Is AI Self-Sufficiency A Pipe Dream?
Ryan Wain, senior director at the Tony Blair Institute for Global Change, on whether countries other than the United States and China can achieve AI sovereignty. He argues that building a national frontier model is unrealistic for most countries, and that they should control what they can, steer where they have leverage and depend on partners for the rest. The episode was recorded before the recent conflict involving Iran.
Transcript
Before we get started, a quick note. This episode of Agents of Tech was recorded before the recent conflict involving Iran. So some references reflect the situation at that time.
The United States and China control more than 90% of the world's AI computing power. In practice, that means most countries rely on either America or Chinese firms, chips, and rules, to access the most advanced systems ever built. Some call it partnership, others though call it dependency.
Our guest today advises governments on how to try to navigate this. And his answer, stop trying to compete. In fact, he calls self-sufficiency a vanity project. Here's the question. If you're not one of the two countries holding the keys to the castle, what leverage do you actually have? And if you are America or China, should you share this power at all?
AI is becoming part of the infrastructure that underpins modern economies. Whoever controls it controls the future. Ryan Wain from the Tony Blair Institute says, "Control what you can, steer where you have leverage, and then depend on those partners for the rest." But is that a good strategy? My co-host, Laila Rizvi, is here with us in DC with her thoughts. And before we get into today's interview, I'm asking you the question, "Do you think there is a best path forward?"
So Autria, I think this is pretty much a catch-22, where I think it's almost impossible for any other country to compete with China and the USA, and compete on the accumulation of both technological power and economic power. But on the flip side, is dependency the best strategy forward?
Right?
I mean, we're seeing so much turbulence when it comes to geopolitics, and even Britain's relationship with America, with this whole seizing Greenland situation and Keir Starmer, you know, strengthening and maintaining his relationship with China. So I really don't think it's a straightforward strategy where you just say, "Let's rely on our big brother, the USA, and this special relationship."
Right. It'll be interesting to see if Ryan thinks that there could be a happy medium somewhere. Somewhere, we'll see. But before we get to that, we do want to hear your opinion. If AI power is concentrated in just two countries, can other nations still be sovereign decision makers, or are they going to be forced into a subordinate position? Tell us what you think in the comments below and tell us where you're watching from. Now, let's get to it today, and let's hear from Ryan Wain.
Ryan, it's such a pleasure to have you with us today. Thank you so much for your time. I want to jump right into the recent report that was authored by your institute, the Tony Blair Institute, about AI sovereignty. And it argues that AI self-sovereignty is really just unrealistic for most countries. So I guess the big question that we want to get started with, is this not a case for these countries that don't have the infrastructure to create their own AI and become reliant on the existing AI superpowers, is it not just going to further entrench the existing power balances and really widen the divide between the haves and the have-nots?
I think as this revolution, it is a revolution, unfolds, these are the things that we need to be talking about. Look, the report, like all of the work the institute does, is clear, right, in how the world is working. And the truth is, whether it's the U.S. or China, billions, tens of billions, hundreds of billions have been invested in building frontier models.
And the argument of our report is that countries need to participate in the AI revolution. And rather than spending billions of money that they do or do not have on trying to build the next frontier model, they need to look at sovereignty in a different way and understand their leverage, build leverage where they can, and make sure that they are getting access to these frontier models. And we think there is a future where you can have genuine sovereignty, but that's different from having to have everything, I guess, inside your own nation state.
Yeah, and Ryan, I wanted to come onto the part of your report where you talked about dependency, and that really it's a bit of a vanity project to be building your own frontier model. And we might as well, the U.K., for example, depend on our ally, America, for these bigger foundation models. But a lot of these companies, like Anthropic, are already cutting off model access to other companies, their competitors. And given the geopolitical tensions we're already seeing, I mean, the threat to Greenland by the USA and Keir Starmer has already supported Greenland, Keir Starmer is in China at the moment, strengthening and maintaining that relationship. So how can the strategy be just to kind of rely on America? What if they just cut us off because we disagree with them, which is already happening?
Yeah, first of all, Laila, let me come back on the idea of building a frontier model being a vanity project. It's not that it's a vanity project, it's completely unrealistic, and we have to take a clear-eyed view of how the world really is working and how this revolution is unfolding. So it's not a vanity project, it's a near impossible project to build a frontier model. And even if countries try, they will waste vast amounts of resources and they will build something that is substandard and watch as other countries' competitors, rivals, collaborators, advance at speed, and they won't be part of that.
So our focus in the report is in how you deliver progress for people through AI, and through the adoption of and embrace of AI. We have a model on sovereignty, which we break into three rough areas. Understanding where you need to have actual control, and that means literally having the infrastructure or the data centres in your remit, in your country. We talk about where you need to steer the technology. So this is where governments have a choice, right? We can make decisions on policy, on procurement, and that gives you power, that gives you leverage over these models and these companies. And then we do also look at where you need to be genuinely dependent, because there is an interest in ensuring that you are able to embrace and use the latest technology.
What I would say is that countries also have a strategic choice, and we know that an AI model in and of itself is useless. That really is a vanity project, it's what we do with that. So we argue that countries should look to play in at various levels, including the application level.
Okay, and then, but what about if we did, let's say Britain is using the models and the application of various models by OpenAI, Anthropic, xAI, whoever. What if we do have to leave the AI infrastructure environment, which is built upon the U.S.'s foundational technology. What happens then, as in that is a hypothetical which could happen and a future that we might have to prepare for. So what would the Tony Blair Institute suggest in that scenario?
It's what you do before that scenario arises. Look I think I'm more optimistic than you on that scenario, but let's play out that that does happen. What you cannot do is be dependent on one model. So we argue that when it comes to things like procurement, you're much more modular and you don't allow there to be a single point of failure. There are multiple models out there, and the U.K., to use this example, should have relationships and contracts with multiple models.
The key thing is what we then do with it. So if we want to deliver progress for people here, you know, we've got to be embracing AI if we want to be at the frontier of the shift towards prevention in health, for example. We've got a huge asset in the form of the NHS and the data that goes with that. We are in control of that. Sovereignty is a choice and we have levers that we can pull that mean that we are in this world, not where things are happening to us, but we're actively shaping it.
And then look, there are going to be some models that we want to build here. And there's also an argument as well for open source. And there's a big conversation, obviously from a European perspective, about what we need to build, and that's where maybe we do deviate a little bit from how things have been built in the U.S. and China. We look at more open source models and we know they're working already in the world, including in conflict zones like Ukraine.
Sure.
I wanted to ask you, you talked a minute ago about being optimistic and that you are a little bit more optimistic than perhaps some of us are about these countries that are going to have to be reliant on existing AI foundations. I think a big concern is that the narrative surrounding AI is primarily pretty negative and people are concerned about, it's going to take my job, it's going to, you know, change my life and have a negative impact. So for these governments that are not going to build their own AI infrastructure and foundation and are going to be reliant on maybe one of the big two, the U.S. or China, how do you really reshape the narrative for the people in those countries and the constituents that they would be serving, to make sure that they are on board with their government being reliant on another country?
Well, look, a narrative is shaped by reality, rather, the best narratives are shaped by reality. So look at the reality, take a country like Kazakhstan. Kazakhstan has the big ambitions to train a million people to be fluent in AI, to become AI engineers. That is a great contribution to what is an interconnected ecosystem globally when it comes to Artificial Intelligence. So you've got to make the case, I think, to your people, around the role that you can play, based on your strengths. Kenya's another great example. Kenya has geothermal energy. We know that AI is energy hungry. They've been able to leverage that and use it to strike up a partnership with tech companies in order to bring AI to their country.
I do think though, you are pointing at something which is, we all need to be aware of, which is trusting that technology. And this is where I think we've got a job to do, to educate the public, to have debates and discussions like this on what is an emerging technology, but also a technology that's already proven to be transformative.
And on that note of obviously trust, there was a lot of anxiety around the collection of data, like when the idea of a digital ID was proposed, which a lot of people were terrified of. And I know in one of the reports you guys advocated for embracing digital IDs and technologies that solve crime, like facial recognition and data-driven policing. But don't you think that's potentially a path to kind of surveilling and repressing the public, or even leading to what could be a totalitarian state, whether that's in a country like China with the CCP, but also in our own democracies, as in you're seeing things even play out in America where people are absolutely terrified of what's going on in Minneapolis. So that is a valid concern of the people, no, and is that maybe not a bigger threat than even the threat of being left behind if you don't adopt those kind of data-driven policies? So how do you balance that and how do you convince people on that front?
Yeah, I think it requires leadership, and it requires leadership that understands this revolution but is also able to harness it. And that means, yeah, having the discussion and debate around key issues like privacy. But we know that we can actually probably have better privacy protection through embracing technology. We see that in Estonia, right? Estonia has a digital ID and it finishes top of all the league tables when it comes to citizens' privacy.
In Britain now we are in this absolutely bizarre situation where if we want to prove our identity, I'm doing this live now as I'm trying to rent a house, I have to take a picture of my passport, send it over Gmail to a landlord that I don't even know. Give me a digital ID that is privacy-protected. You have to make and win the argument. And I think that's what the institute is trying to do. We are trying to show that technology is something that needs to be harnessed. It's a really important verb in all of this, you need to grab hold of it. And that happens at a national level as well when it comes to thinking about the new political economy, which I think does need to be built around embracing artificial intelligence.
Absolutely. But if you were speaking to kind of a citizen of the U.K. right now, how would you convince them that, you know, digital IDs are for-
For their safety.
Yeah, for their safety and for their betterment, as in, what would be kind of, if you had to give it in two sentences, what would be your your, your-
Your elevator pitch? Yeah, yeah, exactly.
I would say to them very simply, it will make life better for you and it will reverse the decline that this country has found itself in. So let's carry on with the status quo, by all means, but no one is happy with that. So we need a different future. And I think it is a job of political leaders to be able to understand how technology can deliver that, harness it, and then make the case to, in this instance, the great British public. Like I really do believe in that.
If live facial recognition, remember you have to be on a wanted list, you have to have committed a crime. The police have to be looking for you anyway. If live facial recognition can catch criminals, get dangerous people off our streets, brilliant. If technology and AI can help you understand where a drone attack might take place in Ukraine and move citizens out of that area, brilliant. If technology can keep a citizenship alive and country running during a time of war, embrace it. So we've got to not neglect it, we've got to understand legitimate concerns, but we have got to make the case and take that to people.
All right, we're going to take a quick pause, but coming up after the break, how do we address privacy fears and the risk of an AI-enabled totalitarian state? Ryan Wain will weigh in.
Welcome back. Today on "Agents of Tech" we've been speaking with Ryan Wain, Senior Director at the Tony Blair Institute for Global Change. All right, you really present very interesting case studies, like the UAE and France, as success stories. But we all know that these are states with significant financial, energy capital, as well as regulatory clout. So I guess my question would be, what framework or what does your framework provide for maybe some of those countries that don't already have those types of advantages?
Yeah, the key point about AI sovereignty is that it is not simply a case of having all of the technology in your nation state, which will be impossible actually for every country, including the U.S. by the way, which relies on a global supply chain for its chips. But particularly for low economic developed countries. What we say is, control where you need to control, steer where you need and are able to steer. And that can be through national procurement, it can be through putting your strengths on the table and building those strengths up. And where you need to depend in order to deliver for your people, do that as well. But for those countries that maybe aren't cash rich, they all have something going for them. Like I'm an optimist on that front, and they all have a contribution to make. But sometimes that might need a bit of leadership domestically as well.
So I just want to pick up on your supply chain point, Ryan, because obviously you said that there's a global supply chain and, you know, which AI advancement is reliant on. But I want to come back to something actually that Dario Amodei, the CEO of Anthropic said, at Davos earlier, which was that he thinks it's crazy that America is even participating in a global supply chain where they're giving chips to China, because he compares AI to nuclear weapons, as in that is how powerful this technology could be.
It's not just kind of telecommunications 5G where everyone's, you know, exchanging various bits of technology for, you know, a global kumbaya kind of vibe. It is, like it's very, very different, and he even says that the American government doesn't seem to get that. So given that Britain has, for example, nuclear weapons, and all of the countries that have nuclear weapons do have some kind of leverage, surely it would make sense for them to at least attempt to create something equally as powerful as the U.S. and China when it comes to frontier models.
I don't think AI's either analogous to 5G, which was a sort of single technology, and lessons were learned on dependencies around that. I don't think it's nuclear weapons either, which was a deterrent essentially. This is a technology that we are going to use, that we're going to see to diffuse right across economies. So I think the point that Dario makes is an important one, like there are genuine national security considerations there.
But we have seen this experiment play out in real time. We saw previous U.S. administrations restrict chip sales to China, and yet we saw China build quite a powerful frontier AI model in the form of DeepSeek. That is the reality of the world that we're in. The frontier model capability's not going to disappear from China. But the key point is that a frontier model in and of itself does not give a country anything. It does not give you prosperity, does not give you security. It's about what you do when you apply it. And that requires political leadership, a direction of travel, and requires energy, literal energy. It requires skills and it requires a regulatory environment that embraces and unlocks innovation.
And I think your point about Europe, it's not to build the frontier model, it's to think about that application level. To quote Sir Richard Younger, who was senior in U.K. national security, "All I want for Christmas is 20 global tech companies started in Europe." That's where America gets its advantage. That's where America gets its defence advantage. There is no reason why we can't have that in Europe, and that is a much lower and more realistic lift than building a frontier model that competes with the U.S. or China.
Laila, that's very encouraging.
Yeah, no, no.
Not a heavy lift. Okay, now to put you completely on the spot, Ryan, but I do want to follow up with a little bit about what you were talking about, about the global supply chain. 'Cause I would just love to know your personal perspective on the U.S. selling chips to China. Albeit not the most recent updated generation of those chip, but nonetheless.
Look, the global supply chain, the AI revolution, the world is interconnected and you cannot pretend that is not the case. So that's where I start from. So looking at it from a purely personal perspective, I have no qualms in the U.S. selling chips to China. I think there needs to be a relationship with China. The dangers presented by AI, and, I mean, making no mistake, there are dangers. This is a general purpose technology, it can be used for bad. That is going to be solved through global AI governance, through global governance. And you only get to that through engaging, through dialogue, through diplomacy.
And yes, of course you need hard power, which is where the point about the applications, and building those applications, making sure that we're investing in talent in Europe, that we are also investing in energy or bringing our energy prices down, that we're investing in compute power as well. All of that matters, but diplomacy and governance are going to be the things that stop this turning from a technology that can empower people, that can deliver prosperity, to one that could cause mass harm.
Okay, but so then to bring it back to the topic of your paper, as long as there is this interdependency, even among the highest levels, say the U.S. and China when it comes to the supply chain, is the actual reaching of AI sovereignty, you know, is that just a pie-in-the-sky dream?
Depends how you define sovereignty, right? That's why we start the paper with a definition. If sovereignty is autarky, if sovereignty is having ownership of everything, every bit of the AI stack from compute and energy down to the applications and the frontier models, it is pie in the sky. No country is going to achieve that, even the U.S.
And I just want to come back to kind of, okay, let's, like looking at the countries who aren't the ones building the frontier models. What strategic advantage do you think that they would have over maybe the U.S? 'Cause Autria and I were talking about even adoption, which we've been talking about, and we kind of disagree where, I think that the U.K., people are adopting AI models actually more readily than the majority of the United States. Where I think, yeah, it's mainly Silicon Valley that knows all these different models, but I would say the average person in the U.K. might actually be adopting it a little bit more quickly.
So you have adoption, which could be a strategic advantage, but even data, I mean I talk about NHS data, you know, it's so lovely and standardized, and we have a really nice historical record of it, where as you guys are completely fragmented in your healthcare data. So I imagine there's a lot of companies that are probably trying to get their hands on our nice lovely standardized data in our welfare state in Britain. So yeah, I'd just like to know what you think is actually our advantages compared to, I mean, Britain's advantages, but also other countries which aren't China and the U.S.
Laila, I think you just made a brilliant case for Britain's advantages there. And I'm looking forward to the election campaign because political leaders need to really own it. The NHS is great, that gives us a huge strategic advantage. Now, of course what we can't do is just release that data into the wild, that is sensitive, important data. But it is a strategic advantage that means that we can have a role to play as AI evolves and also gives us access to those frontier technologies.
But I guess I come back to sort of middling powers, if you like, like the U.K. and Canada and elsewhere. You do just have to get your house in order. You do have to sort your energy out. And the fact that electricity in the U.K. is twice as expensive as America, it's even more expensive than Europe, which is also incredibly expensive. You need to bring the cost of energy down. So that requires decisions to be taken at the highest level of government in order to achieve that. And that allows you then to become a much more attractive and appealing place for the data centres, where you might want to have a bit more control, if you like, over the infrastructure that could be sitting behind the NHS data.
But it also, you want to think about the talent, and I agree with what you're saying around people here are adopting AI, and interesting when it comes back to the trust point. We did a big piece of research which showed that actually there is a growing distrust of AI, apart from when people start to use it. And when people start to use it, they trust the technology more. So that's compelling, but it also shows that the government must try and diffuse this technology, and I was pleased to see this government in the U.K. commit to training, I think, 10 million adults over the next five years. That's great.
But we also can't make mistake using a chatbot, an LLM, with AI powering your economy, or powering public services. The two are very different things. So we need proper industrial strategy that creates the conditions for us to see AI diffused into the economy. Working with the next generation of founders and leaders here in Britain and across Europe to give them the tools and means and support and the capital so that they can build the next gen AI companies. And they will do everything from transforming healthcare to transforming public services, through to making our country safer and boosting national security. So that is the job and the opportunity, I think, for leaders of countries outside of the U.S. and China, and frankly within the U.S. and China too.
All right, Ryan, before we wrap it up today, we've got our rapid-fire questions. These are our questions that we asked all of our guests, 'cause we love to have varying perspectives on the same questions. So Laila, you want to start us off?
Sure, so mine is, what's something that's widely accepted in your field that you actually disagree with?
I am in the world of, I guess you would call it sort of centrist politics, politics that believes in progress. We are obsessed with the climate agenda. It's become ideological and sacred. When it comes to AI it is going to need energy and I disagree with this pursuit for clean energy as quickly as possible. I think we need cheaper energy as quickly as possible, and I actually think that will help us with the climate debate, which obviously climate change is real, it's happening. But this clean energy agenda, I think we need to move on from that.
All right, my question for you. Where do you feel people will draw the line at bringing AI into their lives?
Yeah, I mean you can never tell with people, but I'd like to hope it's in their personal life, their romantic lives. But I think I've already seen that illusion shattered-
Yeah.
on my Twitter feed. So, yeah, that's where I'd hope, but.
It's amazing how quickly you just give everything up if you can do more with less, right?
Yeah.
As in, I'm astounded by myself even-
Right.
I'm like, just run with it, on my entire laptop.
Yeah. I mean I saw something the other day, somebody was saying, "I want AI to come in and do my housework, so that I can do the creative projects that are, you know, are my pet projects." Not do my pet projects so that I can get busy with more housework, so.
Yeah, I've heard of a date with a guy who whips out ChatGPT in real time, works out what the best response is to, it's, that ain't great.
That sounds like a reality show in the making.
And again, that's a good application of AI.
Yeah. Yeah. All right, Ryan, thank you so much. Appreciate all your thoughts today and thanks for taking the time to talk with us. It's been a pleasure.
Thanks so much.
Cheers guys, thank you.
All right, absolutely loved that conversation with Ryan. What stuck out the most to me was his answer that, honestly it wasn't scripted, but about whether he thinks true AI sovereignty is just a pipe dream when it comes to one country being able to create everything, you know, start to finish with our stacks.
Yeah.
Without a reliance on somebody else. And he essentially is like, "No."
Which is crazy because, is that really sovereignty? But I guess that's why they're, you know, Taiwan is a tense-
In play.
Yeah, exactly.
But in my head, I think that's what the United States is truly marching toward is to try, is that their idea of sovereignty?
I mean, if you end up having a country of geniuses in the form of these super intelligent, or very powerful AI models, you're not going to share any of that. Why would you?
Right.
Even, my question, which I didn't actually get to ask Ryan is, okay, yeah, we're talking about dependency, that's the best strategy for everyone else. But if you were America, if you were in that driving seat at the top, would you share, would you be that generous? I mean, they, right?
Well, and if history tells us anything.
Why should they be so-
Right, like generous, yeah.
Exactly.
So I would like to know though, from a U.K. perspective, your thoughts on everything that he shared about being reliant on a foreign entity.
So something which did change my mind a little bit was the whole conversation we had about adoption. So again, a lot of these CEOs of these companies, like DeepMind and Anthropic, I mean, I know Demis Hassabis said that the thing he would advise people is to like use these models themselves, because they actually might not know how you can get the best out of them. Because they're focused on building the next one and the next one, but they're not playing around with them necessarily on a daily basis, 'cause they just focus on building the next one. So actually we might be surprised by how much we can get out and the strategic advantage we might be able to gain, if you just adopt it quicker on like a mass level. So I think that's something which did convince me a little bit, but I just, I guess we don't really have any choice.
We shall see. All right, so, you know, is AI sovereignty real, or is the best that most countries can hope for, is really just being a well-treated client state. We want to hear from you, be sure to weigh in. You can weigh in in the comments below, let us know what you're thinking and let us know if you're living in the United States or China, who controls the AI in your country. We want to hear straight from you. Tell us in the comments below. Thanks so much to Ryan Wain and thank you to Laila for your input and for your thoughts today, and for joining us. And until next time, have a great one.