The West LOST: China’s AI & Robot Revolution is Already Unstoppable | Arnaud Bertrand

## Introduction and China’s AI model The conversation makes a stark claim: the robot age is already underway, but not where Western narratives of technological leadership would have it. Instead of an inevitable cascade of American-dominated AI breakthroughs, what emerges is a Chinese strategy built around diffusion, full-stack integration, and rapid commodification. The core of that strategy is straightforward and strategically profound: treat foundational models as public goods to maximize adoption and then compete on the application layer. If AI is indeed a general-purpose technology on the scale of electricity or the internet, the value accrues where it is applied, not where the infrastructure sits. That inversion—valuing diffusion over exclusivity—reshapes how we should think about competition between great powers. What matters in practice is not winning a contest over the lowest layers of the stack—chips, datacenters, and model weights—but enabling millions of firms and developers to embed intelligence into real products and services. The Chinese approach, illustrated by recent open-source models and aggressive hardware integration, is designed to accelerate precisely that diffusion. By contrast, the U.S. posture—focused on controlling access to advanced models and chips—looks like an attempt to weaponize scarcity. The conversation suggests that such a posture risks both economic inefficiency and strategic backlash: if the largest consumer market is blocked from access to Western models and components, it will build its own stack and sell that capability outward. ## US control strategy and open source A recurring theme is the difference between control and diffusion as strategic choices. The United States has leaned into a control strategy—export restrictions, cloud-access blocks, and a broader regime of tech containment—arguing that denying adversaries access to advanced AI capabilities protects national security. But containment is a blunt instrument. The conversation points out that when the U.S. restricts access to its cloud models or state-of-the-art chips, it paradoxically accelerates the indigenous development of alternatives in places like China. The U.S. choice to limit market access creates incentives for competitors to close their own technological loops. Open source is the counter-move. Chinese players are increasingly offering models and toolchains with permissive availability, lowering adoption costs and enabling local customization. For many firms worldwide, the calculus is simple: why pay many times over for an API you do not control when you can deploy an open-source model on domestically controlled hardware? Open-source diffusion has political and economic consequences. It democratizes capabilities, reduces dependency on any single vendor, and changes the bargaining position of states and firms that previously relied on Western cloud services for AI. Far from being a neutral technical choice, open sourcing becomes a geopolitical lever. ## Friendshoring Europe and the AI race “Friendshoring” has become a buzzword capturing a desire among Western policymakers to confine critical supply chains inside a bloc of allied states. The conversation unpacks why this policy is both attractive and fragile. On paper, building a coalition of chip-producing allies—Japan, the Netherlands, Taiwan, South Korea—could create a secure Western supply base. In practice, friendshoring faces structural limits. Alliance politics are messy; economic incentives pull states in multiple directions; and U.S. rhetoric that frames AI as a tool of domination alienates potential partners. European states face a particular dilemma. They are technically allied with the United States, but their security and economic interests do not always align with an American project to suppress a competitor. For many European governments, a multipolar balance offers diplomatic leverage and market opportunities. Relying on U.S.-centric AI solutions—especially those that channel data to American platforms—poses sovereignty concerns. Thus friendshoring efforts risk pushing partners to favor alternatives that promise technological autonomy, even if those alternatives are Chinese. In short, friendshoring can only go so far when partner incentives diverge and when alternative value propositions—open, cheaper, and locally controllable—exist. ## On-device AI and humanoid robots One of the most consequential trends discussed is the move toward on-device AI and the explosive rise of robotics, including humanoid designs. On-device inference reduces reliance on centralized cloud platforms, mitigates latency and privacy concerns, and broadens the set of contexts where models can be embedded—everything from factories to homes. When models run locally on domestically produced chips, the architecture of dependence changes fundamentally. China’s growing prowess in robotics—both industrial and consumer-oriented humanoids—adds a new layer to this transformation. The conversation emphasizes that Chinese firms are already producing robots at scale and price points that make them household-level goods in domestic markets. That commercialization matters because it shifts robotics from a boutique, defense-adjacent domain into everyday economic productivity. Once robots and on-device AI become ubiquitous in manufacturing, logistics, retail, and domestic life, the side that owns and normalizes those deployments accrues long-term advantages in productivity, data collection, and standards setting. The West has credible technological capabilities, but the rhetoric and policy choices examined in the exchange suggest it is not yet winning the race in deployment or affordability. ## Southeast Asia between two systems Southeast Asia emerges in the conversation as a critical geostrategic hinge. The region sits squarely between two technological ecosystems and faces a balance-of-benefit calculation: which system provides better infrastructure, cheaper services, and less intrusive governance constraints? Many countries in Southeast Asia prize growth and pragmatic access to technology more than ideological alignment. For them, open-source Chinese models that can be deployed locally, with lower costs and fewer strings attached, are an attractive proposition. This balancing act is not simply transactional; it reflects divergent threat perceptions and development priorities. Some Southeast Asian states see Western containment as a geopolitically motivated policy that could limit their economic options. Others worry about dependence on any single external actor. The key point is that the region will not automatically align with U.S. tech policy. Instead, it will choose the path that best serves domestic modernization, and today’s market realities—affordable models, integrated hardware stacks, and regional commercial ties—make non-Western options compelling. ## Asian payments and SWIFT The conversation expands beyond chips and models into payments infrastructure—a less glamorous but equally consequential battleground. The dominance of SWIFT and dollar-clearing has been a pillar of Western financial influence, enabling sanctions and economic coercion. Asian actors, led by China, have been developing alternatives: domestic payment rails, cross-border systems, and bilateral clearing mechanisms that reduce exposure to Western financial chokepoints. These efforts matter because technological and economic independence in payments underwrites broader strategic autonomy. If a state can settle trade and finance outside dollar-based systems, the leverage of sanctions shrinks. The emerging picture is one where industrial policy, payments architecture, and digital platforms interlock: countries that can operate outside Western control in finance and in digital infrastructure gain real room to maneuver in international politics. ## China challenges US sanctions Linked to payments is the broader theme of sanction

The West LOST: China’s AI & Robot Revolution is Already Unstoppable | Arnaud Bertrand The robot age has long begun. Just not in the West, where there is still not a single company able to produce what China is already selling as a household good. What does that mean for the future of the US empire and the multipolar world? Today, I'm talking again to Arnaud Bertrand on China’s open source AI push, US tech controls, Europe’s growing dependence, drones and humanoid robots, Southeast Asia’s balancing act, new payment rails in Asia, China’s response to US sanctions, and the wider shift in global power around Iran, finance, industry, and the fading force of Western control. Links: Arnaud Bertrand on X: https://x.com/RnaudBertrand Arnaud Bertrand on Substack: https://arnaudbertrand.substack.com/ Neutrality Studies substack: https://pascallottaz.substack.com (Opt in for Academic Section from your profile settings: https://pascallottaz.substack.com/s /academic) Merch: https://neutralitystudies-shop.fourthwall.com Donation: https://neutralitystudies. com/donate Timestamps: 00:00:00 Introduction and China’s AI model 00:07:32 US control strategy and open source 00:15:51 Friendshoring Europe and the AI race 00:27:01 On-device AI and humanoid robots 00:32:17 Southeast Asia between two systems 00:35:56 Asian payments and SWIFT 00:42:40 China challenges US sanctions 00:52:24 Iran war and Europe’s decline #Pascal Welcome back, everybody, to Neutrality Studies. This is Pascal, and today I'm joined again by the wonderful Arnaud Bertrand. Arnaud, welcome. #Arnaud Bertrand Hi Pascal, thank you for inviting me again. #Pascal It's good seeing you again. You haven't been on the show for quite a couple of months, which is not because I don't want you. I actually adore all of your writing and analysis that I read almost daily on Twitter, and I don't know how you do it. So congratulations on everything you put out. #Arnaud Bertrand Thank you. That means a lot coming from you. -- 1 of 16 -- #Pascal All right. It's quite fantastic. You're one of the most prolific analysts out there, actually, on Twitter and Substack. And you recently have been working on AI, and I thought maybe we can start with that one. It kind of flew a little bit under my radar. I only read that China came out with a new DeepSeek model, or, well, the company behind it. It's not China itself anymore, but the Chinese AI approach. In what sense is it different from the Western approach, and why to you does that matter? #Arnaud Bertrand Yeah, I mean, it's really a topic that fascinates me also because in my main job, which is actually not writing on Twitter or Substack, I'm a developer. I mean, I'm an entrepreneur and I code every day for my company. So I use a lot of AI, and I get to test the different models, both the U.S. models and the Chinese models. And it's fascinating to see the contrast in both approaches. I think the Chinese approach, the more things advance, the more it actually shows a very smart strategy. So the other day, you had Jensen Huang, who is the NVIDIA CEO, go on a podcast that made a lot of noise, and he described AI as, in his words, a five-layer cake. So you have five layers to AI, starting with the lowest layer, energy. And then, you know, you have chips, the infrastructure, and so on. The last two layers, as he described it, were the models, so the AI models. So, you know, on the American side, you have ChatGPT, Claude, and so on. On the Chinese side, you have, you know, DeepSeek, Kimi, and so on. And then the last layer was the application layer — so what you actually do with the AI. And his point is that AI is a general-purpose technology, much like electricity or the internet or, you know, the phone. So we've had a few of those in the past. With those general-purpose technologies, it's always the application layer that creates the value. So if you think back about electricity, for instance, did it matter much who made the actual energy infrastructure, who owned the electrical power lines, and so on? Not that much. What mattered is who made the fridge, who made light bulbs, and so on and so forth. What created the value was the application layer for that. Same thing for the internet. The companies that became insanely big with the internet weren't the telecom companies that provided the infrastructure. It was the Googles, the Alibabas, the Amazons, and so on and so forth, who actually used the internet, made the applications. There is no reason to think that AI is going to be any different. I think we're going to look 10, 20 years from now and be like, this whole AI race around the lower layers, positioning as whoever wins the infrastructure is going to win the AI race, is going to sound extremely stupid. Because what really matters is, at the end of the day, how you use AI. Whoever wins, to the extent that there is a race, is whoever can use AI in their society in order to gain more efficiency, productivity, and so on and so forth. And that's why the American approach is a bit strange, because if you think about it, what that means is that what matters is diffusion in order to create the most value. So you want everyone out there to have this general-purpose technology, because then the application layer can reach the -- 2 of 16 -- most people, right? So it's like, if you think back to electricity, it would have made no sense to say, okay, we're going to win the electricity race by forbidding anyone else to have the light bulb, to have electricity. So we are the only ones who can, you know, have lights, basically, because then all those big companies that did create value on electricity, like GE and so on, would have a much smaller market if only the Americans could have electricity. They made money by making sure that everyone had access to this infrastructure, and then they could sell the application layer to everyone. But that's the approach that the Americans are taking for AI, saying our models are really what matters, and so we don't want the Chinese to have them. We don't want any, you know, non-aligned nations to have them. So it's a bit of a strange take. Whereas the Chinese approach is, on the contrary, to offer the models free and open source to everyone, to ensure maximum diffusion, because they're taking much more of the general-purpose technology approach by saying, OK, it's a general-purpose technology, diffusion is what matters. Here are the models, everyone, you can have it for free. Let's focus on the application layer. #Pascal Hey, very brief intermission because I was recently banned from YouTube. And although I'm back, this can happen anytime again. So please consider subscribing not only here, but to my mailing list on Substack. That's pascallottaz.substack.com. The link's going to be in the description below. And now back to the video. I mean, the United States' strategy in a lot of areas has always been to gain complete market control, right? Because that then—global market control—that then gives you a huge lever. And we've seen the United States using this lever over and over and over again in the past 10, 15 years with sanctions and sanctions. You know, trying to then, you know, the whole SWIFT system just kicked countries off, with the expectation that their economies would crash. We've seen how they tried to do that, or how they actually did that to Iran in January—tried to just destroy the Iranian economy overnight by market manipulations, centrally done from coordinated forces from Washington. And this is, of course, only possible if you are in firm control over certain market segments and also technologies. And they seem to try to replicate that here, but then the Chinese approach now is a completely different one, apparently. So, also not trying to control everything, but just let it go, including open-sourcing stuff, huh. #Arnaud Bertrand Yeah, exactly. I mean, if the Americans' goal is to control everything, then it's failing dramatically. Because, you know, Jensen Huang just gave an interview in a magazine and, by his own words, NVIDIA, which is the biggest American chip company, is down to 0% market share in China. He said that. It has 0% market share. And this is directly linked to their approach, which is, of course, you know, you Chinese can't have our chips, you can't have our models. So if from China you try to access anthropic.com, which is the place where you get to use cloud AI, or if you try to go on -- 3 of 16 -- ChatGPT and so on, you can't access those models. It's not China banning those models, it's the Americans banning those models in China. #Pascal I didn't know that. I thought it was the Chinese Great Firewall, because I actually was just in Shanghai for a couple of days and I couldn't. I had to route my traffic through my Hong Kong eSIM in order to use Anthropic. Ah, it's the Americans who block it. #Arnaud Bertrand Yeah, absolutely. It is the American side who blocks those models in China, and they do the same for the chips and so on. So, you know, by definition, that gives you zero control because you're saying the largest market in the world when it comes to semiconductors—which is a fact, China actually buys more semiconductors than any other country in the world, including the U.S. We voluntarily... well, voluntarily is the American government. Actually, the NVIDIA CEO is extremely pissed off about that. But they're saying we voluntarily don't sell chips to them, at least the most advanced chips. And so what they're forcing China to do, which China actually did, is to build... and so by that action, they're helping, they're very much encouraging China to become a mighty competitor, which they are in models. So you consistently see the Chinese models competing, you know, keeping track with the American models. So, including if you look at the latest DeepSeek V4, if you look at most of the benchmarks, it's on par with the American models and increasingly on chips. So, the DeepSeek V4, for instance, can run on the latest Huawei chips, Ascend chips, and it can run very well. So right now you have the Chinese in control of their full stack, of their five-layer cake, which we were speaking about earlier, and what they will do and what they're increasingly doing as well is going to other countries with an incredibly strong value proposition, which is we have the whole stack, the same as the Americans, so the five-layer cake. But ours is open source, so you guys are in full control. You can even modify the model and so on. We don't care. And it's 30 times cheaper, by the way. So that is their value proposition, which is insanely strong. It's literally, at this stage, you are a complete idiot if you have a company like iView and you're paying for cloud or OpenAI services. API, you're a complete idiot because you're paying 30 times more for the same thing and you give all your data to OpenAI or Anthropic. Whereas if you have a Chinese open source model on your own server, they're not communicating with China at all. You're in full control of your customers' data. #Pascal So China is by now providing the planet with a public good. Isn't that quite fascinating? I mean, it used to be the United States that had this kind of approach, at least some of the companies, to do -- 4 of 16 -- things open source and provide. But now the US government is stepping in and making that kind of model completely impossible, thereby actually increasing China's digital sovereignty and the abilities they have. I mean, isn't that just increasing the gap between, well, the attractiveness of actually the two development models? #Arnaud Bertrand Yeah, it's very much increasing the value proposition that the Chinese are providing, that's for sure, by contrast with them. I mean, it's not the first time in history that this happens. We saw the same thing between the UK and the US, actually. I think it was... I can't remember the technology exactly. I think it might have b