What AI's nightmare November means for politics, power and business

Hello, and welcome back to Independent Thinking, the weekly podcast from Chatham House. I'm Bronwyn Maddox. Today, with stock markets sliding and one of the big tech chieftains on the front pages for reasons he definitely did not want, is the AI bubble bursting. The Democrat victory in the U.S. midterms has had a dramatic effect on equity markets, after an initial short-rise tech stocks, which dominate the U.S. indices and therefore many global ones, fell dramatically on fears that Congress would now boost regulation, block new data centers, and that the eye-watering investment in AI would slow down. Anthropic, which devised Claude and is preparing for its initial public offering, that is a flotation on stock markets, has had to accept that the value of the company may be only $1.5 trillion, a quarter smaller than it first hoped. Yet questions remain over whether the AI boom can last. Can Trump or any U.S. president keep backing big tech in the face of a voter backlash, which seems to be driven by fears of the cost of data centers, job losses, and potential harm to children and teenagers? If he doesn't, can the U.S. keep its narrow lead over China in this technology? Or has it already lost that lead, given China's own ferocious investment perhaps helped by a bit of spying? Is AI the secret of world growth and the transformation of medicine? I was at the Francis Crick Institute last night with cancer researchers who could not have been more excited. Or is it something that could, as the prospectus of anthropic warns, be an existential threat to humanity? Joining me today in the studio, I'd like to welcome first David Lubin, who is the Michael Klein Senior Research Fellow in our Global Economy and Finance Program. Hi, David. Hello, Bronwyn. And also, we have Alex Krasodomsky, Director of our Digital Society Program. Hey, Bronwyn. Great to have you both. Well, look, let's just start off with some initial questions. Alex, to what extent is AI a real problem for Americans right now? I think the results of the midterms suggest it is a big concern. Over half of Americans polled suggested this was a major or crucial factor in determining their vote. Down ballot, I think some of the data centre measures were rejected, and particularly on the Democrat side, actually. So you are sort of twice as likely as Democrats to sort of be feeling a bit uncertain about data centres as you are as Republicans. So, look, I mean, you can compare that to questions like the Iran war, the cost of living crisis, government corruption. These were all issues that ranked much higher for voters than data centres. But nevertheless, I think over half is pretty significant. David, I'm going to ask you even more broadly, what is going on? Well, before the midterms, for most of this year, U.S. financial markets were suffering two conflicting forces. On the one hand, the AI investment boom was building hugely optimistic expectations about the future of corporate profits, about the future of the U.S. economy. But on the other hand, the inflationary effects of the Iran war were causing interest rates to rise, and that was potentially a big threat to financial stability and to profitability in the corporate sector in the United States. It turned out that that rise in interest rates, we saw the 10-year bond yield go from 4% at the beginning of this year to around 5% now. That increase in the 10-year bond yield in the United States didn't have any devastating effect on U.S. equity markets. And so everything worked out. What we have now is exactly the opposite of what we had before the midterms. In other words, AI optimism has fallen, and that has led to the sell-off in equity markets that you described earlier, Bronwen. But at the same time, inflationary expectations have fallen because the lower probability that the war will continue to be prosecuted eases expectations about inflationary pressures resulting from the war. And so we now have a situation where AI optimism is diminished, but inflationary risks are too. And so we're left with a different kind of uncertainty than we had before the midterms took place last week. You put it beautifully. Let me just take us back. The AI optimism is about whether the technology transforms corporate profits. It makes companies across the world much more productive. Is that the idea? Yes. But before you get to that, there's something else going on in the first place. The best way to think about it, if I can put it this way, is that right now, AI is affecting the demand side of the economy. In other words, there's data center construction and that sort of tech investment, which is now running at about 8% of US GDP, is causing this huge surge in physical economic activity. And AI-related investment probably accounts for more than one half of US GDP growth this year. And AI represents something like 40 or 50% of US equity market capitalization. So right now, what we've got is not so much the productivity gains that result from AI, but the sheer economic activity that results from AI-related investment. The productivity gains that I think are the basis of your question come later. And there's still a huge amount of uncertainty about whether the sort of positive benefits of those productivity gains will outweigh the existential risk to humanity that you highlighted. And I think that's one of the reasons why we might be in this sort of uncertain moment is that I think the productivity gains were promised early and haven't materialized. I think a recent MIT study found that 95% of projects being rolled out across a number of corporate partners were generating absolutely no return on investment whatsoever. And we sort of heard earlier this year, the PIP starting to squeak around Uber and Meta and this sort of sense that actually using this technology was enormously expensive and not generating a return as they might have expected. And Alex, what do you think the impact of the Jensen Wang story is? I was speaking at an FT gathering earlier this week where they were just going around and this was not the topic of the congregation. They were just going around saying, what was Jensen thinking? This is, of course, about the CEO and founder of NVIDIA, the chip maker, being revealed to have had an affair with an alleged Chinese spy. I think, you know, immediately in the wake of this scandal, a number of other sort of Silicon Valley luminaries and indeed the president of the United States came out in his defense. But while a lot of folks are looking to sort of brush this under the carpet as an indiscretion of one kind or another, others are pointing to the fact that if indeed the U.S. is in a race with China and this situation led to the margins of that race much narrower, there are a lot of people who are extremely worried about that, both from a national security perspective and from the sort of AI safety existential risk questions that you outlined earlier. So let's dig into just that question a bit more, particularly from the American perspective on this. When we last did a podcast on this, we were examining whether the competition between countries and between companies was so fierce that any thoughts of cooperation or governance were really for the birds. Let's just look at this. The U.S. versus China question on it. Where do we think the race is on this? The U.S. has a technological edge. How big that technological edge is, is open to question. You know, one of the things that has been revealed through the Jensen Huang story is the ease with which China has access to compute via data centers built in Southeast Asia and has access to chips via Singapore. Despite the U.S. prohibition on selling many kinds of chips, not all, but many kinds. Yes, although technically there's no restriction on Singapore on selling chips to China. And remarkably, the firm at the center of the Jensen Huang story is not on the U.S. entity list. By which you mean? It's not subject to U.S. export control. So there's nothing restricting the firm. So NVIDIA was selling to this firm and the firm was selling on to China and all that appeared to be completely legal. Absolutely. So, you know, China, has all sorts of capacity to catch up. And that's underpinned by the fact that it's got access to a lot of very cheap electricity. There's a lot of data center construction. And so at the point where its ability to secure chips is more ensured, China's ability to catch up, I think, will escalate very dramatically. There has been some head scratching as to how China has maintained what I think at the moment is roughly described as a three or four month lag on the U.S. in terms of capability. The most common accusation is one of intellectual property theft, something called distillation, when you sort of use a U.S. model to train your own model and make it smarter. And this is something that is banned under the terms of service. And I think a lot of the hawks in the U.S. put a huge percentage of China's progress down to this effectively theft of intellectual property. I'm a little bit more skeptical of that. I think that you could put a lot of extremely talented AI engineers and a real push from the center in China to advance this particular technological revolution. The one unanswered question was how on earth were they doing it without access to all of the hardware? You know, around 60% of NVIDIA's semiconductors would be going to the U.S. Other partners in the U.K., Europe, elsewhere, would also have access to these semiconductors. And of course, Malaysia and Singapore as well. And perhaps that is the answer that folks had been looking for. In fact, that China did indeed have access to these chips, just not directly. directly, but via proxies in Singapore and Malaysia, exactly as David outlines. How important is this war, if you like? How much does it matter to be in the front? On the same day that the questions around whether or not NVIDIA chips that shouldn't have been going to China were indeed going to China, another warning emerged from the labs that they had underestimated how quickly their AI models were improving. This was a report that came out of Anthropic and effectively it suggested that using AI to train new AI was going better than they had initially forecast. Now, this is obviously extremely significant and the kind of thing that if you believe and if you are worried about, poses the kinds of questions on the existential risk side that really suggest that if this is going to happen and the forecasts are superintelligence by 2027, whatever that means, and however far you believe the claims being made, these labs want that moment to happen in America and not in China. And I think historically, the link between technology and war preparedness is so close that that is what really underpins President Trump's insistence that this is a race that the United States cannot afford to lose. And we're thinking of wars, not just the conventional kind we're talking about, but cyber wars and indeed wars in space, things that need to be in the lead in the control of data. Absolutely. Let's just dig in a bit more to what the feeling is in America, what America is doing on AI and indeed where Europe and the UK are. Where do you think this apparent public backlash is coming from? I think that there are very obvious proximal causes. Texas, which flipped Democrat, is one of the sort of data center heartlands. There are hundreds of these things being built. And frankly, living near a data center like this is a pretty tough sell. They are noisy. They create a lot of pollution. They push up the prices around electricity, water, and so on. And there are lots of examples across the US midterm elections of, you know, leaflets being sent out saying, well, you know, if you don't vote Democrat, they're going to build one of these things right next to your house. And I think that most folks suggest that that's something that they didn't want to happen. So I think that there are very obvious, very, very real sort of local concerns that I think are replicated in the US. And they don't want it because they feel, and this is true in some cases, because of the way electricity is priced in parts of the US, that it pushes up their local costs as well as potentially threatening their jobs and so on. Sure. And when these data centers aren't or can't connect to the grid, that's okay. Then they'll stick in a gas turbine that will, you know, pump out huge amount of pollution in order to power these things. I think just two of these that are proposed in the UK, and there are hundreds proposed in the UK, would smash our ability to meet our targets under, you know what I mean, COP is ongoing, right? And to meet those targets, data centers make that very, very, very challenging, particularly if they are being run off fossil fuels. And in fact, I think that there are some nuclear power stations being built specifically to support them in the southwest of the US. Yes, but there are many more gas turbines being built. We've had data centers for a long time. Why is there such a problem now? The scale, I think, is unforeseen. The question then might be, well, what can data center builders do? We need these things for our economy in order to make them more palatable to the public. And that is a question that they are going to have to now answer. Aside from the nimbyism, which emphatically exists, there are a couple of other things that I think are worth referring to in the explanation of why there's such a high level of rejection in the United States. One is wealth, and the other is trust. It tends to be the case that higher income countries are more suspicious about AI. Lower income countries that have obviously more natural ambitions for income growth and catch up to advanced economies are more willing to accept the risks, whatever they happen to be. The other thing is that, according to Pew Center research, the United States is the country where there is the lowest level of trust in the ability of the authorities to effectively regulate AI. And so I think that, you know, there's a bunch of different things that come together to explain why we've reached this political moment in the United States, where a midterm election was really, in some respects, governed by concerns about AI. And I'd also be interested in your views on this. But to my mind, there is a longer tail here. AI is, you know, very front of mind for lots of folks now. I think that there is a skepticism towards the kind of technological revolutions that we've been undergoing for the last 10 years that has been building up both in the US, but also here in the UK. And I think that there is a real sense that a handful of folks have become very wealthy off the back of this technological revolution. And a lot of other folks have felt rather precarious as a result of it. You know, the rise of the gig economy, the rise of social media and the abuses around it. I think all of this does, in fact, weave into a single sort of tapestry that ends up at AI, where there is just a general lack of trust in these American technology companies to deliver the kinds of gains that they talk about for the average person. And frankly, giving them a kicking at the response to 10 years of perhaps feeling left out of the progress that has been trumpeted in Silicon Valley. And frankly, when your sales pitch for AI is a best case scenario is job disruption, and a worst case scenario is extinction of humanity. It's sort of at the apex of that disconnect. I wonder whether it's a bit different here, though, because there's a lot of environmental and sort of aesthetic concerns. We don't want that ugly data center right here. There was absolutely not very well formed, but unease about jobs. And what is this going to do to jobs? There's a lot of concern about the harm of social media for teenagers and adults. And I think there's a lot of public support for what governments in UK and Europe have done on content control. But it's not it seems to be not quite the same as the US where the people who have become so wealthy are right there in America, in their leather jackets or whatever, but also putting in some cases their money into US politics, that is, they are trying to shape the very. Fabric and change the democracy of the country that they're in. And we don't have quite that factor here in the UK, I think. Yes, but also following the point that it's kind of rich countries where the greatest level of anxiety exists. You know, there are a lot of countries standing up saying, we'll take your data centers. I mean, you know, Argentina now, President Malay is saying, you know, look, there's the whole of Patagonia, albeit that Patagonia is slightly has a water shortage. You know, there's lots of data center capacity here. There's already been almost $60 billion worth of investment in the data center. And I think that's a lot of data center capacity here. There's a lot of investment in Latin American data centers, most of it Brazil, but kind of spread around. So maybe the future is, in response to this sort of nimbyism and low level of trust in the United States, that data center construction kind of spreads across the world, because there are plenty of developing countries that want to see some of that activity. David, there is a long history of American ambivalence towards big business, isn't there? Building up and championing these enormous companies and then using the antitrust mechanisms of the US or regulation to break them down. Yeah, antitrust politics is a huge theme in American political history. And it may be that one of the things that's kind of underpinning the buildup to the election, the election results themselves last week, is that we're, you know, moving towards a more antitrust kind of US politics. Alex, do you think it is now hurting the tech companies that President Trump has made such a thing of their favorite place in his circle? I'm thinking of one financier who said to us, we're normally in the front row of inaugurations, and we're in the second row, and all the tech people are in the front. No, I don't. Alex Karp, the CEO of Palantir, came out yesterday and reminded everybody that he is, in fact, a Democrat, and was brought up a socialist, or at least in a very progressive household. So I think just as first under Obama, there was a swing to the left, and then to the right and then to the left. I don't think Silicon Valley will be tarred with Republican brush for very long, if that's indeed what it's going through at the moment. And as you've described, the great distancing is beginning. Where do you put China in this, in terms of embrace of AI? Oh, there's a very high level of AI optimism in China, very high level of AI acceptance. You were there recently. Yes. And there are lots of different ways of explaining it. You know, it probably doesn't have much to do with authoritarianism, because there are lots of non-authoritarian societies in Asia and elsewhere that have similar levels of acceptance of AI. It may have something to do with demographics. You know, talking about existential threat, demographic decline, is an existential threat to public financial stability. And so the future cost of pension obligations in an aging society, if AI can be a new source of economic productivity that can generate the kind of income that can pay for those pensions, fantastic news. And so for all sorts of reasons, cultural, societal, political AI acceptance in China is really very high. What do we think the Democrats, now that they control both houses of Congress, might do about all this? I'm thinking of everything from approval of data centers to not pursuing Europe quite as ferociously over its data laws, which the Trump administration has taken a huge dislike to. I think that we can expect Democrats to attempt to make good on some of these promises that they have very clearly and very concretely made in their manifestos. I would expect changes to things like tax breaks, to encourage data centers, really, an attempt by Democrats to push more of the costs of data centers back onto the developers who are trying to build them. But more than that, they have promised things like moratoriums, greater public consultations around these data centers and where they are put. And I think they're going to have to make good on that, particularly in states where data centers were a really core part of the pitch being made. I would say that, you know, the two sort of broad concerns are that data center construction will be restricted. And I think Alex is absolutely right, you know, in the way he describes the risks to that. The other concern is that, you know, that is often described is about AI regulation. And I think that is a little bit more uncertain because regulation is an executive function. So even if Congress finds ways of kind of compelling the administration to regulate AI, you can bring a horse to water, but you can't make it drink. And so it may be that the risk of more dramatic AI regulation is more in theory than in practice. There are dozens of AI regulation bills now passing through, not necessarily at a federal level, but at a state level. So I think, you know, in California, in Connecticut, Michigan, there are lots of examples of attempts to govern the use of AI at a state level. And I think given the changes in the House and the Senate hopes to then move those up to a federal level, where at the moment the governance of AI is extremely informal, decisions being made in the dark, and often by folks who you may not be expecting to make those decisions, that may then become more formalized under it. And of course, the states have lots of leeway under the U.S. Constitution to do this, and all kinds of ways. As they've been doing, for example, on energy. But then you have the spectacle of this ultimate fluid technology that barely recognizes borders at all, being regulated in completely different ways at the subnational, at the state level. Let's just go on for our final question to this question of governance, which you're taking us into. And when we had our last podcast on this, we were wrestling with this, of whether governance was a dream or whether these different initiatives that there are around the world, not just a head off. Well, I think it's a great question. I think it's a great question. But to try to get something that reassured the public and reassured societies that this technology would be pointed in good directions, not bad. Whether these were going to come to anything. So we've got the U.K. with quite a flexible model, looking very much at sort of outcomes, putting weight, as the U.K. is prone to do, on a regulator. You've got the E.U. with its very legalistic, meet all these requirements right now, whether or not they turn out to be outdated, or you face huge fines. And you have the U.S. in, as we've been discussing, a different place. Where do you think this is going to go? Well, it hasn't gone far since we last spoke. AI was a hot topic at the U.N. General Assembly this year. And there were a number of countries calling for more control. A real focus, I would suggest, on two things. One, the safety, as we described in the second, on whether or not middle powers can somehow collaborate to own a stake in this economic system. The other place where there's been a fair amount of movement has indeed been at the state level, as we talked about. So the current push, particularly being led by California, is for third-party audit of these labs, something they call independent verification organizations. And then the question might be, well, if that is the path that we are heading down, where if you have an AI model, you have to bring in a third party to make sure that it's safe, who are those third parties? And I think that will be a big question for the U.K., given that AC, the AI Security Institute, is one of the jewels in the crown. I think there would be a real hope that if indeed we do move to a model where third-party audit is the process of choice for making sure that AI models are safe or meet any other sort of sets of standards that folks might hope for, could AC be one of those independent verification organizations? Let me just stop on that point for a second. This was the institute created by Rishi Sunak saying, look, we've really got to do some things. And he was prescient in this. But isn't the weakness of this institute that if companies or countries don't want their stuff to be regulated, they just take a little sidestep around it? And we've had one or two cases of that. That's right. And I think if you look at why the U.K. decided not to sign up to President Stubb and President Storr and President Ruto's call for control at the U.N., the sort of generally accepted reason is because they did not want to jeopardize the relationship with the U.S. and therefore jeopardize the AI Security Institute's access to these models. But you're absolutely right. Access remains voluntary. And I think questions on what kind of access and how far that access can be compelled will be front of mind for the U.K. going into the G20. I think in the context of U.S.-China relations, the good news is that since Trump went to Beijing in May, we've been in a period of constructive strategic stability. And that's a kind of detente. And, you know, if you take that analogy of detente back to the 1970s, well, it was detente that allowed the U.S. and the USSR to start negotiating strategic arms reduction. So with that analogy in mind, can you be optimistic about the possibility of the U.S. and China getting together to regulate AI? I really don't think so at the moment, unfortunately. So in a way, we're in this kind of weird situation where the political environment may be a bit conducive to some kind of agreement between the U.S. and China on AI governance. But I really don't think it's likely. And, Alex, you mentioned various U.N. initiatives and the Global Governance and Security Center at Chatham House, which your program as part looks very much at whether there is still life in these old, if you like, multilateral, to use the jargon, institutions. Any chance of that getting somewhere? I think on AI specifically, it's unlikely. But I think that where there will be dialogue, it will probably remain within these institutions for countries to at least signal that this is something that they care about and they want action on. You know, the U.K. took that opportunity this year. Australia drew attention to a rogue AI incident that targeted their government infrastructure that they blamed OpenAI, a U.S. company, for. So I think it's definitely a platform. I think it's definitely a platform where concern about the future of AI will continue to be raised. Whether it is a platform where anything might be done about it, I think, is a different question. With that, we are going to have to draw to a close. So let me first thank David Lubin. Thank you. And Alex Krasodomsky. Thanks, Bronwyn. And let me just say a couple of things at the end. First, a really exciting announcement. Our Chatham House shop is now open online for gifts celebrating 100 years of the Chatham House rule. That's going to be 2027, but you can buy them for Christmas now. And they look great, and they are selling quite fast already, notebooks, mugs, and so on. And if you want a more substantial present, Chatham House membership as well. And this isn't just a routine promotion for it, but we are going to be doing so much in 2027 that this is exactly the right time to come and join us. And just to give you a taste of that, do look at our website this week, chathamhouse.org. There is so much. We had the Chatham House Prize given to the women of Afghanistan. A really very, very. very moving evening. We've had the Iraq conference. I would say every corner of every staircase filled with people discussing things in the margins of that incredibly lively and important annual conference. We have a big event as well on the Nigerian elections and more. So do look at all that, and I'll see you next week.