Now, I will say, I think if that happens to be the case, they will absolutely gain huge amounts of profit. So it is potentially a winning scenario. I don't know whether it's going to be the case or not. I don't think anybody really knows, but their hope is their drive is. Their fundamental business strategy is to build this sort of thing. On the flip side, I think there's an argument to be made. This is something I've only come around to a little bit recently. There's an argument to be made that AI is somewhat different from previous general purpose technologies in the way that it captures value from downstream applications. Because previously, what would happen was you would get a sort of, you get intellectual property over this fundamental innovation that you had made, and then you could license it out and make some money from that. But there was limits to licensing. There was limits to your ability to capture the value through that sort of legal mechanism, which is why most of the value ended up going elsewhere. What's really interesting in AI is that maybe there's a new mechanism which is emerged, which is the API, the application programming interface, which is the way in which if you're going to build a business, for instance, on chat GPT, you're using the API to access its capabilities. Now, this is distinct from licensing because it gives the model developers much more insight into what's happening downstream. It allows them to see what is becoming a successful business, what are they using it for. It also allows for instantaneous price adjustments in a way that's not possible with licensing, for instance. The field that's really interesting to me to follow at the moment is the coding sphere. You've got a number of different coding startups, which are in terms of general-purpose technologies. They're a specific application of this general technology. What's happening in that coding sphere is that the coding startups are finding a lot of success, but when they start to become too successful, the model companies are raising their API prices and effectively taking a much bigger cut of the value as a result. So there's an element of control here, an ability to capture value that I don't think was there in historic cases of G.P.T.'s. So I don't know whether this four shadows where it might go in the future, but it seems to me a really important case to follow. So is it fair to say that the expectations of AI someday unlocking profits massive enough to justify these hundreds of billions of dollars in capital expenditure that that's that's just essentially the belief or hope that these represent costs that AI will save capitalists in every single company and every single industry by eliminating just massive numbers of jobs through automation and then splitting the resulting savings between the AI oligopolists and the particular capitalist in whatever industry? I think that's definitely a part of it. Automation is the ultimate end goal. And there are a lot of people in the AI space worried about the consequences of that. And you get sort of quasi-Marxist conclusions in multiple cases. People sort of independently finding out, oh, there's surplus populations, that sort of thing. That's definitely the end goal is sort of that massive amount of automation. But I think even before that were to happen, is the oldest story of capitalism that if a company finds a new technology which increases their productivity, every other company then has to follow in that path and adopt that technology or they go bankrupt, they lose their competitive edge. And AI is the same sort of thing. At the moment, I think it can make people more productive in many, many cases. And that's going to likely improve over time. So even without automation, there's going to be that drive for all these companies to adopt and pay subscription fees to open AI and stuff. Just how capable is AI right now? And I know you said nobody knows, but how good might it get? Because obviously when we think about AI, as you mentioned, we think about consumer facing AI products that hundreds of millions of people now use every week like chat GPT. But again, as you've suggested, enterprise AI is where the tech industry expects, perhaps just fantasizes about making the huge share of the profits that it needs to justify all this capital expenditure. And we're talking about applications for everything from law, medicine, pharmaceuticals, education, manufacturing, including through the development of autonomous agents that can complete full tasks without direct human oversight. So my question here is what is the technology on track to deliver the sort of giant profits that would justify the these corporations spending loads of money on enterprise AI? You write that, that this oligopolistic market structure could quote, ensure that the potentially immense value created by AI ends up in the hands of a diminishingly small number of people posing perhaps unprecedented concentrations of power and wealth. That is one really bad outcome. Is it is it also possible though that that those people make a ton of money from an AI bubble that that pops never to be reinflated? Yeah. So we can talk more about the bubble and the implications and the potentials of it later on. But I will say I think even if there's a bubble and I think there probably is, even if it pops, AI is here to stay. And I think this is one thing that I think that the left hasn't given enough thought to. There's a sort of denunciation of the hype around AI and appointing to its limits and appointing to its biased implications when it's applied and all of that is useful and important. But AI is not going to disappear when the bubble pops. It's going to continue on and it will continue to become more and more capable in many, many ways. On the question of how capable it is, I think this is a really difficult, interesting question and it's also a really important one. At the moment, I think the best way to characterize the capabilities is what's been called jagged intelligence. So the idea being that AI can, its intelligence can be off the charts on certain points and certain times when you question it and give it, you know, ask it for an answer. And other times, it can be just, you know, give you absolute nonsense, you know, hallucinated nonsense. So jagged intelligence points to that sort of pointedness of like its intelligence. It's not how we would expect sort of human intelligence to function. That being said, effectively what's been going on, the story of, say, the past five, six years of AI research is that first of all, what's called pre-training is really important for these models. So you train them on the, an internet's worth of data and you use massive amounts of computing power to do that to you. You use all these algorithms to train it and you end up with the model. And as we, as we introduce more and more computing power, we found that these models were getting better and better and better. So this is sort of one vector of improvement for these models. That seemed to have stalled this year, although there's questions about whether or not that was actually the case because effectively we haven't had larger compute clusters to train these things on. So we haven't really been able to test whether pre-training still gives the sort of expected improvements. But what the story of 2025 was, is about reasoning and agents on some level. And here the point is that companies started using what's called reinforcement learning to do post-training. So after you train this model, you do some additional training afterwards through reinforcement learning. And the thing about reinforcement learning is that you need some sort of signal that acts as a reward. So reinforcement learning works really well for fields where you've got a definitive answer for coding, for mathematics, for these sort of like hard, concrete answers. And that's exactly where we've seen improvements this year in the capabilities. Models which are winning gold medals in the most prestigious mathematics competitions. Models which are winning programming competitions as well. So there's been lots of improvements in those directions. But again, there's still, you know, if you use these AIs in any, you know, with any frequency, you can also see their limits quite often as well, despite all of these improvements over the past year. Even if AI gets better and better, can it actually reach AGI, artificial general intelligence, which is what at the end of the day justifies this enormous capital expenditure, because when people like Mark and Dreson talk about AGI, they talk about it like the rapture. It's a discourse that can be suffused with outright messianism. But it of course also can take place in a less sci-fi, more capitalistic register, recently Microsoft and OpenAI, signed an agreement that defined AGI as the development of AI, of AI systems that can generate at least $100 billion in profits. Is AGI a fantasy or an economic terms, an objective benchmark that will at some point possibly justify all of these resources from the investor perspective? I mean, I think it's both. I think, personally, I think AGI is a theoretical possibility. I think there's no reason that intelligence only has to be instantiated within brain matter. I think it can be instantiated in other physical media as well. So I think it is a theoretical possibility. And I think there's good sort of philosophy of mind reasons to think that whether that's what companies are building at the moment, I think is a different question. It's what they, it's their stated goal, and it's certainly what they're telling investors. And you read some personal stories about some of the people in these AI labs. And you very quickly realize it's not just a sort of story to tell investors. It is a real lived belief that is driving these people to do all this research and to, you know, to do so much work for it as the sort of term goes in the field. They feel the AGI. So I think it's all of these things. It's a live belief. I think it's a theoretical possibility. It's also an investor story. The best sort of thing I can compare to is the way in which Moore's law for semiconductors improving was not actually a law. It was sort of a prediction based upon, you know, a short period of time. And then vast amounts of resources poured into maintaining that law over time. And a similar sort of thing I think is happening with AI, where there's a belief that these capabilities will improve in the sort of the same way that they have been improving. And that if we just, you know, make a few more major steps, we'll have AGI. But there's no law there. There's no physical necessity of that. There's no theoretical necessity. It is largely just about pouring huge amounts of resources into it to try and find that sort of next step, which this year has been reinforcement learning. And next year it may be something different, you know. You on Musk helped found open AI as a as a nonprofit organization in the name of ensuring that the technology that AI technology is developed in the interests of humanity, specifically guarding against the possibility of a dystopian terminator style sky net scenario where artificial intelligence turns against us in some horrible way. And I mean, these days those pretensions are pretty hard to take so seriously given that same ultimate converted open AI into a very much for profit operation while Musk's X AI does does the same in Musk's account. Of course, this is okay because he needs to be the one in charge of developing AI so that it doesn't kill us all. And at the same time though, there there's still these very vociferous tech world critics of AI concerned about so called at X risk, existential risk. Most notably, Ellie's or Yodkowski of the machine intelligence research institute. Meanwhile, some on the left charged that the entire AI safety discourse is just this huge distraction from more mundane harms caused by AI right now. What accounts for these strange dynamics around the debate over AI safety I find it very, very confusing. And then in your opinion, should we be worried about X risk the prospect that AGI will become ASI or artificial super intelligence in the 111 at that. Yeah, so it's it is really complicated this sort of stuff. I think partly because you know some of these discussions around AI safety have been going on for 15, 20, 25 years now. And there's quite rigid communities that have built up over time. And these communities as chat GPT came along have sort of transferred their earlier positions into the the era of chat GPT and Yodkowski is, you know, a good example of it is is one of the major criticism of his book that came out this year was precisely that he could have written it at any point over the past 20 years. It wasn't dependent upon you know it wasn't reflective of the technology of today in any sort of way. So there is I think there is a group of people who extrapolate from the sort of increases in capabilities and then speculate on what might happen as a result. And the most plausible scenarios I sort of find are not that some sort of malevolent AI comes about if ASI ever comes about and has some sort of disposition towards humanity, it will just be boredom and curiosity rather than fear or something like that. So I'm not that worried about a malevolent ASI but there is more sort of subtle ways in which you know an ASI may just simply see humanity as a bunch of atoms that could be rearranged in better ways more efficient ways. And that's the sort of thinking that you know it's highly speculative. But I think sort of somewhat plausible if you follow these lines of thinking to the end. I will say in terms of the communities is complicated because not only are these sort of internet-centric communities in many cases you know they've got their sort of hubs of discussion on the internet. But in many cases they're also backed by vast amounts of money whether it be through crypto or things like that. But they end up having a lot of money and they end up having quite a lot of say over how people are thinking about these issues. And one thing that's very confusing about interpreting the AI safety debate from afar is you have different people of the tech world who appear to be on very different sides of that debate. And then you have certain tech critics on the left who say that the entire debate is itself a material and materially interested smoke screen. So I think the sort of worries about existential risk definitely at times are a matter of projecting hype about the capabilities of this technology. And they do serve a sort of you know a function of getting investment from funders. But on the other hand there are I've met a number of them there are people who genuinely fear for humanities you know outcome if this sort of stuff comes to pass and they think that it may come to pass quite quickly. So I think the best way to sort of present it is that people like Sam Altman and Elon Musk I would say nowadays very much weaponized this genuine sort of fear from a group of people and weaponized in such a way that it helps them get more money and probably helps justify in their own heads why they need to be the one to be leading the charge here. So I do think there are genuine real concerns. I think my disagreement with a lot of the AI existential risk community would be over timelines I don't see it happening all that soon. But I do think it is again a sort of theoretical possibility that we need to be alert to in the same way that we are alert to asteroids potentially hitting the earth. So I think that's unlikely to happen. But it has happened and it could happen again and better to be prepared than not. I want to turn to the the strange interlocking financial relationships that drive a lot of the perpetual money motion machine feeding this explosive growth of compute. Of course, is making a ton of money selling GPUs while TSMC is making a lot but much less money manufacturing them. The hyperscalers are profiting from selling compute provided by the data centers that they're building. But there's this huge question as we discussed as to as to how the AI model building companies like open AI how they'll generate the profits to justify this spending on hundreds of billions of dollars worth of compute the spending that is ultimately required to sustain the whole ecosystem. And part of the way this works now pretty astoundingly is that companies like Nvidia and Microsoft have have created various mechanisms to finance customers like open AI and in tropics so that the latter can purchase really large amounts of the former's product. How are these forms of what you call control without ownership structured and and what do these financial relationships reveal about about the nature of the entire industry and about about the lines of capital and power that that transect it. Yeah, I think it's a really interesting element of the AI industry and this idea of control without ownership is a term I borrowed from Cecilia Rikap's work and she's got a fascinating piece on this topic as well. But the core sort of idea is that so I mean part of it has to do with the somewhat contingent historically speaking contingent element of the Biden administration's antitrust turn. So the ways in which Lena Khan came in and Tim Wu and Jonathan Cantor came in and basically closed down a lot of the ways in which big tech had been trying to amass power over the previous decade and particularly through something like mergers and acquisitions you know just buying up competitors or buying up you know companies that seem like they were going to become successful. And that that died out over the Biden administration period it's you know it's a fascinating shift that I think is you know still yet to be really fully explored the implications of it. One implication that's really important for the AI story is that Microsoft couldn't just buy open AI, for instance, and Google couldn't just buy Anthropic and Amazon couldn't just buy them. The regulatory environment, while these companies were at that early stage, it wasn't there for them to be taken. So instead, what happened was loads of different investments. Particularly for the hyperscalers, the cloud computing companies, it was a matter of investing money into these companies, but oftentimes just providing them with cloud credits that could only be used on your platform. So for the hyperscalers it meant, A, you get some equity in these companies, these rising startups, but B, you also get a customer, which is now building their models, their services on top of your cloud platform. And it's, once you've built on top of a particular platform, it's not impossible to leave, but it does pose challenges in additional costs. So getting these major AI companies to build on your platform was a really important step to controlling them. In addition, with that investment, you ended up getting insight into what are these companies doing? Have a say as an investor in where these companies are headed and you may be discussing with them about what should they be doing next, what should their strategy be? How are their technical advances going? So by investing in all of these different companies, the major hyperscalers ended up getting a quite penoptic view of everything that was going on in the AI startup world. So that's the control with our ownership that ends up being, I think, quite significant for understanding how power is operating today in a quite different antitrust environment than what we had under say Obama. And yet at the same time as we have the rise of these forms of control with out ownership, we still have the conglomerate strategy, which seems like precisely the opposite way to go about it, which is through direct control up and down the value chain, you write quote, the conglomerate strategy means moving into the model and app layers, but not yet becoming an end user within the industry. For instance, Google and Microsoft are developing AI-based healthcare tools, but are not, again, yet owning and running their own hospitals. Are you suggesting here, at least in a friendly enough antitrust environment that the vertical integration of a company like Google or Microsoft could continue to extend indefinitely to the point of taking over these giant consumer facing sectors, in which case we're really thinking about big tech taking on a sort of stature within capitalism, traditionally occupied by finance. And I would say even deeper than finance has ever gone, the sort of level of power and the level of control there is far more than what finance has ever had. I think there is definitely the potential for these companies to continue to extend out into these different verticals. And particularly when you look at a company like Google, Microsoft has a slightly lesser degree and Huawei has a slightly lesser degree as well. There is very much this intention of moving into different verticals and particularly focused on lucrative verticals like healthcare, multi-trillion dollar market and education, another sort of trillion dollar market. It puts the digital ads business, which I think is maybe a trillion dollar business now. I forget exact figures. But it puts the digital ads business to shame. These are major markets that these companies are moving into. And part of the promise for these companies of artificial intelligence agents is precisely that you could go beyond just offering a tool for these industries, but instead offer the workers, the labor force for these industries. At which point, doesn't matter who owns the hospital if Google is running all the nurse agents and doctor agents and radiologist agents and whatever the case might be. You know, at a certain point, it's all Google, for instance. And so even in the conglomerate strategy, in that case, you have an extension of control without ownership. Well, actually, I think in this case, it still remains more and more about ownership because it's these agents owned by these platforms, which are effectively acting as the labor force. You know, that's where these companies are aiming for. Can you just to concretize that briefly? And we've described what an agent is, but can you give an example of what a particular agent might be and what it might do? Yeah. So we have two sort of clear examples of agents right now. One is coding agents where you tell a coding agent that you want to build something, to build a website or an app or something. And it goes out and it takes time and it thinks through the process. And then it carries out that process step by step by step. And different agents will have more or less human oversight. So they might return to you and say, oh, I'm not sure what to do here. What do you think? And then you sort of give them a response. Or they might just go and do the thing and then come back at the end with, you know, a finished product to show you. So that's an agent. It makes a plan. It carries it out. It takes a lot of time. The other example is the sort of deep research things that are available through most of the main models now where you ask, go and research this particular topic. And again, it sets out a plan for how do I find sources on this material? How do I organize all this material? How do I think about salient points? And then how do I present it all? And it takes dozens of minutes for that sort of thing to get carried out. The idea of the future of AI agents is that this stuff will continue to improve. It will be able to work longer periods of time, which is a crucial sort of benchmark. It will be able to work in different fields beyond just coding, which is really optimized because of that reinforcement learning stuff. But also beyond just deep research and instead work in these very specific industries. And then thirdly, they start to work collaboratively. So it's not just that you have, for instance, one big AI agent running a hospital, but instead you've got a team of different agents that are doing different things, interacting together and functioning as, you know, a sort of organization. So that's, you know, that's what the AI company is trying to build at the moment is. All of this sort of stuff to, to again, to get AI to work better in the world and ultimately to be able to automate people. Just to linger on the, the, the, the labor question a little bit further, is it possible that the, that the outcome is not that, and I'm thinking of Corey Doctorer's arguments here that AI doesn't effectively replace workers, but replaces some of them while making other people's jobs way crappier. And in the process produces a much crappier service or, or product. I think that's absolutely case. I think, it's, automation is rarely clean. It's never like a very rarely, I shouldn't say never, very rarely the case that a technology simply replaces a job wholesale. Instead what tends to happen is that technology improves productivity in certain parts of a job and that job sort of gets redefined and it may turn out that you don't need as many people doing that job as was once the case. So it's a complicated process, but the, the sort of general thrust is yeah, fewer workers in a particular position and the remaining workers are under more pressure to be productive and to work according to the machine rather than, you know, their, their own sort of normal rhythms. In terms of, you know, Doctor sort of inshidification thesis, I think this is a really, a really interesting point because I think one highly understated reason why people like using chatbots at the moment is because it's not inshidified yet. You know, it's, you can go there and get a really clear answer to your question in the way that Google promised to do and sort of did back in the day. And the fact that these ways of accessing information have become so inshidified and so incapable of giving you a direct answer to the question that you might have is partly the, the novelty of these chatbots of these AI systems and why I think people are so attracted to them. I mean, it's part of the reason why I don't mind using them because this, for certain questions, you just ask it and you get a nice clear direct answer. No ads, no nothing. So you don't have people, for instance, doing 10 minute YouTube videos to answer a one-line question. just so they get advertising dollars and just hears the answer. So I think we're at a moment where AI hasn't been inshidified yet, but it probably will be in the future. And you know, it very likely will end up getting worse over the next few years as especially companies like OpenAI become more and more desperate to meet their spending commitments. >> Well, and then in the case of the app Sora, it's dedicated to making a slap. It's sort of inshidified on principle as it's premise. >> Yeah, there's this, it's a fascinating thesis. And I think horrifying thesis in many ways from this, this analyst Ben Thompson, who has sort of pointed out that, you know, what are the sort of radical ways in which meta could benefit from AI? And there's the obvious ways in which they use artificial intelligence to rapidly spin up highly customized advertisements for people that are, you know, more persuasive than what they are now. So that's the obvious way in which meta can benefit from using AI internally. But the sort of thesis that he has is that actually, you might end up moving from this period of social media, where your newsfeed was filled with people that you knew, people that you were close with and had some connection with. Increasingly today, our news feeds are filled with, not by people that we know, but instead by recommendation algorithms that are feeding us what we think, what they think we want to see. And the sort of next step of that is like, well, what if AI just generated all of this stuff? And what if it wasn't humans who was doing it? So you move from this period of like news feeds being highly social to now just being highly individualistic and highly personalized. And, you know, that's what I think is sort of implicit behind Sora. And it's, I think explicit, increasingly explicit in the ways that meta is using AI as well. Well, yeah, I think a horrifying vision. But yeah. Yeah, I mean Zuckerberg earlier this year said that the average American has quote, three people they would consider friends. And the average person has demand for meaningfully more. I think it's like 15. And his idea is that, well, AI agents can be your friends. Yeah, I mean. You're welcome. (laughing) I think especially when it comes to him, there's such a visceral reaction against that idea. And it is sort of horrifying in many ways. But I do think as well that there should be some sort of nuance conversation about AI companions, especially given how I think influential they already are. And how influential I think they're going to be for generations growing up right now. The sort of knee jerk just denunciation of it all. I don't think is sufficient to address what is actually going on here. So I think, yeah, Zuckerberg's saying that is horrible. But there's other things that need to be tackled in this area too, I think. I want to return to the financing in a minute, but I want to push a little deeper here. Because I certainly feel a visceral, I guess abolitionist instinct towards AI being our friends. I guess given that AI is already becoming, people's friends complicate that position. But why does one need a more nuanced perspective on something like that? I think because, so for instance, when my wife Helen Hester and I were researching domestic technologies in the house, and one of the examples that often gets brought up and immediately denounced is care bots in various ways. And when we went and sort of researched it and looked at people who had interacted with them, there was a genuine benefit to be had from these care bots. I think that needs to be addressed, like pointing to the ways in which society produces loneliness at the moment is I think totally true, totally necessary. But there's still something more going on here. It's not just a symptom of a decrepit society. It is, my suggestion would be that even in a society that was far less lonely and far more communal, people would still turn to these things to have as potential friends. You know, whether there's sort of fantasy scenarios of some sort or whether, you know, it's just meeting needs that aren't being met in their local community. There's something that requires like a proper ethical, philosophical appraisal rather than just denunciation. So I don't have the answer by any means, but I think that there has to be a more subtle take on it than, yeah, just saying, this is horrible. But it does seem, it does seem analytically significant that so much of the disintegration of the social fabric seems to have been accomplished by tech, specifically various forms of social media, not all of it, but it seems to be a driving force to have the same industry offering new products as the solution to that. Yeah, it's, I mean, that's appalling in many, many ways. You know, produce the problem and sell the solution. But yeah, I think that there's, you know, there are people doing research on this stuff in ways that I think are reflect the sort of actual complexities of the case. And it is, you know, it's quite a novel circumstance to have a completely artificial companion that can talk to you and respond to you in meaningful ways. You know, I can't really think of any other sort of historical precedent for that sort of thing. So it's a unique social phenomenon, and I think it's only going to become more prominent. And we as a society sort of need to think about how should we handle this stuff? You know, the same way we didn't give enough consideration to what social media was doing to us. We need to be giving serious thought to what AI companions are going to do, because I don't think they're going to go away. - Right. I mean, this all returns to the basic problem, which is that under capitalism in general and oligopolies in particular, you have a small number of people making giant decisions about the future of humanity. - Exactly. Yeah. Whether you want to do not, this is what you're getting. - Returning to the finance. Where does the rest of the money outside of these weird circular arrangements we were discussing, where does it come from? How much is coming from various forms of debt, private credit? Also, when it comes to the hyperscalers, like Amazon, that earned giant revenues from their actual profitable business lines investment from straight up cash. What does the rest of the financial picture look like? - Yes. So this has been like a rapidly developing part of the AI ecosystem, I'd say. And it's sort of the, it's one area where if I had to correct my book, I would make a massive change here, because I write at one point in the book that I don't think the AI bubble is all that significant because the hyperscalers are funding this capital expenditure out of their own balance sheets, you know? They make immense amounts of profit. They have loads of cash, and they can afford to spend all of this money on capital expenditure. And if it happens that the bubble pops, it hurts their balance sheets, and they might get knocked back in the stock markets for a little while, but it's not some massive catastrophic bubble bursting. Since I wrote that, there's been about $200 billion, I think, of corporate debt that has been accumulated over the past like six or eight months or so. And that is the massive change over the course of 2025, is that the capital expenditure is now reaching the point that even the biggest companies in the world making vast amounts of money, hand over fist, are not making quite enough to be able to afford all this, this capital expenditure stuff. So they've had to turn to debt markets. Primarily it's been bond markets, so selling off corporate bonds, and there's been a huge uptick in corporate bonds from the big tech companies this year. But the other thing, which is far more worrying, is the rise of these sort of special purpose vehicles, which are effectively designed to keep the debt off the balance sheet of these major tech companies. - Right, they show up as operational costs 'cause these tech companies are just, quote unquote, renting them, and they've basically given the risk to someone else or someone's else. - Yeah, exactly. So they're designed in this legal, technical manner in such a way that it doesn't count as debt on their balance sheets. But at the same time, they are written in such a way that those big tech companies are effectively guaranteeing the debt, so that if for whatever reason, something some partner along the line can't make their payments, the big tech companies will step in and make those payments. So if you're buying the debt of these special purpose vehicles, you're effectively buying the debt of these big tech companies, extremely safe, you know, they're going to meet their debts. Microsoft has a better credit rating than America. But at the same time, you're also getting a higher yield, you're getting more money than what you would if you bought a traditional corporate bond. Like 10%? Yeah, yeah. So there's, there's quite significantly higher yields in some cases. And that's the sort of allure for investors. And for the big tech companies, it's a way to, to keep their credit ratings because it doesn't show up as debt on their balance sheets. But there's a weird thing. And I, I mean, I don't know the financial space well enough, but like the ratings agencies must also be aware of this and sort of just be like, well, we just won't focus on that. We'll pretend that debt doesn't exist. You know, everybody's sort of in the now on this. But it doesn't count for the credit ratings agencies. Right. Well, it's sort of like too big to fail until it does, which we'll get into more at the bottom of the interview. I want to turn to a core subject of your book, which is the intensifying geo economic conflict with with China and how AI ended up at the center of that. The US over the past decade, increasingly pursued an aggressive strategy of geo economic conflict. That aims above all else to block China's technological and economic advances on AI and high end semiconductors. And that's all you write rooted in deep economic changes, shifting fractions of the capitalist class and changing relationships between the state and quote, the leading sector of contemporary capitalism, which of course is tech capital. And specifically, you argue that there's been a broader hegemonic shift from a neoliberal pro free trade order that dominated the US until the mid 2010s, which you call the Silicon Valley consensus, a shift from that to, to an emergent, still not hegemonic order guided by nationalist geo politics and geo economics and you write, quote, here, there was a broad agreement across the political elite and tech elite about the role of technology in the world. About what was required in order to allow that technology to flourish about what purportedly American values they embodied and about the requirements for capital accumulation in the technology sector. Before we get into why this went into crisis and what has emergently been contending to replace it, let's describe what it was. What what was the Silicon Valley consensus on an ideological level and and then what material form? Did that consensus take in terms of US political economy, specifically the relationship between the sector and the state that allowed the tech sector to grow to such gargantuan size and wealth? I think the Silicon Valley consensus has been even if only implicitly fairly well recognized for a number of years now. And I think Obama sort of the peak period of this, but it really goes back to 1990s and arguably even earlier. The core sort of idea here is that particularly the internet and the technologies and companies which have emerged in the wake of the internet and the worldwide web and everything that's come since then. The development of those technologies are global that they bring along a sort of liberal idea of a global community of people who can interact on these digital services, whether that be socially politically, economically. And that for the American state, all of this was beneficial because it was American companies in the driver's seat. It was American companies which owned the commanding heights of the digital economy and which were rolling out their infrastructure around the rest of the world. And this had benefits in terms of American capitalism, obviously massive amounts of money flowing to America and Silicon Valley more specifically. But it also had virtues in terms of, as people like Henry Forell have pointed out, virtues in terms of the ability to surveil other countries, the ability to weaponize this infrastructure in such a way that if a country or a group of people sort of went against America, you could weaponize this technology to eject them from that system or coerce them to change their behaviors through the use of these. Global technological infrastructures. So on both points, you know, from the sort of Washington DC perspective, this was great for American strategy and from the Silicon Valley capitalist perspective, it was great for capitalism as well. And that I think was the sort of Silicon Valley consensus that Washington said effectively to Silicon Valley, go and grow as big as you want, go around the world, we will support you. We will not regulate you, we will not hinder you. You are serving our interests as you serve your own interests. So that was the consensus there. Before we move on, I want to, I want to ask just methodologically and theoretically about about why you employ this sort of hegemonic analysis, why, why is this sort of Gramscian framework key to understanding what's actually happening here at these sort of like deep tectonic levels of of political economy to to clarify what's what's otherwise obscured and more superficial or partial assessments. Yeah, it's a really good question. I think, I mean, on this particular issue, I've been quite influenced by Jeremy Gilbert in my other co author Alex William. So they've written a book on this topic and focused on Silicon Valley and applying a Gramscian analysis to it. But in sort of broader strokes, more theoretical strokes, the sort of hinge point here for me is that capitalism as a system that structures particular incentives and demands particular behaviors on behalf of its economic actors and social actors and political actors. That system. At an abstract level still needs to be more concretized and that doesn't mean as a sort of mainstream economics would do that sort of micro foundations and individuals. But instead thinking about significant groupings of distinct interests of capitalist. And for me, that then lends itself to this, you know, the Gramscian point and the ways in which you know, I've always found it a really fascinating question as precisely the Gramscian question of how can a small group of people lead over a massive complex society. And understanding these different capitalist groupings, which you know, they're rarely formalized as such, even if they are embodied in trade associations and industry groups and lobbying associations and things like that. They're rarely formally articulated as such and it's sort of a process of empirical deduction to say, well, here is a core group and here's their core interests. And then as I try to argue in the book, there's an emerging group of people with a distinct set of capitalist interests that are, you know, they're not the neoliberal ones. There are far more national security, economic nationalism sort of perspective. But I think this is the sort of analysis of like what you might call a sort of me so level of capitalism. I think it gets to some of the big important changes in capitalism and helps us to understand where things might be going, why things have gone the way they have gone over the past, you know, particularly since 2016 in the US. But then also understanding, I think, really importantly, the internal tensions of the enemy, you know, I think far too often, the the tech right especially is presented as a unified entity with a relatively homogenous set of interests that is more or less all powerful. And in fact, I think is far more imported and useful strategically to understand the tensions and understand the ways in which this grouping is at odds with itself on quite key issues. So that's again, I think the Gramscian analysis really lends itself to understanding the formation of a consensus and the ways in which that is always a sort of ongoing negotiation. And China had had its own consensus to a quarrelary to the the Silicon Valley consensus and you write that what prevailed was quote, broadly similar harmony between the Chinese state and its domestic tech giants. As in the Silicon Valley consensus economic interests ruled over national security interests and importantly, it was a consensus in pursuit of coordinated national development, but not at the time toward high tech self-sufficiency self-sufficiency wasn't a goal. You write until the US launched its trade war and strategy of due economic containment against China China up till the that moment, it was committed to the free trade system right up to the moment that the US blew it up. But let's not get ahead of ourselves. What was the Chinese corollary to the Silicon Valley consensus? How did it operate? What sort of technological advancement did it support? And what sort of a role of the tech sector in Chinese society economy politics did it promote? So in the Chinese case, I think it's a similar consensus, but it's reflective of the different environment that's involved. So if America is this leading frontier economy that's global and all dominating, China by contrast is still a middle power, a middle income economy that is struggling with all the challenges of economic development in the 21st century. And so the sort of consensus that emerges is one that first and foremost is based around economic growth and fostering economic growth in the domestic economy. There's not, as you get to America, there's not as much concern about that sort of international expansion of these tech giants. They certainly have those ambitions and they carry them out in various ways. But it's not the primary focus and it's not what the state is most concerned with. It's particularly after the 2008 crisis when the sort of previous growth model of the Chinese state was halted and you know, had been stumbling along in some provinces for a little while at that point. And there was a gradual turn over the course of the 2010s and I think really solidified in the May and China 2025 document that in fact what was going to be the driver of the Chinese economy was innovation and it was going to be these high tech sectors. So gradually you start to see the Chinese government really focus on that with all the sort of industrial policies that they have available, all the sort of regulatory allowances that they give these companies. And then the companies, the platform giants are basically given free reign to sort of go and do what they want within the Chinese economy. And the second sort of key function I think of these companies at the time was also about providing a sort of infrastructure to knit together particularly the rural elements of China into a cohesive national economy. So especially as e-commerce is starting to arise, there's been some fantastic work on this I think is a book called Click to Boom or something like that. But it's all about the ways in which this infrastructure also provided for that sort of integration of a domestic economy through that infrastructure, you know, instead of local markets or something like that instead everybody could go on to Alibaba and JD and whatever and get their goods. So yeah, there's these two key elements I think which they're all about developing the economy, developing the country, moving it up, you know, that sort of development ladder. And yeah, I think it worked quite successfully for a number of years in the Chinese case. I'm Naomi Klein and you're listening to The Dig, my go to podcast for the most thoughtful in depth conversation on the left. It's an incredible place to be exposed to new ideas and new writing and if you can please become a sustaining supporter at Patreon. This episode of The Dig is brought to you by our listeners who support us at Patreon.com and by Equator, a new magazine of politics and ideas. The ongoing moral catastrophe in Gaza shattered the moral authority of the Western establishment and its media. Equator believes that the end of the West is not the end of the world. Indeed, we're in a moment when new political horizons can be glimpsed and radical forms of action and imagination are possible. Equator.org is launching a new cultural platform which challenges the narratives that equate Western interests with universal truth. Instead, we're building a future where equality and justice define the global conversation. Here's include dig guests like Naomi Klein, Rachmane Dresa, Adon Gattacho, Nikhil Paul Singh and many others. But Equator is more than a magazine. It's a rallying point for a global community built through reading groups, online seminars, film screenings, workshops and events. I've been reading Equator and I love it. This work needs your support. Join the community. Digital memberships start at just $6 a month. Find the essential voices you've been missing at equator.org. In the US, the Silicon Valley consensus coming apart, it marked or caused a crisis in the hegemonic order, which in turn precipitated this shift to a more conflictual relationship with China. This political shift was not driven first and foremost by new ideas and think tanks or new instincts of presidents. But by crises and changes in the deep structure of the global economy that then rippled up through American political economy and into foreign and trade politics, how as you argue and you're drawing here on Ho Fung Hung, who I interviewed a while back on the podcast, how was it that secular stagnation across the entirety of the global economy alongside China's attempts to move up the value chain, up the development ladder into foreign markets in the wake of the 2008 economic crisis. How did that all upset the dynamics of what had been the hegemonic coalition governing US political economy? So I think a big hinge here is the relationship of America to China and the perception that America has of China. And drawing on Ho Fung Hung's work, as you say, for me, the real key part of the story is the ways in which American capitalists were the drivers of this pushed integrate China into a liberal world economy with the hope from some think tanks and politicians and whatnot that as China integrated into the global economy, politically they would also liberalize and they would become a sort of non-threatening great power and America could continue its dominating role. So that's the sort of, I think, a key element of the story here is that it was US capitalist interest, which won out against the interest of the American national security wing because they were filled with China hawks and always thought China was something that needed to be feared and needed to prevent their capacity to grow and develop into a real potential threat to America. So the capitalist interest won out and what happens over the course of the 2010s in particular is that those capitalists start to lose a sense of the benefits of integrating with China, whether it be through inability to access the Chinese market. So Google, for instance, pulling back its services from the Chinese market, Facebook unable to operate there, things like that. And then also the sort of rising competition of Chinese companies in these spaces that have been dominated by American companies. So e-commerce is now facing with Xi'an and Timo. So these sort of Chinese competitors moving into spaces that have been dominated by American companies and the American capitalist increasingly seeing China not as a potential market that they could benefit from of cheap and educated labor, vast amount of consumers to sell to, and instead started seeing it as a potential threat and as something which was competitively going to undermine them. So that shift in capitalist perceptions then means that the US national security wing can start to push their Chinese perceptions much, much more. And I think you see this starting to emerge under Obama. I think the pivot to Asia is maybe the first major sign of this. You've got this sort of, you know, the turn away from this rigid focus of the war on terror throughout the 2000s and suddenly saying, no, actually our strategic focus is now going to be on the Pacific and on Asia. So there's that element. And then you've also got, I think it's in 2013, the first national security investigation of Huawei. And a little bit later, you've got Obama using this legal tool for the first time to block the Chinese acquisition of an American semiconductor company, which could be a story from today. In fact, it was more than 10 years. years ago. So you've got the inklings of everything that we're seeing today starting to emerge in those early 2010s. Trump then comes along and with his trade war really just sort of blows it open. Part of what I really wanted to do in this book was to write a story that didn't depend on who is the president, especially because I did most of my writing before I knew who was going to win between Harrison Trump. So there was this real constraint that I have to write a story which is about structural elements rather than the personal idiosyncrasies of Trump or whatever the president might be. So yeah, this is a story which goes across administrations, goes across parties, it reflects deep economic changes and changes in the coalitions and perceptions of key geopolitical enemies and then that feeds into changing perceptions of the international order. So that old order becomes destabilized and then there's this emerging US order which which again has not yet achieved real hegemony is defined by national and economic security. It's exemplified and in some ideological sense perhaps led by a techno nationalist fraction of Silicon Valley capital that's that's embracing increasingly hard right politics hostile to China abroad to the administrative state at home and these are military hard tech firms like Andrew in the Gundo, Nathsack, DataFirm, Palantir, far right accelerationist venture capitalist Peter Teal and Mark and Dreson and then even fractions of tech capital that retain a strong interest in orientation toward trade and openness with China figures like Bezos Zuckerberg, Jensen Huang, they're increasingly aligned with Trump and that really did pay off recently for Huang and Nvidia which will get to more in a little bit. But meanwhile Musk who of course has had this strange very important if ultimately very messy alliance with Trump, Musk's own relationship to these politics are materially very messy because Tesla profits are dependent upon China in a big way even as SpaceX seeks military industrial revenues in the same way that a company like Andrew does. How do all of these various fractions of tech capital and their emerging interests and agendas, interest in agendas that post contradictions not only between fractions but within them. How does that all constitute the contested emergent block seeking hegemonic power over the United States? Yeah and this is why it's so complex at the moment. You know one of the things I really wanted to try and argue was that the seeming contradictions and you know the flip flops and everything that go on with Trump's strategy particularly in foreign policy and economic security, all the ways in which US strategy is constantly changing and seemingly going back on itself at certain points in time is not simply the reflection of a sort of senile old man but in fact is a sort of objective reflection of objective conflicts of interest amongst the hegemonic blocks in America that the ruling class that the Silicon Valley consensus doesn't hold anymore and that there is an ongoing competition at the moment amongst different fractions of the ruling classes to push for their vision of US strategy, their vision of the world onto the rest of you know onto the rest of the world. And so yeah you have I think still although I do think it's starting to disappear, you have this sort of neoliberal big tech wing embodied by companies like Amazon, Microsoft, Google and Meta, and the export-oriented companies like Nvidia I should say and the semiconductor industry in America more broadly, all of those sort of companies their economic interests are aligned with traditional neoliberal globalization, they've benefited from it in the past and they will continue to benefit from it massively and all this stuff around you know export controls and tariffs and economic warfare, all of that is a real hindrance for them, they don't want to have to be dealing with it. By contrast you do have this emerging fraction of techno-nationalists as a sort of shorthand way to to call them that are embodied first and foremost by the defense tech wing, so your talent is your andrels as you say, backed up increasingly by venture capital which is a highly concentrated industry, so whatever the sort of leading venture capitalists say is where much of the industry ends up going and they now see defense tech as a massive growth market. And so for them you know irregardless and political beliefs, it's a massive growth market so they're chasing investment there. And just a note there, Anthropic Google Meta Open AI were all opposed to applying AI to military purposes as recently as the beginning of 2024 and boy has that changed. Yeah it changed over the course of a single year and it was it was extremely rapid and I think really telling that actually two of the changes Anthropic and Meta came in the same week of the US elections. They were news stories which you know unless you're following the stuff you may not have seen it but yeah in the week the Trump got reelected. Anthropic partnered with Andrew I think and Meta said oh yeah the US government can use our AI models for military purposes. So there's a massive change and you know there's been ongoing integration of the big tech companies into the US military which has been a long process since about the 2010s. I think Amazon was the first one but it's been a sort of growing process with notable pushback particularly in the case of Google against a project maven which workers managed to successfully push against Google's participation in that project. That's all changed now though and those sort of protests just they can't get off the ground in the way that they did back in the day. And in fact one of the more sort of frightening things I read in my research was that even the workers themselves are starting to look back on that sort of anti-military stage of the tech worker scene and just say that that was anti patriotic. It was anti-American and in fact you need to be working for military companies to serve this sort of patriotic purpose today. So there's a real culture shift ongoing in the tech worker scene as well which is worrying because I think the tech worker scene is one of the few points of leverage that we actually have over these companies and to lose that I think would be a real loss for us. And in this sense the defense tech firms, the Andrels, the Palantirs, they might be in terms of size and revenue etc. are far smaller than big tech firms like Alphabet or Amazon but through an agromsian lens they can be playing a leading ideological role as a fraction in attempting to make this new hegemonic block a reality. Yeah exactly and I think one of the more telling elements the sort of self reflective awareness of this group is the fact that they have been forming a variety of industry associations and there's been a number of the defense tech companies which have allied together to take on major military contracts against the defense primes, you know the major defense companies of the past. So they do see themselves as a collective group with a shared set of interest and they are lobbying for it, they are pushing for it, they are fighting for defense contracts and yeah you know people like Teal are obviously playing an ideological role as well. So yeah it's an ongoing struggle I think at the moment and it remains to this day unclear where things are going to go, you know will the sort of techno nationalist wing win out or will a sort of digital neoliberalism win out. I think it's still up for grabs but yeah it's you know one of the main claims I want to make in that sort of discussion in the book is simply that there is the struggle going on that the ruling classes in America are not actually cohesive at the moment that there is surprisingly vocal criticism from both sides when the government doesn't do what they want them to do. So there is a sort of awareness by both of them that like there is a wrestle for power at the moment. Yeah and again we keep referring to it but in video score to big W for tech neo liberalism recently that that we'll get into but but you do point to some evidence historical evidence with some strong conceptual support for why the military industrial tech right might have an advantage or might have kind of momentum behind them, you write that history shows that the military is generally very helpful, maybe uniquely helpful in moving general-purpose technologies toward diffusion and adoption because the federal government, when it comes to militarism and repression, it will spend a ton of money over very long periods of time. And it's not a new dynamic. You write, quote, "such state capitalist relationships have been central to earlier formations of imperialism, Lenin famously characterized the imperialism of his era as a merger between monopoly capital and great powers, and they remained influential throughout the 20th century." And in this case, the prospect that this fascist authoritarian project's expansion of repression, whether we're talking about mass deportation or the buildup of the US military, the prospect of this creating a material basis to bind the most powerful capitalists in the world to that project, that seems very chilling. It does. I think it's an absolutely terrifying prospect, and I you know, I never wanted to say that we might pine for neoliberalism to come back, but it is terrifying the sort of prospect of this emerging hegemonic structure. And I think precisely because of the systemic logic of how it needs to build that consensus, the outcomes are terrible for the vast majority of people around the world and the vast majority of people in America as well. So yeah, it's something, you know, I think by trying to distinguish between these different fractions of the ruling class, you know, I want to also be able to say that like, there's angles to play, there's strategies that are maybe available that aren't available if you just think all the tech companies are the same and all of them are part of the tech right. And I don't think that that is actually the case. And I think that there are ways in which this sort of emerging militaristic technonatialist coalition can be broken apart. So having a clear understanding of the enemy is, I think just yeah, vitally important in this case. Yeah, to further clarify our understanding there, we've been talking a lot about economic factors, but there's powerful ideological work underway, notably the widespread reaction among tech elites against woke, particularly their woke workers who dared to challenge the progatives of executive and founder power. And you mentioned the very specific, very costly to the bottom line case of Google canceling its participation in project maven with the Pentagon. You write quote, much in the same way as communism acted as a unifying term for the post war right. So too does woke act as an organizing bogeyman today. The term also enables the techno nationalist right to find alliance with openly reactionary ideas ranging from eugenics to climate change denial to race science to gender essentialism to Christian nationalism and more. Is is anti woke in a sense the glue provisionally reconciling all of the contradictions between all of these fractions of tech capital reconciling vast contradictions if only provisionally to stitch together this far right power block contending for hegemony. I think it's certainly an ideological element of it and a key ideological element. But I think the sort of more consolidating element has to do with the sort of material flows of money. And you know you mentioned that historically general purpose technologies have relied upon the military as a customer in order to get off the ground and to get going. And I think a similar thing is happening today, you know, open AI is desperate for money because unlike Google for instance, they don't have an arm of their business which is just a money printing machine. So they are desperate to find money and to do whatever it takes to get money. Same thing I think in a more sort of curious case is anthropic which is explicitly the most focused of all the major AI labs, the most focused on AI safety and doing the sort of ethical correct thing into the developing of AI. But their CEO as Kometan said, we're in a civilizational sort of war between democratic AI in the one hand led by America and authoritarian AI in the other hand led by China. And that there is this binary world order that needs to be, needs to have democracies win. And on top of that, they're now, you know, there is desperate for money as open AI and now going to the Gulf States to get money. And there are statements from the CEO where you can sort of sense the reluctance that he knows morally, this goes against his business, but the reality, the economic reality is such that he has to chase these dollars. So there's that sort of integration of all of these different elements of the ruling classes together. And that I think is the most sort of frightening element, venture capital seeking to fund defense tech, defense tech, desperate for money from the government, the government seeking increasingly anti-democratic systems wherever it possibly can, and increasingly saber rattling with China and even Canada in the case of Trump. So all of this, you know, just is feeding into a system where, you know, it's quite terrifying that it's a system where more money flows into defense and into the military and that feeds the ideologies of economic nationalism and geopolitical conflict. And all of these companies getting more and more integrated into the system, more and more dependent on it and dependent on its continued survival and its expansion. And that's historically been a recipe for disaster. Before we go deeper into the US political economic situation, let's catch up with China, where you write that the shift in hegemonic order was from prioritizing high tech as part of a larger developmental strategy to move up the value chain within a globalized economy, a globalized economy defined by free trade, a shift from there to a new order that above all else, prioritizes developing high technology so as to eliminate all dependency upon a hostile United States. And a signal moment in this transition you write was the 2020 2021 crackdown on Chinese tech companies that followed Alibaba's Jack Ma making publicly critical comments of the government. And it was sort of interpreted in the media that just sort of like, wow, Ma really fucked up and now he's in deep trouble. If I remember the contemporaneous coverage correctly, but you argue that the crackdown reflected far deeper shifts that were underway. What did this crackdown reveal about about the Chinese states emerging program to discipline and remake capital, remake tech capital toward particular new ends? Yeah, so I think the seeds of this sort of emerge in 2018 after new sanctions put on Huawei and some other major Chinese companies. And if I remember correctly at the time, it was seen as a sort of existential threat for Huawei. You know, it wasn't clear that they were going to survive as a company, the way in which America was cutting them off from all of these economic supply chains and the businesses that they'd built up over time. And there's a fascinating article called Driven to Self-Reliance. It's an international studies quarterly. And it shows quite clearly the ways in which the Chinese state, when it saw these actions from the American government, increasingly took self-reliance as a primary goal. There's a long history of self-reliance within Chinese history, you know, from Mao onwards. There's lots of talk about technological self-reliance and things like that. But there hadn't been this intense focus that you get, start to get in 2018 onwards. And that is quite clearly, as this paper shows, the result of American actions in China reacting to it. And what I think happened in 2020, and I should say nobody knows for sure, my theory, though, is that what was partly going on here was the Chinese state increasingly frustrated with the platform giants that it had allowed to flourish over the 2010s, increasingly frustrated that these companies and a variety of other companies were focused on sort of frivolous things. And they weren't focused enough on that increasing pressure to self-reliance technologically. And so the Jack Ma comments are just sort of the spark for what had been an ongoing change. And what I think is really curious, and I don't think it's been noted enough in discussion of this crackdown, is that it wasn't a tech crackdown. Because lots of tech companies escaped, unscathed, including Huawei, and including the lithography companies in China, and the semiconductor manufacturing companies in China. None of them had any sort of fines or investigations or anything. It was only the platform companies which saw this massive investigations and massive fines thrown out, and certain industries completely demolished. It was a crackdown on doing frivolous shit that wasn't advancing the national interest. Yes, yeah, exactly, in a nutshell. I think you get this, again, a sort of similar to the way in which in America, the consensus breaks down, you start to have a fracturing of the tech elite. In China, I think you can see a similar scenario, which is being patched up now, that the companies like Huawei who'd been more focused on following what the government wanted directly, less interested, although not uninterested in chasing profit, but less interested than say Alibaba was. They get risen up in the wake of the platform crackdown, and the platform companies get pushed down a level in terms of the hierarchy. Now with AI coming along, Alibaba and Bike Dance and Tencent to a lesser degree, they're now taking on this real focus on tech self-reliance, and participating in that primary aim of the Chinese state. Another thing I want to mention, though, just very briefly, is that there's often, I think, an assumption that what the Chinese state dictates is what the capitalist then follow, but it's not actually the case, especially given the power of the Chinese capitalists. In the same way that Western countries are dependent on their capitalists for economic growth and things, China is also dependent on their cap was for economic growth. It's not a one-way system of power. There is mutual dependency here, even if the asymmetry of power is quite a bit more towards the state in the Chinese case. Right. In one case, executives can be executed for. Economically, the emerging hegemonic order in the US has been defined by the state imposing a ton of export controls on AI-related technology, including on exports from third countries, containing even the smallest amounts of US tech. So this is particularly important in terms of the US constraining the Netherlands, the sole country manufacturing, the extreme ultraviolet lithography machines that alone make the highest end chips, and also Taiwan, where TSMC makes those chips. But in recent months, things have changed pretty dramatically. In response to Trump's continuous escalation of the trade war, China responded with a powerful trade weapon of its own, tightly restricting the export of rare earths, an astounding demonstration that China can indeed stand up and protect itself against US bullying by imposing real economic pain. Trump folded, dialing down the trade war, and most recently, he just lifted export bands on Nvidia's second most powerful GPUs, the H200. And this was an export control. This was an export control scene as a key tool for geo-economically containing China. What's going on here? Was it just that G demonstrated to Trump that China was playing a much better hand economically than the US had let itself to believe? Or does this also have to do with these contending fractions of of tech capital, of American tech capital? Because in video, of course, they want to sell the maximum number of chips to the maximum number of people everywhere, very much including to China, was this more G-beating Trump or more in video winning out, or some of both? And what is it all reveal about this emerging still very highly contested hegemonic order and the balance of power between these various fractions of tech capital? Yeah, I think it's a bit of both, as you say. I think it's a bit of Nvidia and Jensen Wong very much wooing Trump. And it's also a bit of China for the first time really able to show their hand. And I think that they've had the capacity to do this sort of thing in the past. But quite notably in the past, the reactions to the trade war were pretty much proportional. If the US did one thing, China would do a similar sort of thing in response. The rare earth step was a major step. And from what I can gather, particularly affected the automobile industry in America. Real concerns from the CEOs of all the major auto companies that this was going to have a highly negative impacts on them. So there was that sort of element that various capital's interests in America were going to be directly harmed by it. And we're calling up Trump and saying, you need to do something about this. And then Nvidia, yeah, also lobbying extremely hard to to be able to sell to the Chinese market, which at one point, they were reported to have 95% of the market. And today, they effectively sell nothing to that market. So, you know, billions and billions of dollars for Nvidia at play there. And I forgot to mention that Trump orchestrated this crazy part of the deal where the US government gets a cut of every Nvidia sale of these chips to China. Yeah, wish. From what I can gather, nobody's quite clear that that's legally admissible. But yeah, it's very much a Trump sort of specialties to add in these deal things to the yet to that thing. But yeah, it's interesting. I think in terms of the sort of battle between different elements of the tech elite, seeing the reaction to Trump's decisions about first of all, the age 20 chip, which was Nvidia's chip, which they designed specifically to meet the Biden criteria that it wasn't too powerful to be sold to China. It just sort of went underneath those limits. And the age 20 chip, they weren't able to sell for a while and then Trump changed his mind and said, you can sell it. And this is definitely a Nvidia lobbying. But the part of the interesting element here is that the age 20 chip is in terms of training AI models is not all that powerful. But where it's actually really quite good was in running AI models. So if you're concerned about diffusion and adoption of AI, actually controlling the age 20 was quite an important step. So yeah, lots of the techno nationalist wing critiquing the age 20. And then more recently, the decision around the age 200 to allow it to be sold to China, which hadn't been the case. But now appears as though it will be the case. One thing I will say that's changed though, is that Nvidia's next generationships are going to be coming out relatively soon, and that they will be quite a bit better than what's being sold to China. So I think there's a case to be made that the Trump administration may have been wooed by the fact that these chips are going to be relatively obsolete by American standards in a short time period. So selling them is sort of a way to you know, as I think Howard Lutnik put it, get China addicted to the American tech stack. It's reflective of that sort of neoliberal vision of the tech world of having the American AI stack spread everywhere, including to enemies, simply because having that control is more important than giving them access to this technology. And the irony here is that it's Nvidia and China reimposing these neoliberal norms through China unveiling and deploying a powerful geo economic weapon. Yeah, yeah, exactly. Under under Biden, geo economic strategy was this multi faceted effort to advance US innovation and market share while while containing China, not just on AI, but across a range of so-called technologies of the future and those technologies supply chains. So not just AI, but also clean energy, battery technology, et cetera. Trump, of course, quickly gutted federal investment in renewables and and doing so really winnowed the US's geo economic agenda down to AI plus critical minerals. Meanwhile, as Tim Wu writes in a recent FTSA, China is investing way, way more in batteries, EVs and solar panels than it is in AI. So unlike the US, which is betting everything on on the pursuit of AGI, China sees AI as just one piece of a much larger puzzle is the US in general, but very much specifically under Trump with his gutting of the IRA, just fundamentally misunderstanding what technologies will define the future of the global economy due to these rigid ideological commitments of the Magah far right. And then also more specifically in terms of AI, are they misunderstanding the role that AI can play in economic development, whereas China sees it as part of this broader industrial strategy, US tech by comparison and US tech policy by comparison seems to figure AI is more of like an economic miracle unto itself. Yeah, I think very much that the US approach is AGI or BUST. It's an attempt to try and build this speculative technology as quickly as possible to be the first to get there and to sort of reap the benefits from building this technology and controlling this technology. And by contrast, as you say, is very much focused on the diffusion of this technology as it exists today, the diffusion of it throughout all these different industries. And there's loads of different efforts being made in China across different industries to try and figure out how can AI help improve the productivity of the sector and boost the economy. So it's a very different approach and like you mentioned, it's much more diversified approach in terms of investment in future technologies. I think again, this is partly a reflection of that crisis of the American ruling classes that you don't have a single definitive tech elite that can say, we need solar power, we need it now and you should be funding it to the Hilt to get it. You instead have a vocal Magonashelist who are scaring off foreign researchers or kicking them out of the country and are fighting against anything that smells of climate change. All of this is, you know, it's a core tension of the American ruling class at the moment that to get into power, these sort of groups are part of the coalition and that you don't have a dominant element of that coalition, which can just point the direction that needs to be taken for the sort of benefit of that fraction. Whereas in China, it's obviously a much sort of simpler situation in that respect that the Chinese government can dictate key priorities and those can then be interpreted in a variety of different ways by provincial governments and leading tech companies. And so it's possible here, not only the obvious reality that the US is going to miss out on all of the economic growth related to green tech, but also will miss out on the profits generated from the AI that American capitalists have poured so many hundreds of billions of dollars into into training because your book suggests that that it might very well be be China that's so much better poised to dominate diffusion and application is is this what the success of of deep seek trained very, very cheaply because it piggybacked on pre existing innovations. Is that what deep seek suggests? It does to some degree, I think the extent to which deep seek is different from the American frontier companies is I think somewhat overstated. So the you know, the state of training costs of the one model, I think was five or six million dollars, which is really cheap, but it wasn't the model which got all the attention. So there was more training needed for that additional model. And it was only the cost of one training run when in fact there's often many training runs and it didn't count the you know, the cost of buying the GPUs and stuff. So deep seek was still an expensive proposition. It wasn't a cheap one. I think what China does show is that there is there is the capacity for being a fast follower and seeing what America has done, what American companies have done and sort of reverse engineering the techniques to get to where they've gone. So a lot of you know, I think a surprising amount of money that these companies are spending in America has to do with R&D. It's part because machine learning is not like a formal science. It is a very empirical science. It requires testing ideas. And you know, one of the things that got us to chat to you was just that some people thought, "Oh, we've got this technique, but what if we threw a lot of computing power at it? What would happen?" And nobody really knew until they went and did it. And so a lot of the developments are extremely expensive to get to. But then once you see that they've been developed, you can follow them quite. I don't want to say cheaply, but you can do it at a much cheaper pace. You still require huge amounts of talent because it's not, you know, deep seek is a top tier team. They are, they're not the team who knock off of open AI. They are up there with the best researchers. But they're working with a way less compute. Yes, yes. But again, as a matter of not, you don't have to, I mean, without getting into too much of the technical details, you can trade off some of the pre-training compute needed and build up models in post-training. And you can sort of balance these things out in a way that they can run on less compute and still be as capable. And you can distill elements from advanced models. So they do with less compute for sure. And yeah, I mean, is it a threat to America's approach? I mean, kind of, this is where, you know, this is partly where the intense pressure comes in America's because of this belief that you can reach an AGI point. And as long as China is three months, six months behind, it doesn't matter because you will have three months of your artificial general intelligence, self-improving itself, and quickly taking off into the singularity sort of thing. And that's all you need, at least according to, you know, some of the narratives around this stuff. So I mean, if that turns out to be true, then yeah, America's bets will pay off. If it doesn't pay off, if it doesn't turn out that way, then, you know, it will be $1,000,000, sort of poured into all this computing power that, who knows, how much we'll need in the future. You write, quote, "There is a distinct lack of any consensual element to the new ruling coalition within the United States, which might bring the public along." If the Silicon Valley consensus promised new devices and conveniences, along with cheap credit to buy consumer goods, the promises of the new period are more uncertain. One is tempted to reach out for more libidinal explanations, with the hatred, harassment, and humiliation of particular others, such as immigrants in the trans community, substituting for economic benefits. I think that seems very correct, but fortunately, the political return on fascist racist gap-goating and persecution, it seems to be diminishing somewhat. And so I'm hopeful that the fact that these tech overlords are really unpopular could, thank God, pose a real problem. The tech industry, of course, has flexed a ton of political muscle and receneers to block regulation on corporate consolidation, cryptocurrency, AI, etc., etc. They've had huge benefits from this, including something we haven't mentioned, which is that one reason Silicon Valley firms really benefit from prostrating themselves before Trump is that so many of these companies are benefiting from Trump's war on European regulation of the sector. But none of this is possible. People seem increasingly hostile to the tech sector in general and to AI in particular. AI pulls horribly. It's increasingly synonymous with ugly, ecologically destructive energy hogging data centers, fake slop content, systematic literacy, the destruction of ordinary people's jobs. And I wonder, does this mark a contrast with the role that crypto played in the 2024 election? and where the industry, I think, was perversely able to perform a sort of predatory inclusion that allowed for ordinary crypto buyers, or even aspiring crypto buyers to believe that the industry's enrichment was tied to their own. Has this sort of capacity of tech to articulate a shared collective future? Has that seemingly come to an abrupt end? Yeah, I think so. I think one of the differences with crypto, I would say, is that for the average person crypto just wasn't an element of their life. So it was a relatively inconsequential thing and sort of crypto lobbying for getting rid of regulations. Most people didn't have any sort of informed opinion about it. With AI, it's different because it is, if you like it or not, it's being forced down our throats to use it everywhere. I was reading some PC other day that was pointing out there's like four different ways that you can access co-pilot on Microsoft's apps nowadays and it's everywhere. Absolutely, everywhere. I can't open a PDF and Adobe without it, like trying to shove its AI assistant down my throat. Yeah, exactly. So it is being thrown everywhere and everybody has an experience of it for better or for worse and far too often it is for worse. And so people have a real sense of that. They have a real sense of, you know, I've been surprised, happily surprised, by the sort of vocal reaction against companies using AI to generate advertisements and things like that. Or Andrew Cuomo using AI to generate anti-Montani ads. Yeah, exactly. Yeah. So I've been surprised by that nicely because I think it speaks to an awareness of how important the creative labor of society is and the need to not just let it be subsumed under AI. And I, I mean, this is completely off topic here, but I do think there can be good AI art, but it requires the same finesse that using any technology does to produce good art, you know. So I'll leave that aside for now. But the other thing is the data center issue is a really live one. And I think it's going to be a growing and important one in America. Because more and more communities are going to have data centers near them. And there's reported issues around water, there's reported issues around. You give them tax incentives to come to build a data center in your county and then you don't make that money back. So like you've just given them loads of money and you get like 10 jobs back. And the energy issue is like the major one. Yeah, exactly. So electrical bills, I think is going to be one of the big ones. And the midterm elections and, you know, going into the future as well as even if there is no causal relation, if people can sort of say, oh, a data center came here and I saw my bills go up. They will connect the two and it's going to become an increasingly political issue. So AI affects people in ways that crypto never did and almost never would. So I think it's going to become this very political issue. And I mean, one thing I would love to do in future research is sort of map out emerging coalitions around AI here. So you've got, you know, people protesting against data centers, people protesting against, you know, the automation of labor. But then you've also got increasingly sort of religious elements built up around AI. So you've got people on, you know, in the Republican Party who see AI as an affront to God. And you've got people in Silicon Valley who see it as a new religion. Check out the Ross Doothat interview of Peter Teal if you would like a contrast of those two perspectives. That's pretty entertaining. Exactly. And so you're going to have these sorts of groups. You'll have groups who are who are convinced that AI is conscious and deserving of all the human rights that we are supposed to have. And you'll have a group which is saying, well, no, they're not conscious. They're clearly not conscious. And, you know, there's no there's no empirical fact you could sort of point to to like adjudicate that dispute. That's just a political fracture that is emerging at the moment and will only grow larger in the future. So mapping out all of these different things and where they're going to play out in the future, I think it's going to be of growing importance to understand where AI politics goes. In November, open AI quickly backtracked after after their CFO said that the federal government might need to backstop the industry to keep investment flowing. The month prior to that, Jeff Bezos said that AI is a bubble, but that we shouldn't worry about it because it's an industrial bubble rather than a financial bubble. And so when it pops, it'll leave world changing technological advancement behind, is this thing going to go pop? And if so, what will the impacts be? Could it precipitated generalized economic crisis or might it be relatively contained? So I think the impact of a bubble popping is massively worsened over the past year, precisely because of these financial interconnections. And another key channel for a bubble popping and having impact is through massive declines in the stock market, as you were saying, the amount of gains that the stock market has seen from AI companies is huge and has been for quite a while now. So if that were to all be rolled back, it would be a huge debt to the stock markets and people's pensions and all the sorts of financial investments that households have. And that would have pretty dire economic impacts as a result, especially given that there's not much other dynamism aside from the so-called dynamism of the stock market right now in the US economy. Yeah, I mean, so I think it's hard to say for sure because I think you can imagine a counterfactual where if all this money wasn't being poured into AI, it might be poured into clean energy, for instance, or something like that. So it may just be that the American economy might find some other source of dynamism, but definitely at the moment it's a bet on AGI, AGI specifically. And as near as I can tell, whether it's in a bubble and whether it pops, almost solely hinges on whether or not OpenAI can meet its spending commitments, which is just whether or not it can improve its revenues at the rate that it says that it's going to. Because so much of the spending is tied up with them, all these different cloud providers, all these different hardware providers, the so-called NeoClouds, which are typically crypto miners, which have now pivoted to being AI hardware companies. All of them have massive tie-ins to OpenAI. And basically have deals in place that OpenAI is supposed to be paying them, in some cases hundreds of billions of dollars. And if AI can't meet those commitments, those companies then suffer. And that's when the sort of ripple out would seem to happen. And these huge negative investor reactions to all of these announcements of giant capex suggest that investors are increasingly worried that OpenAI will not do so. Yeah, it's very much a correction, I would say. And it's a slightly nuanced correction because you know, the companies which are being put under pressure oracle most notably, they're also objectively the ones which are the most precarious in terms of their position. It's not, you know, Google's not suffering, Microsoft is not suffering, Amazon, I don't think it's suffering either. It's oracle. It's meta, which is promising all this capital expenditure, but without any clear way of making money. You know, these companies are the ones which are getting hit the hardest in the stock markets. So I don't think that sort of nuanced take, you know, that some companies are declining and others are doing fine. I don't think that suggests that the bubble is popping. I think it is a correction that like actually can oracle, you know, do what it says is going to do. Probably not. Can meta do it and still make money? Probably not. So it reflects, I think opinions more on individual companies than on AI as a whole. Well, I appreciate you checking my somewhat motivated reasoning here, but we shall, we shall see. Before I ask kind of a big picture closing political question, I should also ask you, is Bezos right? Will AI's promise as this productivity transforming super technology will that emerge relatively unscathed from the bubble bursting or given everything we've just discussed about the broader dynamics of the broader global dynamics of the sector and the strong possibility that China could eat the US's lunch on this? Is that not even the right question to be asking? Yeah. So I'll give you my most optimistic scenario for the bubble bursting. This is my like dialectical take on it all, which is that we can imagine a scenario where the bubble bursts and all these data centers are sort of half built and we don't really need them. But also we've rolled out loads of cleaner energy sources that now we can use and that actually the sort of transition to far more renewable forms of energy suddenly is just there for the taking if the bubble bursts. That would be the most optimistic scenario. I think highly unlikely, but in my more wildly speculative moments, I like to believe that might happen. I think the bubble bursting, it would have I think impacts on the development of new capabilities of AI. So I think it would be far harder to make an investor case that to train the next generation of models, we need to spend a trillion dollars on compute. That can't happen again if it pops. Yeah. Yeah. That's a tough case to make. They need to invent another reason to spend a trillion dollars. Or as it goes, you know, AI has gone through these bubbles and bursts before. You just wait 20 years and people forget. But even if that bubble bursts, I think people companies will continue to use AI and by all available evidence, that's quite a big portion of the compute right now. That's being used. So I think lots of the data centers will actually still continue to be used. But it will almost certainly be consolidated into the hyper-scalers hands. So Anthropic will get bought by Amazon and OpenAI might get bought by Microsoft. That sort of thing might happen if there's a bubble bursting and, you know, all investor enthusiasm sort of goes away from for a while. The cloud companies will take it up and just provide that AI as a new part of their service and that will be it. What do you, you might have been gesturing to this a moment ago, but what do you make of the argument that Advait Arun of the Center for Public Enterprise makes in a recent report that the bubble popping would provide an opportunity to snatch up stranded energy assets that could be redeployed in a socially and ecologically beneficial manner? I mean, in a different political context than the one we're currently living in. I think absolutely. I think it's something that government should be planning for. If we're going to be rolling out all this new energy for data centers and maybe we don't need those data centers in a few years' time, absolutely. We still need the energy. We need cleaner sources of energy. And, you know, most of the new energy being rolled out in America is natural gas based for the data centers. But there is solar powered stuff. There is wind powered stuff. And, you know, with a different administration, there might even be vastly larger amounts of that sort of renewable energy being put out there. Even if it's just natural gas, which is being put out, though, the ability to just shut off every coal plant would be a partial wind. So it's not ideal, but it's certainly better than continuing to run coal, for instance. You write, quote, "all these tensions between leading tech firms and their governments might not be significant if states were able to easily dominate and regulate their tech companies. But crucially, platforms today have more power in autonomy from their states than earlier periods of capitalist state alliances." And you gestured at this earlier when you were like, "No, this is worse and way more powerful than finance has ever been." And so I want to close by asking how then do we build a left political project that can achieve this historic task of overcoming the enormous political economic domination of tech capital? It's just really incredible given what are obviously the core challenges facing humanity at the moment, that this is what such a vast share of our human financial ecological resources are focused on. And I'm not just talking about focused on AI, but I am talking about that, but also the specific ways in which AI is being deployed for Sora Slop, for students learning how not to read, etc. All of these resources are finite, and yet here we are building houses for chips instead of for people at a time of a historic housing crisis. And prioritizing energy demand for the sake of more compute instead of transitioning our entire energy system to renewables, it's so maddening all because we live in an oligarchy where a tiny number of people decide everything. What would a left wing program look like that could meet this moment? Should we be calling for the mass destruction of data centers? I mean, I don't have all the answers, and as much as I wish I could come and just say, here's what we need to do and be able to give a proper full answer to that. But I think there are things you could point to as key ways to influence the direction of travel. One has to do with the tech work sector, and the organizing that has gone on in amongst tech workers, particularly in Silicon Valley, and the ways in which they've been able to push against some of the worst elements of these tech companies. There's that part. There is the obvious point as well about simple regulations, whether it be around antitrust, whether it be around social media, whether it be around, you know, children's usage of this sort of stuff. There's all sorts of regulations that could be put in place. You could also, and this has been done in a couple of countries when American hyperscalers come in and say, we want to put a data center in your country. What are you offering? You could make the case, for instance, that data centers are only allowed if it's off of renewable power. You could make the case that data centers are not allowed tax breaks, and they do have to pay their taxes, which they can actually be from what I can gather, a fairly significant amount of taxes that can be gained from them. But that requires that you're not giving them tax breaks in competition with other parts of the country. The other sort of thing is more ideological, and I've been cheered on by the fact that in the sort of AI community, building stuff that produces AI Slop is increasingly really looked down upon. So workers who have gotten paid extremely well to go work for Meta, for instance, and Bessacher Berg was reportedly offering billion dollar salaries at one point. So huge amounts of money to go work for Meta. But the AI community is sort of looking at those people now and saying, you're a genius at this AI stuff, but you're just using it to produce Slop. And same thing with Sora, you know, Sora was really looked down upon precisely because people didn't see any social value in it, and rightly so. Though Andrew all of those kind of companies respond and say, build drones and missiles instead. Yes, exactly. Well, this is exactly the narrative they give that America has been far too focused on consumer frivolities for the past decade and needs to be focused on good, manly, hard tech now. Yeah, I mean, there's so, so much ideological condensation involved in these narratives here. Yeah, so I mean, there's this ideological battle. There's battles I think in workplaces to be had about whether and how AI is adopted, and what it means for the people actually doing the work. All of these things I think are points where we can push and have some say. I think the other key one, because you know, one of my big concerns is an increasing geopolitical divide between China and America and that sort of split, and ultimately a potential technological decoupling is I think it lays the groundwork for future conflicts and I think I think that would be historically appalling. So that sort of situation has to be avoided. And maintaining technological interdependence between the two countries is one way of doing that sort of thing. So I think one thing that can be done by researchers, but also supported by governments, is building up alternatives that are open source and that are building up the tools and the scaffolding that makes these things so that big companies can't capture people. So I'll give a very simple example. If AI agents are part of our future here, if they're dominated by particular platforms, and then that platform gets your personal data, it's very difficult to then move your data out and you're sort of stuck with that platform, much like we have with social media today. You get stuck with an AI agent, except to be far worse in social media because they would know every element of your life. By contrast, if we build systems which mandate that data is, personal data is interoperable and can be moved very easily, then it's suddenly much easier to be moving away and moving all, you know, yourself all around and that ability to concentrate wealth, power, resources becomes not impossible, but technically much more difficult for these companies to do. And I think we're at a brief sort of moment where interventions into this sort of scaffolding around AI can have quite significant impacts, you know, building up the protocols which ensure that we have an open system rather than a closed system. So lots of different, you know, avenues, I think, for pushing in a better direction, even if, you know, everything is sort of dense set against us. - Well, Nick Sirnick, thank you very much. Thank you. (upbeat music) Nick Sirnick is Senior Lecturer in Digital Economy at King's College, London. His books include the book that we discussed today, "Zillokin Empires," "The Fight for the Future of AI," as well as "After Work," a history of the home in the fight for free time, co-authored with Helen Hester, platform capitalism, and inventing the future, post-capitalism and a world without work, co-authored with Alex Williams. Thank you for listening to the dig from "Jackamon Magazine." As Mark Swin said, after noting that, technology discloses man's mode of dealing with nature, the process of production by which he sustains his life, and thereby also lays bare the mode of formation of his social relations and the mental conceptions that flow from them. While other podcasts suddenly interpret the world in various ways, our point is to change it. We're posting new episodes most weeks. The dig was produced by Alex Lewis. Our associate producer is Jackson Roach, music by Jeffrey Brodsky. Our operations manager is Sylvia Atwood. Our senior advisors are Theoreal Franko's and Ben Maybe, special thanks to Ben Tarnoff for helping me prep this episode. Check out our vast archives and newsletters at TheDigRadio.com. Follow us on Twitter and also Instagram at TheDigRadio and please do find us wherever you get podcasts and subscribe to this one. If it's an iTunes or Spotify, please leave us a nice rating and review. Those reviews help introduce us to new listeners, but what really and truly does that is you telling people to check out TheDig, you like it, maybe they will too. Please make propaganda for us. And do find us at patreon.com/thedig and make a monthly or annual contribution to keep this operation up and running strong. Even a few bucks is huge. (upbeat music)