EPISTELEM

  • Français
  • English

    • Auteur/autrice : Rémi Bourgeot

    • The Geoeconomics of AI Scares

      The Geoeconomics of AI Scares

      This piece is published in partnership with the French Institute for International and Strategic Affairs (IRIS).

      The debate about AI’s existential threats can no longer be separated from economic and geopolitical interests. It sheds light on the nature of the global AI race, between the US and China, and between closed and open models.

      Real risks deserve proper assessment and regulation to prevent malicious use. However, progress is also needed to achieve better and safer models. Successive warnings about extreme scenarios inspired by science fiction, in which AIs turn against humanity, appear intended to prompt a type of regulation that would limit domestic and international competition, especially from open models.

      A slowdown forced by regulation could help to preserve the unsustainable financial dynamics of leading model providers like Anthropic and OpenAI ahead of their IPOs. This trend undermines necessary efforts toward regulatory coordination, since it has now reached the point of spurring a global backlash. Even partner chip companies like Nvidia, which have a more resilient business model, have distanced themselves.

      The controversy points to control of AI’s economic value. While closed American AI proves vulnerable to the collapsing price of inference (model usage) and the rise of increasingly powerful Chinese models, China is pursuing a strategy inspired by open source, which Europe too should have implemented to develop its autonomy.

      The Nature of AI Risk: From Sci-Fi to Policy-Making

      Former and current Anthropic employees, followed by CEO Dario Amodei and OpenAI executives, repeated warnings about the possibility of AI causing human extinction. However, claims that an AI could simply replicate itself indefinitely to escape control, self-improve and take over various systems tend to ignore the hardware and infrastructure constraints involved.

      Questions already began to arise in recent months with a series of romanticized disclosures about « rogue agents » escaping testing environments or building “AI civilizations.” The communication around these events obscured faulty engineering behind the threat of human extinction. The Hugging Face platform, which hosts open models, was one of the organizations targeted in such incidents, just weeks before it was acquired by Nvidia.

      Though the risk related to swarms of autonomous agents is obviously real, such usage would rather be linked to criminal intent with access to massive compute infrastructure. The fact that agents that are given excessive access can cause damage does not demonstrate that they have developed a will to escape or control human civilization.

      No plane manufacturer would reasonably claim that an aircraft developed a malicious conscience, decided to overrule the pilots’ instructions and harm its passengers. The fact that modern planes are highly complex mechanical and electronic systems that can operate with limited human intervention does not imply that they have acquired “agency” or the mental abilities of a bird.

      Autonomous systems can create dramatic security problems. AI agents can be given excessive permissions, make mistakes, exploit poorly protected systems and be used in cyberattacks with unprecedented efficiency. These risks will be better managed if safety efforts are not distracted by fanciful philosophical claims.

      Regulatory Capture as a Solution to AI’s Economic Conundrum

      Leading players can use existential warnings to push for a type of regulation that they would influence, creating a situation of regulatory capture. The companies that already have the computing resources, researchers and capital to develop frontier models are also those best positioned to comply with expensive new regulations and comfortably implement a pause. Barriers that appear to be designed to protect society can therefore make it more difficult for new entrants to challenge the status quo and actually develop more reliable and safer models.

      This issue is all the more crucial since the economic model of leading US AI companies appear uncertain. OpenAI and Anthropic are spending gigantic amounts of capital on training, inference and computing infrastructure. They need to turn this spending into a sustainable business model, especially as both companies have declared their intent to go public. However, the models are becoming increasingly standardized (or “commoditized”) and token prices are collapsing.

      Agentic AI generates far greater demand for inference because agents can perform long sequences of actions and consume large quantities of tokens. But this does not necessarily mean that OpenAI or Anthropic will capture the value created by that demand. The same compute can be used to run models that are cheaper, open or adjusted by companies themselves, which can then run it on dedicated servers. The model does not need to be rented from its original provider.

      The AI boom has created a distinction between the companies developing models and the companies supplying the infrastructure. Nvidia is at the center of this system both from an industrial and financial perspective. The GPUs it designs are required to train and run AI models, within a broader ecosystem including data centers, electricity, memory and semiconductor manufacturing.

      The interests of chip companies and LLM providers thus differ. Nvidia benefits from the continued expansion of compute. Jensen Huang’s relatively positive stance on open source can be understood in this context. Open models can create additional demand for dedicated servers and computing infrastructure. A slowdown can be presented as a response to existential risk, but it can also reduce capital expenditure. A temporary reduction in the pace of the race could help a rebalancing of revenue between model providers and chip companies.

      China’s Open-Source Challenge Upends US Bubble Dynamics

      China has been moving in the opposite direction. Chinese companies have increasingly focused on open models as a way to catch up with the US. DeepSeek became the most obvious example, but it is part of a wider ecosystem that includes Alibaba’s Qwen, Moonshot AI’s Kimi and Zhipu AI’s GLM models. Cheap Chinese models change the economics of AI. They can be downloaded, adapted and deployed locally, allowing companies to build their own systems. This is particularly important for many users from the Global South to US multinationals wary of their AI bill. China can thus develop its own technological sphere of influence.

      There is also an important difference in the structure of the Chinese and US AI industries. Both countries have large digital companies that can finance AI development as part of broader businesses involving cloud computing, infrastructure, and, in the case of China, the development of domestic alternatives to Nvidia chips. Both countries also have specialized AI labs. However, in China, specialized labs such as DeepSeek and Kimi, while important, do not constitute the systemic center of an asset bubble like in the US.

      Besides, China has little reason to trust that a pause would be implemented in the current geopolitical environment. The strategy adopted by OpenAI and Anthropic undermines the efforts to find common ground on AI regulation, which was supposed to be at the center of upcoming talks between the two governments.

      Safety Requires Open Innovation

      The technological foundation of modern AI came from open research and an ecosystem in which companies like OpenAI and Anthropic could use publicly available research and data. They have now built closed systems and have an interest in protecting their position. Their claims that they are approaching AGI and solving the most advanced mathematical problems, while still heavily drawing on human expertise, are part of the attempt to demonstrate that their systems reached a unique level of technological advancement. They need investors to keep believing that today’s spending will lead to a profitable oligopoly.

      However, progress cannot stop, precisely because LLMs are highly imperfect and risky. Good, rational research and engineering can make them both more effective and safer with appropriate regulation. Future progress will not necessarily come only from scaling existing architectures with more computing power. Different approaches to data and physical AI can produce different advances.

      Europe’s Regulatory Impasse

      Europe too had the possibility of using open source to catch up with the US without reproducing its entire capital expenditure trends. Mistral was relatively well positioned as its founders came from major digital companies and entered the field early. But the EU’s AI Act raised regulatory barriers when Europe needed more open innovation. Bureaucratic rules and jargon detached from technical reality can be beneficial to the administrations in charge and to strong competitors in other world regions. Several national governments, including those of France and Germany, tried to persuade Ursula von der Leyen to revamp the AI Act before its implementation, to no avail. By comparison, in the summer of 2025, Donald Trump faced little resistance when detailing the amendments that the US ecosystem had requested, amid broader trade instructions.

      Besides, Europe has a major strategic position in semiconductors through ASML, which produces lithography machines sold to semiconductor manufacturers like TSMC in Taiwan. Europe could have developed a more ambitious and integrated strategy in semiconductors, AI infrastructure and models. While China was pursuing the opportunity linked to open source with determination, Europe once again deepened its dependence on US technology.

      Conclusion: Regulation for Safety, not Oligopolies

      Regulation should address actual risks rather than developing new prerogatives based on science fiction. It should avoid giving leading AI providers the power to define the regulatory framework in ways that secure their own market dominance. The current controversies therefore need to be understood in their economic context. The issue is whether demand for AI compute will translate into profits for model providers, allowing them to survive the bursting of the bubble, or primarily for chip companies, and whether open models will redistribute the economic value of AI.

      China’s strategy combines open models with the development of domestic chips. It does not need to reproduce the American system, but to build an alternative capable of producing sufficiently powerful AI at lower cost and distributing it widely. The political reaction to AI risks is consequently also part of a much broader race for technological and economic power.

    • Conflicts: Are Economic Sanctions Effective?

      Conflicts: Are Economic Sanctions Effective?

      I took part in Radio France Internationale’s Today’s Debate, hosted by Romain Auzouy. Below is a transcription of my contribution to the discussion.

      RFI: Rémi Bourgeot, this “Economic D-Day,” in Donald Trump’s words, so this new U.S. strategy toward Iran, which obviously concerns Tehran first and foremost, but also, in order to target Iran more effectively, all countries that would maintain relations with Iran: do you think this could ultimately have more of an impact on the regime in Tehran than the American bombs?

      Rémi Bourgeot: That is how Donald Trump presents it to the rest of the people shaping opinion in Washington, and particularly to neoconservative circles. He needs to tell them that he is continuing this drive to crush Iran. Since it has been a huge failure militarily, he claims he is going to continue the war on the economic front. Then, when you look more closely, there is a huge gap between these stated objectives, which are to economically suffocate Iran, and the reality of what the United States can actually do in terms of additional sanctions.

      We are talking about secondary sanctions, about threatening Iran’s economic and commercial partners so that they stop trading with Iran as much as possible. So one very important country in this equation is China, which is Iran’s main partner when it comes to hydrocarbon imports, and also when it comes to supplying Iran with all kinds of components, particularly for industrial and military purposes.

      For China, it is clear that there is no question of enforcing these secondary sanctions. It has instructed its companies, particularly its banks, not to be intimidated and not to implement these sanctions. In a way, it has already been through this with the trade war, by rejecting certain American threats and threatening to restrict rare-earth exports last year. So on that basis, it is responding very differently than it did a few years ago to the question of enforcing sanctions, and it is rejecting them.

      And on that basis, we are seeing a number of countries that are fairly close to Iran commercially following the same path. Even Pakistan has said it will not enforce these sanctions, these threats of secondary sanctions. So that gives you an idea of just how limited the impact actually is. Turkey is in NATO, at the same time it has relations with Iran, but it has absolutely no desire to follow the United States down this road. People say Iraq would be the case, but that also shows how limited the effect is, in a way.

      Then, I don’t think it is unreasonable, from Donald Trump’s personal point of view, to go and tell those neoconservative circles I was referring to: “I’m carrying on.” We can see that he is stuck. There is a kind of regime crisis unfolding in Washington. He knows he is in the middle of a fiasco, that going to war with Iran was a huge mistake. He has no military way out, and this can partly be a way of escaping this situation, this military dead end, by pursuing the war more on the economic front.

      But beyond that, I don’t think he is particularly interested in the actual effect, which is limited. Obviously, we should remember that Iran is already under very substantial sanctions and is economically strangled in any case. But this idea of targeting all the relationships, all the trade that Iran currently has, particularly with China, remains a very distant prospect.

      It is perhaps more of a political act, then, something the United States knows well: the U.S. has been sanctioning Iran virtually since 1979. Doesn’t China’s attitude already show the ineffectiveness of sanctions, which you were talking about? For example, China buys 90% of Iranian oil. So, once China is able to ensure that these sanctions don’t affect it, particularly in Beijing’s case — this is the whole issue with secondary sanctions, which cut off access to the dollar, the essential channel for international transactions — but China uses the yuan through its own financial system that it has built. Does all of this mean that, whatever form they take, sanctions cannot bring about the kind of economic strangulation the United States claims they can?

      You have the situation linked to the Strait of Hormuz, where exports are nevertheless limited. You have the American counter-blockade, so there is still some traffic taking place, but it remains very limited. Then we can see an architecture adapting to these American threats, and above all to their erratic nature. So you have this architecture becoming more firmly established in terms of payment systems.

      But then there are also issues of trust and dependence that remain very significant when it comes to the dollar. But if we take a broader view, we still have this enormous bloc centered around China, with Russia as well, which is a crucial factor when it comes to sanctions.

      Now, with the Iranian front, we have a huge bloc that has come together. It is not completely unified. Obviously, each country is defending its own interests, but they interact, they trade, they provide one another with essential things, and above all, they now have formidable industrial capacity. That is what is lacking on the Western side, with this crisis in industrial production, and that is what is on the minds of all decision-makers, particularly Donald Trump.

      There is the political message, as you said. It is interesting to bring up China’s attitude toward Russia. China is much more discreet in the Russian conflict than it is when it comes to Iran. Would you say that, ultimately, perhaps we have never had so many states using — or at least, perhaps, economic sanctions have rarely been used so extensively — to threaten states, to distance them, and never to pursue what Michel Duclos was saying? Ultimately, perhaps sanctions have never been circumvented to this extent, as in China’s case, which may be at the very heart of this issue, because China is the country countering the sanctions weapon today?

      The idea of sanctions really became significant during what we call the unipolar moment, after the collapse of the Soviet Union, with this overwhelming power of the United States, and with this idea of economic warfare, of shaping the world through interventions in some cases, but also through all kinds of economic measures, particularly sanctions, economic warfare. And since then, a lot has happened.

      We have seen the emergence of certain countries, major industrial and technological powers like China, and even countries such as Iran and Russia have also developed their own industrial and technological momentum. They train a higher proportion of engineers than Western countries. That is the direction they are moving in today, and we are seeing a kind of convergence, in fact, taking place.

      We see it particularly in terms of military production capacity, where there is a huge problem on the Western side in the various wars being fought in Ukraine or against Iran, or on all sorts of issues. On the other side, we have more of a crisis of overproduction centered on China, and a whole group of countries that are part of this industrial and technological development dynamic, that train huge numbers of engineers and still know how to make use of them.

      And so sanctions are, in a way, somewhat anachronistic when it comes to these countries. Sanctions were designed at a time when these countries were much less developed, much less resilient. Since then, they have not only developed considerably in industrial and educational terms, but obviously they have also learned how to circumvent sanctions in all sorts of ways.

      For example, Russia, at the beginning of the Ukrainian affair, around 2010–2014, was still at a different stage of development. It was much less interested in industrial issues. The sanctions actually pushed it to make an effort in that direction and really set this dynamic in motion.

      So obviously, sanctions do have an effect, but they also encourage countries to be inventive, to find ways around them, through different financial means and different channels, but also by going much further in terms of educational and industrial development.

      We said that it obviously impoverishes people. The first group affected by these economic sanctions is the population; they become poorer. Can this turn populations against the regime in question? Could that also be the objective of sanctions?

      That is the objective. But in Iran’s case, for example, we can see that the threat is existential. Trump has not hidden that. At the beginning, we were talking about Iranian women football players and I don’t know what, and a few weeks later Trump was talking about annihilating Iranian civilization. So we were no longer really in what you might call a humanitarian discourse.

      So in Iran, what we have seen instead is a rallying effect. That is often what happens in the case of massive bombardments, when a nation faces an existential threat.

      And today, there has indeed been a change of regime, not in the sense of the neoconservative theories, but the regime has effectively changed, and today we have more of a military dictatorship, with these figures coming from the Islamic Revolutionary Guard Corps, these veterans of the Iran-Iraq War, who are much less religiously conservative, who have also relaxed the enforcement of religious rules considerably, particularly when it comes to wearing the veil. We see many women who are not veiled in Iran’s major cities, particularly in Tehran.

      So what we have seen is more of a form of partial unification, obviously, but a form of rallying by the population in the face of the existential threat posed by the United States and Israel. That is more on the military attack side.

      Then sanctions do play a role, of course, but without this whole context over the past twenty years, one could imagine that an emerging middle class, much more visible in Iran, with precisely these ideals of political change, might have led the regime to adapt much more, or even to genuinely change.

      This whole highly antagonistic context with the United States has probably tended instead to keep the regime in place.

      The regime in place, yes, because one could also have imagined that, in this desire to impoverish the population, there was this idea that, effectively, sanctions make that situation even worse.

    • Economic War against Iran: Rhetoric and Deadlock

      Economic War against Iran: Rhetoric and Deadlock

      Interview on France 24 in French – Read the full English transcript below the video.

      Rémi Bourgeot is with us live to discuss the US sanctions against Iran. Hello, and thank you for joining France 24. You are an associate researcher at IRIS. « It will work, » says Scott Bennett. « It worked in Venezuela after the blockade. It worked in Cuba, and it will work in Iran. » Is it really that simple? Can weakening Iran’s economy actually weaken the Iranian regime?

      Iran’s economy is indeed severely weakened, but there is a massive gap between the US announcements and the claim of waging a total economic war. The main obstacle to this aspiration is China, Iran’s most important trading partner. China is both a major importer of Iranian oil and a supplier of essential components to Iran.

      Yet China refuses to comply with these sanctions. Given the current state of Sino-US relations—and China’s previous pushback against US threats, even commercial ones—Washington cannot afford to take an offensive stance against China with so-called secondary sanctions.

      So there is a real disconnect between the announcements, the intended effect, and reality: the measures remain limited. Nevertheless, taken together, they are already very significant and exert enormous pressure on Iran’s economy, particularly through the US-imposed blockade.

      If I understand correctly, Rémi Bourgeot, these are just empty threats? They won’t actually be enforced?

      There is a sense of desperation on the US side. Donald Trump knows the military war is lost. It’s a strategic disaster: the US has lost much of its effective presence in the Middle East.

      The idea is to announce strong measures to reassure neoconservatives in the US that the offensive continues, while at the same time trying to disengage militarily.

      In terms of actual sanctions, Iran’s trading partners refuse to enforce them. This was not the case a few years ago, when US sanctions threats—particularly against banks in these countries—created real panic, and the sanctions were effectively followed.

      Today, in China, the government prohibits banks from reacting and thus from enforcing these sanctions.

      Other key players include Turkey, which cannot afford to follow US orders as it has no interest in doing so. The United Arab Emirates, for their part, were more aggressive a few months ago, urging the US to « finish the job, » but they have also sought to find a modus vivendi with Tehran.

      Overall, the US lacks the means to impose massive secondary sanctions on Iran’s trading partners. This means that China, Russia, or Turkey, if they do not comply with these sanctions, will not face US retaliation.

      These countries are in very different positions relative to the US. China, for example, has been in a standoff with the US for two years now: it rejects some US sanctions and prohibits companies—especially banks—from complying with them. We saw this during the trade war, when China threatened to restrict rare earth exports in response to US tariff threats.

      This standoff has now extended to geopolitical sanctions targeting other countries. China is building on this momentum and banning the enforcement of sanctions on its soil.

      The measures remain limited to networks directly linked to Iranian trade. But the idea of blanket sanctions—targeting anyone with even the slightest financial connection to Iran—is unacceptable to China. Moreover, Beijing has the means to reject and retaliate if the US pushes in that direction.

      Other countries take a different approach. Turkey, a NATO member, is close to the US but also pursues its own interests and has no intention of blindly following US orders. That is not its agenda.

      As for Russia, already under sanctions, it maintains a special relationship with Iran, particularly in military terms.

      One last question, Rémi Bourgeot: Who currently controls the Strait of Hormuz?

      The Iranians effectively control it, but the US adds its own additional blockade. Traffic is currently very low, especially toward Asia, but the situation is not completely blocked.

      Washington went all-in with this summer’s offensive, but it was another resounding failure. Even economically weakened, Iran retains military and industrial capabilities—particularly through its ballistic missile and drone programs—which prevent the US from resolving the conflict militarily.

      Today, we see a shift in US rhetoric toward economic pressure. But after two years of trade war, the US’s actual leverage is very limited.

      This is a way for Donald Trump to send signals to neoconservatives as he tries, step by step, to extricate himself from this desperate situation.

      Thank you very much, Rémi Bourgeot, for joining us today.

    • Behind the US Defeat in Iran: A New Industrial Paradigm

      Behind the US Defeat in Iran: A New Industrial Paradigm

      Opinion piece published by Les Echos – 22 July 2026

      The conflict with Iran has exposed a shift in global power dynamics—one less tied to the sophistication of weaponry and more to industrial capacity. Superiority no longer depends solely on cutting-edge technology or GDP but on the ability to design, manufacture, and integrate diverse components rapidly, then deploy that firepower in service of a coherent strategy. What is most striking is not Iran’s performance—despite its economic struggles—but the United States’ failure to adapt to this new reality.

      Western economies struggle to translate their historic scientific lead into tangible production. Weapon shortages are a symptom of this. The misallocation of resources into financial bubbles—AI, real estate—at the expense of industrial expertise only deepens the problem. Yet the often-underestimated Generation Z represents a potential turning point.

      Engineering as a Source of Power

      Emerging nations are leveraging their growing base of versatile engineers and technicians. In Iran and Ukraine alike, drones, ballistic missiles, and electronic warfare systems exemplify this approach: mass-produced, simple devices prove more decisive than a limited number of ultra-sophisticated systems. The Shahed drones, for instance, owe their effectiveness to a modular industrial architecture—standardized components, accessible manufacturing, and continuous improvement.

      This operational necessity has even forced the United States to rethink its drone and missile production, shifting toward cheaper, more scalable models. Innovation now emerges not just from research labs but from production lines and the battlefield, where the ability to rapidly adapt systems has become a critical advantage.

      The depth of technical expertise matters more than the stock market valuations of a few firms. Emerging nations—particularly the geopolitical bloc of China, Russia, and Iran—now produce a proportionally higher number of scientific and technical professionals than the West. Organizations led by individuals with a technical background make decisions grounded in technological realities rather than bureaucratic or political logic.

      Modularity and Industrial Sovereignty

      Supply chains are now structured around modular architectures, combining electronics, embedded software, sensors, or radio frequency components, often sourced through international networks. China occupies a central role, lowering entry barriers for many countries. Full mastery of every industrial process is no longer essential; the advantage lies in integrating components and swiftly substituting critical elements. Iran has capitalized on this, relying on Chinese supply chains for affordable electronics while developing certain components locally under sanctions.

      Industrial sovereignty is no longer about absolute self-sufficiency but about resilience: identifying critical dependencies and ensuring production can continue despite logistical disruptions or infrastructure destruction. Economic warfare, once an American prerogative, has now been turned against them—whether through Iran’s control of the Strait of Hormuz or China’s dominance in rare earth materials.

      China’s progress stems from the gradual integration of value chains in key sectors such as batteries, electric vehicles, and robotics. While U.S. restrictions on semiconductors may temporarily hinder China, they also accelerate its investments in domestic capabilities, particularly around Huawei.

      The Decline of Western Productive Culture and the Potential of Gen Z

      The United States retains major assets, such as its university system and energy independence. Yet an increasing share of capital flows into financial activities disconnected from production. AI investments, though promising, often focus on consumer applications or ad-driven models rather than structural transformations of the productive apparatus. Moreover, cronyism often prevails, as seen in the civilian nuclear industry, where so-called « nuclear bros »—entrepreneurs whose nuclear physics expertise is barely more advanced than that of top diplomatic negotiators—exert undue influence.

      In Europe, the situation is even more concerning. Despite world-class scientific sectors, strategic decisions are increasingly driven by administrative or political logic, as evidenced by the shortcomings of the FCAS program or the limitations of the Chips Acts, which overlook the integration of value chains (electronics, materials, software). Failures in battery production, such as Northvolt’s collapse, highlight the urgency of a more incremental substitution strategy.

      Yet recovery remains possible. Despite educational challenges, Generation Z is developing a strong technical culture, often outside traditional institutions, thanks to unprecedented access to knowledge and open communities. European economies will not regain their industrial momentum through short-term spending plans or calls for financial unity but by empowering those who understand technology, control production chains, and can turn innovation into productive capacity. This is the battleground on which the hierarchy of global powers and prosperity will now be decided.

    • China’s AI Rise: Open-Source Diplomacy and Engineering Culture

      China’s AI Rise: Open-Source Diplomacy and Engineering Culture

      This piece is published in partnership with the French Institute for International and Strategic Affairs (IRIS).

      China’s rapid progress in AI challenges US leadership and the narrative of its “financial exceptionalism” that underpins the asset bubble. It demonstrates what can be achieved when open-source collaboration, technical expertise, and industrial strategy align. Rather than a genius breakthrough, this advancement is rooted in engineering skills, rational investment and industrial organization. China’s weak domestic demand also means it will increasingly include AI models and semiconductors as part of its mass export strategy.

      At the same time, China is pursuing a geopolitical ambition. By positioning itself as a champion of open-source AI, the country is leveraging technology as a tool for global influence, offering an alternative framework to proprietary systems built on unsustainable funding. This is striking given that open-source is ubiquitous in western digital technology and can even be seen as its foundation. Many U.S. AI labs, like the ill-named OpenAI, have drifted away from this logic amid the hype and financial excesses.

      Current developments rely on globally-shared engineering skills. The path to technological competitiveness is open to any nation – including those in Europe – that chooses to support broad scientific education, empower its younger generations, and reward inventiveness over bureaucratic inertia. With determination, resources, and industrial realism, catch-up remains within reach.

      The AI Race and the Open-Source Strategy

      The global AI landscape has been undergoing a shift far beyond Silicon Valley’s hyped labs since last year’s ‘Deepseek moment’. The United States still leads to some extent in cutting-edge AI models, which continue to set benchmarks in reasoning, coding, and multi-modal tasks. However, the rise of Chinese AI models like Moonshot AI’s Kimi K3 or DeepSeek V4 has introduced a new dynamic, with performance that rivals US models at a fraction of the cost.

      China is challenging its dominance in ways that seemed impossible to envision just a couple of years ago, based on the US exceptionalism narrative, but easier to anticipate with a pragmatic sense of traditional math and engineering skills. While lacking the funding scale of global AI labs, French startup Mistral demonstrated, even before Deepseek, that it was possible to develop a competitive model, initially with just about thirty top scientists.

      Chinese competition is economically and financially disruptive, to the point of challenging the stratospheric valuations of US tech giants. As a result, it is also starting to shake semi-conductor companies, notably in Asia, which have enjoyed a massive transfer of cash flow from American ‘hyperscalers’ spending unlimited amounts on compute and infrastructure.

      The cost advantage is particularly compelling in the era of AI agents, where iterative actions drive exponential token consumption. Companies are increasingly routing tasks to Chinese models to slash costs. In the race to scale AI, affordability is becoming nearly as important as performance.

      Beyond pricing, the philosophy and politics behind AI development is key. At the World AI Conference in Shanghai this July, Chinese leaders and tech executives positioned their country as the global champion of open-source AI. Unlike the US, where proprietary models now dominate, China framed its AI advancements as open, accessible, and collaborative, in contrast to the closed, corporate-controlled systems of Silicon Valley giants.

      By offering accessible, low-cost AI tools, China is courting the Global South and beyond. The idea is that AI should be seen as a public good more than a corporate asset. Of course, the devil is the details and it will be interesting to monitor how Chinese companies and policymakers will deliver on this commitment over time. However, it is striking that China is officially making its own the approach which has long been a discrete pillar of western tech, around Linux and far beyond. OpenAI was also originally launched on this promise.

      China’s open-source strategy serves three key objectives. It undermines US tech dominance by providing a viable alternative to proprietary systems. It builds a new tech alliance, particularly among developing nations that lack the compute resources for expensive models. And it lays the groundwork for a parallel tech ecosystem, one that reduces dependence on the U.S. and includes domestic hardware, alternative cloud platforms, and even standards-setting.

      While America relies on proprietary systems and export controls to maintain its edge, China is betting on openness and accessibility to win global influence. The temptation for the US to respond with a ban of those models risks backfiring, by dissociating its top companies from the achievements made in open-source AI, including at the national level, and leaving domestic users at the mercy of pricing schemes dependent on the ebb and flow of the bubble.

      Semiconductors and the Battle for Tech Sovereignty

      The AI revolution is not just about algorithms. It is fundamentally tied to hardware. Since 2022, the U.S. has progressively banned the export of advanced NVIDIA chips to China, citing national security concerns. These restrictions reached a fever pitch in June 2026, when the Commerce Department blocked all foreign nationals from accessing models like Claude Fable 5 and Mythos 5. The move forced a global suspension of these models, demonstrating the far-reaching impact of U.S. export controls.

      China’s response has been strategic. Huawei’s Ascend 910B, while still behind NVIDIA’s H200 or even H100, is now powering notable Chinese AI advancements. China’s leading semiconductor manufacturer, is inching closer to advanced nodes, though it still lacks access to top EUV lithography techniques mastered by Dutch firm ASML. To accelerate self-sufficiency, China has poured nearly unlimited money and effort into its semiconductor industry, aiming for a vast majority of domestic AI workloads to run on local chips within a few years. US restrictions, intended to cripple China’s AI ambitions, have rather accelerated its drive for self-reliance. Meanwhile, American firms like NVIDIA and AMD are progressively losing access to the lucrative Chinese market, as Jensen Huang recently acknowledged.

      This semiconductor race is not only about economic competitiveness, but also technological sovereignty. The US has long relied on its ability to control the flow of advanced chips to maintain its technological edge. But as China builds its own ecosystem – from chips to cloud platforms – it is creating a parallel market where its tech industry can thrive with fewer and fewer US components. The result points to an increasingly bifurcating global tech order, with implications on standards, supply chains, and alliances.

      These innovative strategies echo those employed by European nations – like France under Charles De Gaulle’s presidency – to catch up with American technologies, until a deep cultural shift allowed critical dependencies to go unchecked, culminating in the current industrial impasse.

      From AI to the New Military-Industrial Landscape

      The competition in AI and semiconductors is increasingly spilling over into military and strategic domains. It is visible in the rise of asymmetric warfare, where nations like Iran have demonstrated how low-cost, mass-produced systems can effectively challenge US and Israeli military superiority.

      China, too, is leveraging its technological advancements for military purposes. Its civil-military strategy ensures that breakthroughs in AI and semiconductors benefit its defense capabilities. Hypersonic missiles, autonomous drones, and AI-driven cyber warfare tools are examples of how China is combining commercial and military technology. This constitutes a key advantage in the new bloc confrontation, where the ability to rapidly deploy and scale advanced systems can determine the balance of power.

      Decades of deindustrialization have left the US and Europe dependent on global supply chains for critical technologies, from rare earth materials to advanced semiconductors. In the US and to some extent in Europe, chips acts are attempts to rebuild semiconductor capacity. Yet, for now, TSMC in Taiwan remains the linchpin of the global chip supply chain, leaving clients vulnerable to disruptions. Meanwhile, China’s increasing ability to produce its own hardware, however lagging behind some of the latest American designs at this point, gives it a degree of resilience that the US currently lacks. The democratization of military technology, driven by advances in AI, drones, and missiles, means that an increasing number of countries can now challenge traditional military superpowers.

      The Future of Tech Diplomacy

      As the AI and semiconductor competition intensifies, the Global South is emerging as a critical political battleground. The World AI Conference in Shanghai was not just a showcase for China’s technological prowess, but a platform for its technological diplomacy. By positioning itself as a leader in open-source AI, China is appealing to developing nations that are eager for affordable, accessible alternatives to Western systems.

      For many, the choice between expensive, proprietary American models and cheaper, open-source Chinese alternatives is a no-brainer. African and Southeast Asian startups, for example, are increasingly adopting Chinese AI models for applications in agriculture, healthcare, and finance, where cost and accessibility are essential. By offering training, funding, and infrastructure support, China is cultivating a network of partners that progressively reshape the global tech landscape.

      The future of international power will be determined by mastery of AI, semiconductors, robotics, and industrial production. Europe, in particular, must return to the engineering culture that drove its post-war industrial success – before mass bureaucratization, misguided industrial policies lacking scientific culture, and the importation of the US cultural crisis, without its top scientific programs or funding.

      Power will also be shaped by the nations who can offer the most compelling narrative for the world’s technological future outside of unsustainable financial trends. China’s open-source gambit is a clear signal. In a more multi-polar order, the ability to provide accessible, innovative, and inclusive tech solutions will be essential.

    • Behind the US Defeat Against Iran: A Crisis of the Industrial System

      Behind the US Defeat Against Iran: A Crisis of the Industrial System

      This piece is published simultaneously by the French Institute for International and Strategic Affairs (IRIS).

      The outcome of the US-Israeli war against Iran reveals a profound shift in power dynamics, centered on industrial capacity and a new paradigm for innovation. Superiority now depends less on the most advanced systems—or on GDP—and more on industrial organization. This organization must follow a modular logic: designing, modernizing, and producing systems rapidly by integrating diverse components and sustaining high replacement rates, to support a rational deterrent strategy.

      Beyond the strategic fiasco for the United States, what is most striking is not Iran’s military-industrial prowess—despite its economic constraints—but America’s difficulty in embracing this new global industrial paradigm. Western economies struggle to translate their historic scientific base into productive capacity. Chronic shortages of armaments are just one revealing symptom. The concentration of resources in asset bubbles, such as AI or real estate, plays a key role, as does the sidelining of essential skills. Yet, contrary to stereotypes, Generation Z—despite the educational crisis—would be capable of driving a rapid recovery.

      Industrial Capacity and Engineering Density

      Emerging economies, in all their diversity, increasingly rely on the classic skills of engineers and technicians—versatile and trained in large numbers. Many nations, like China, are pursuing industrial ascension on this basis, reflected in their economic policies. This shift is evident in military terms, both in Iran and on the Russo-Ukrainian front.

      Recent conflicts mark a crucial turning point. Drones of all kinds, ballistic missiles, and electronic warfare systems are reshaping the economics of war. The ability to mass-produce relatively simple devices is becoming more decisive than possessing a limited number of highly sophisticated systems.

      The appeal of Iran’s Shahed drones lies less in technological breakthroughs than in their industrial architecture. Design choices prioritize available components, standardized electronics, simple manufacturing, low costs, and continuous adaptation to operational constraints. Ballistic missiles follow the same logic of incremental improvement. This approach mirrors what emerged in the Russo-Ukrainian conflict: innovations do not come solely from design offices but also from production lines, the battlefield, and the ability to rapidly modify systems. This challenge is already pushing the United States to attempt a new approach to drone and missile production—massive and less costly.

      Behind the fantasy of replacing engineers and coders with AI, this evolution places the skills of younger generations at the heart of the race. Innovation depends on a vast number of engineers, technicians, and geeks of all kinds, capable of solving new problems. Emerging countries—particularly China, but also Russia and Iran—now train a higher number of scientific profiles than most Western nations. While attention remains fixed on the most (over)valued companies, the real difference lies in the depth of the technical fabric.

      This also influences decision-making processes. When an organization is largely composed of individuals with scientific understanding, choices naturally rest more on technological realities than on politico-administrative considerations. This is the paradox of the rise of centralized yet effective countries, as was the case with several European countries like France under Charles De Gaulle’s presidency, during their post-war industrial boom.

      Iran has many structural weaknesses. Even before their influence grew during the war, the IRGC already wielded considerable economic power, monopolizing resources and, naturally, prioritizing military objectives—often at the expense of civilian development. However, the industry is well-suited to the demands of a war economy, where industrial planning, the decentralization of production capacities, and organizational resilience are top priorities.

      Modularity Redefines Industrial Sovereignty

      The organization of value chains has evolved. Systems now rely on increasingly modular architectures, composed of electronic subsystems, embedded software, sensors, batteries, or RF components that circulate within international supply chains. In many of these segments, the Chinese industry now occupies a central position, lowering entry barriers for numerous countries. Developing a system no longer requires mastering all the necessary industrial processes from the beginning. The advantage often lies in the ability to integrate components, organize production that evolves with needs, while pursuing a gradual substitution strategy for autonomy purposes.

      Iran has largely built its military-industrial apparatus on this logic. Chinese supply chains have given it access to abundant and inexpensive electronics, while the modularity of systems facilitates their continuous improvement. Sanctions have forced Iran to adapt—modifying architectures or developing certain components locally.

      This openness does not diminish the importance of industrial sovereignty. On the contrary, competition now focuses on mastering critical dependencies. The goal is to identify the links whose loss could disrupt production capacity and concentrate efforts on securing them. Industrial resilience depends less on absolute autonomy than on the ability to continue producing despite sanctions, logistical disruptions, or infrastructure destruction. Moreover, the concept of economic warfare, developed by the United States, is now being mirrored by Iran—with the Strait of Hormuz—and China—with rare earth elements.

      China’s industrial lead stems from the gradual integration of value chains. Batteries, electric vehicles, electronics, rare earths, and robotics illustrate this strategy. U.S. restrictions on semiconductors exemplify the dynamic: they temporarily slow certain Chinese developments but also create strong incentives to invest in domestic capacities, such as those around Huawei.

      The rise of Turkey’s defense industry is another example of this industrial development trend, particularly in drones—where it established an early foothold—but also in armored vehicles, weapons, and ammunition, even generating interest in its fighter jet program. While it had long been confined to a subcontractor role in the automotive sector, the country now develops its military exports to the EU, addressing significant needs after decades of underinvestment.

      The Erosion of Western Productive Culture and the Potential of Gen Z

      The United States retains exceptional assets, such as its university system and energy independence. Yet, a growing share of capital is directed toward activities whose valuation depends primarily on hyper-inflated financial markets. Passive investments, major stock indices, and abundant liquidity create self-sustaining mechanisms where the most highly valued companies attract new flows.

      The technological potential of AI is undeniable. However, a significant portion of investment remains focused on consumer applications, driven by advertising revenue or the growth of digital platforms. Applications capable of durably transforming the productive apparatus—such as robotics—still occupy a secondary role.

      These shortcomings also affect skill allocation. The United States continues to attract some of the best scientists and engineers. Yet, cronyism often prevails, as seen even in the civilian nuclear industry with the so-called “nuclear bros”—entrepreneurs whose knowledge of nuclear physics is often no deeper than that of the two lead diplomatic negotiators.

      The situation in Europe appears even more concerning. Despite the persistence of cutting-edge scientific sectors, strategic decisions are increasingly driven by administrative or political logic. The FCAS fighter jet program is a case in point, with its insurmountable flaws evident from its politically motivated launch. The “Chips Acts” overlook the value chains linking electronics, materials, industrial equipment, software, and manufacturing demand. Issues are addressed in isolation, without an integrated vision or sufficient scientific expertise. Analyses often revolve around calls for a great financial leap forward in European construction, as seen in the Draghi report, rather than technological foresight. Setbacks in key sectors, such as batteries—with the bankruptcy of Northvolt—now invite a more gradual approach to substitution.

      As shown by various emerging countries, recovery remains possible. Never before has a generation had such broad access to scientific and cultural knowledge, open communities, or educational and catch-up content. Despite the educational crisis, much of Generation Z is developing—often outside institutions—a technical culture that familiarizes them with the design of complex systems.

      An economy does not regain its industrial dynamism through an accumulation of short-sighted political spending initiatives, but by empowering those who understand technologies, master production chains, and can transform innovation into productive capacity. It is on this terrain—far more than on that of spending, including military spending—that the hierarchy and prosperity of global powers will be decided.

    • AI Financial Exuberance as a Shield Against Reality

      AI Financial Exuberance as a Shield Against Reality

      Op-Ed for Les Echos, Friday, June 5, 2026. The text below is a more detailed version of the article published in the print edition of the newspaper.

      The general rebound led by AI and semiconductor stocks, since the onset of de-escalation in the Middle East, is not enough to dispel concerns about the current investment cycle. On the contrary, this exuberance raises questions about its sustainability, against the backdrop of a race among LLM providers to go public.

      The already massive and growing weight of the sector in stock indices fuels a self-reinforcing dynamic, driven by passive investments: the more the sector grows, the more it attracts waves of capital, which in turn drive further growth—until a shock finally disrupts this mechanism.

      From this perspective, markets have quickly dismissed material risks—such as energy shortages or supply chain pressures for essential semiconductor inputs—as temporary. This reaction is not merely diplomatic optimism. For three years, the financing model has relied on the stratospheric expansion of LLMs, while downplaying questions about their business models, the consequences of pricing adjustments in the era of agentic AI, or the intrinsic limitations of these models in terms of reliability.

      The strong performance of cloud and semiconductor companies, combined with abundant liquidity and the dominance of a handful of firms, has reinforced the notion that demand for generative AI infrastructure will remain indefinitely robust.

      Circular Capital Flows and Technological Concentration

      This dynamic stems in part from the increasingly circular structure of financing. In 2026, projections for investments by Nvidia, Alphabet, Apple, Microsoft, and Amazon in « hyperscale » infrastructure range between $600 and $725 billion. The interconnections within this ecosystem are particularly tight. Nvidia occupies a central position as both the dominant supplier of GPUs and a key investor, reinforcing a loop in which investments, demand for computing capacity, and production capabilities are mutually dependent.

      Microsoft has invested $13 billion in OpenAI, whose computing costs rely heavily on Azure—further boosting Microsoft’s cloud revenue and its ability to sustain its investments. Google has poured several billion dollars into Anthropic, which has simultaneously committed to massive cloud infrastructure contracts, split between Google and Amazon. Amazon itself has invested around $8 billion in Anthropic, which then developed subsidized access for developers through the cloud infrastructure of these same companies.

      However, usage-based pricing models (per token) introduce uncertainty about whether this growth can translate into sustainable revenue, particularly if mechanisms of « subsidization » or indirect support were to wane. These mechanisms sustain growth expectations but blur the line between independent demand and self-perpetuating capital flows. A significant portion of the sector’s apparent strength rests on a small number of companies that simultaneously finance the infrastructure, provide the computing power, and support the applications consuming that power.

      The rise of passive investing further amplifies this phenomenon. As major AI-related companies see their valuations climb, their weight in major indices automatically increases, attracting more financial inflows and intensifying market concentration. The « Magnificent Seven » now account for between 30% and 45% of the S&P 500’s market capitalization, depending on the period, while Nvidia’s market cap has surpassed $5 trillion.

      Material Constraints and Financial Fragility

      At the same time, the material foundations of this expansion are becoming increasingly critical. Cutting-edge AI depends on massive growth in electricity consumption, semiconductor manufacturing capacity, cooling systems, and data centers. Semiconductor production itself relies on complex industrial supply chains involving LNG, helium, specialty gases, copper, and stable electrical power.

      Helium exemplifies this dependency. Qatar is one of the world’s leading exporters, and any disruption to maritime routes in the Gulf could quickly impact semiconductor manufacturers in East Asia. In Taiwan, several industrial groups have already expressed concerns about the security of LNG and helium supplies.

      Moreover, the island sits at the heart of the U.S.-China diplomatic chessboard, with Trump’s approach amounting to a refusal to engage on the issue. Meanwhile, China has embraced the U.S. strategy of restricting semiconductor exports and is betting on building its own autonomy—centered around Huawei—which, in the long run, could challenge the dominance of American giants and their financial constructs.

      Additionally, the rapid obsolescence of infrastructure adds another layer of fragility. Data centers built around current GPUs could lose a significant portion of their competitiveness for advanced computing workloads in as little as 18 to 36 months, even if they remain usable for inference or secondary applications. Yet, accounting depreciation periods typically span three to five years, potentially obscuring underutilized infrastructure and delaying visibility into long-term financial obligations.

      This is not about questioning the AI revolution itself, but how financial markets treat LLMs’ growth as limitless—underestimating physical, financial, industrial, and geopolitical constraints… as well as the opportunities of alternative models. Large language models fit naturally into the current financial architecture because they deploy efficiently through cloud infrastructure and (partial) subscription-based business models. In contrast, physical AI—particularly in robotics—operates under a different logic. It depends more on real-world deployment and longer development cycles, which align less neatly with current financing mechanisms.

      This dynamic echoes Minsky’s financial instability hypothesis, which posits that long periods of stability gradually encourage increasing risk-taking. The limits of financial and industrial resilience may soon force a rude awakening, perhaps triggered by profit-taking after a wave of IPOs.

    • France Might Become Europe’s Data Center Hub, but Where Does It Stand in the AI Race?

      France Might Become Europe’s Data Center Hub, but Where Does It Stand in the AI Race?

      Interview with Atlantico on France’s AI Infrastructure investments, following the 2026 Choose France Summit announcements (Excerpts).

      Are the Choose France announcements a sign that France is winning the AI race—or just the data center race? Behind the €93 billion figure, how much actually goes toward developing AI technologies, models, and intellectual property compared to infrastructure?

      These investments do not mark a decisive victory for France in the AI race, but they do position the country at the heart of Europe’s AI infrastructure. While they bring industrial benefits, the creation of intellectual property largely remains in the hands of international players. The challenge now is to leverage this attractiveness to develop a national AI industry.

      Amid an energy crisis, France—with its largely decarbonized electricity and stable grid—has become a European hub for AI infrastructure, drawing in players like SoftBank, Brookfield, and Ardian. These firms are investing in data centers and sparking industrial partnerships. Schneider Electric, for instance, is mobilizing its expertise in energy efficiency, cooling, and automation. These projects help develop high-performance data center management skills.

      However, the development of key technologies and models remains limited. Despite initiatives like the Bull/Foxconn project on motherboards, the focus is more on infrastructure than R&D labs or GPU production—the latter being where much of the sector’s real value lies. Europe still lags far behind in semiconductors, despite some promising efforts.

      We must also consider the financial exuberance surrounding AI, particularly in infrastructure financing. A national strategy cannot be built on speculative promises alone. As seen in global initiatives, France should develop local funding sources and protect the integrity of its tech companies, both in terms of intellectual property and capital resilience.

      We often distinguish between inference data centers (which execute queries) and training data centers (where large AI models are developed). Is France hosting the most strategic parts of the value chain, or mainly data centers that benefit from our energy advantage?

      Overall, the projects cover both types, though the distinction isn’t always clearly defined. The Ardian/Verne « AI Gigafactory »—combining high-performance computing and research activities—appears to be the most training-focused. Training centers are more strategically valuable from a geopolitical and industrial standpoint, as they require massive resources (energy, cooling, GPUs) and are difficult to relocate.

      Inference data centers, on the other hand, are less strategic since they rely on pre-trained models and optimized chips. Yet they complement France’s offering by enabling large-scale AI service deployment with reduced latency for European users. Their value lies in proximity to end markets.

      The key challenge is avoiding the role of a mere host. France must capitalize on these infrastructures to develop technological partnerships, attract R&D centers (by conditioning support on technology transfer commitments), build ties with local industry (for sector-specific models), and ensure it doesn’t remain just a provider of electricity and land.

      Arthur Mensch, Mistral AI’s CEO, told the National Assembly that AI is first and foremost a heavy energy industry. Does France truly understand that the AI battle won’t be won on talent or software alone, but on the ability to produce and deliver electricity? Can France meet the electrification challenge posed by AI’s demands?

      Mensch rightly reminded policymakers that AI isn’t just about models—it’s about physical and energy infrastructure. Behind every model lie data centers, semiconductors, cooling systems, and power grids. Beyond leveraging France’s tradition of mathematical creativity and its versatile engineers and researchers, the country must also exploit its abundant, stable, and competitive electricity supply. Here, nuclear power remains a key advantage, even if the sector has been weakened by strategic indecision and hindered by a flawed European pricing framework.

      Why does France face a two-year window of opportunity in AI? Are we witnessing an industrial revolution where today’s decisions will shape global power dynamics for decades?

      We’re entering a phase of consolidation. The early years of generative AI were experimental—models were developed, pricing was fluid. Now, the players controlling compute, data, talent, and energy are locking in their positions. The parallel with industrial revolutions is clear: those who dominate foundational infrastructure set the technological, financial, and geopolitical standards that follow.

      Yet we must resist the dominant narrative. We’re in an era of excessive valuations, with circular financing mechanisms between semiconductor companies, hyperscalers, and model providers. Many use cases remain unproven, while markets anticipate massive future revenues—even as some models continue to operate at a loss.

      Europe shouldn’t blindly copy the U.S. hypergrowth model, fueled by deep capital markets and a high tolerance for deficits. We lack the financial firepower and the same risk appetite. Instead, we must pursue more selective, industrial, and efficient pathways.

      Open source is a strategic lever: it enables cost-sharing, broader access, reduced dependence on American platforms, and the development of specialized models without requiring tens of billions in capital. As Yann LeCun has noted, much of Meta’s early Llama development happened in Paris. The real challenge is turning conceptual strength into industrial power.

      According to estimates, AI could require up to 40 additional gigawatts of power in France. Should nuclear be seen as the absolute condition for digital sovereignty, or is a more pragmatic mix—nuclear, solar, renewables, and grids—now unavoidable?

      Nuclear is essential if France wants to maintain a controllable, decarbonized, and competitive electricity supply at scale. Industrialized AI cannot rely excessively on intermittent energy sources. But the challenge extends beyond nuclear: it’s about the entire energy system—grids, storage, hydropower, energy efficiency, and cooling capacity.

      Meanwhile, China is taking a more pragmatic approach: lower-cost infrastructure, more compute-efficient models, and aggressive hardware optimization. Beijing is also working to replicate Nvidia’s capabilities in the face of U.S. export restrictions.

      The global AI race is now moving at a pace incompatible with France’s bureaucratic inertia—not just in energy, but across the board. We need to recreate industrial and technological free zones: streamlined regulations and tailored tax incentives for innovation and critical infrastructure. The French paradox is that we once had one of the world’s most competitive energy and scientific systems, only to then systematically deindustrialize ourselves.

      Is France missing the AI value chain upgrade, left providing only energy, infrastructure, and expatriated talent while the U.S. captures the real value?

      The risk is real: France could end up confined to the lower rungs of the value chain—supplying power, hosting data centers, and exporting talent—while the U.S. monopolizes the high-value segments. But to reposition ourselves, we must first understand the sector’s current state, with its flaws and emerging opportunities.

      Beyond the inherent limitations of LLMs, much of the AI sector today is driven by highly speculative financial expectations. Many use cases remain difficult to monetize, even as compute and capital demands skyrocket in the age of agentic AI. The next wave may well come from AI deeply integrated into real industrial systems: robotics, automation, maintenance, defense, logistics, industrial simulation, and healthcare. The goal isn’t just to imitate OpenAI but to drive productivity gains through integration with physical production chains.

      With OpenAI’s rumored IPO at $850 billion, Anthropic at $900 billion, and SpaceX at $2 trillion, American giants will have the capital to lock in compute capacity and energy resources at a scale Europe can’t match. Is there still a realistic path for France and Europe to close the AI gap with the U.S. and China in the next two years? Could robotics be part of the solution?

      These valuations underscore America’s financial dominance. These companies can raise sums that secure semiconductors, data centers, and energy contracts on a scale beyond Europe’s reach.

      Europe has also trapped itself in a regulatory labyrinth, particularly with the AI Act. Multiple European states and companies have pleaded for adjustments to preserve industrial competitiveness—only to see their concerns overlooked. Meanwhile, Donald Trump merely had to demand that Ursula von der Leyen fall in line with U.S. interests—and she complied.

      France and Europe can still build strong positions in areas where we have industrial, scientific, or even energy advantages. Robotics is a prime example: it combines software, sensors, mechanics, power electronics, and industry—even if we don’t cover every link in the chain.

      Physical AI offers an alternative to consumer-focused, chatbot-driven applications. Advanced industrial robotics delivers direct gains in competitiveness, productivity, and reindustrialization. This is likely where we have the best chance to create synergies with our industrial base.

      Read the full interview on Atlantico.

    • Reopening the Strait of Hormuz: Toward a Frozen Conflict amid US Political Paralysis

      Reopening the Strait of Hormuz: Toward a Frozen Conflict amid US Political Paralysis

      English transcript of my interview on France Info TV – 29 May 2026.

      Hello, Rémi Bourgeot. At the heart of this issue, as always, is money—the economy. In Donald Trump’s decisions, just minutes ago, we were told the U.S. president wanted to take his time before possibly announcing his decision on the Iran deal. And now, we see things accelerating. Do you see an economic dimension here?

      Rémi Bourgeot: Absolutely, this is the primary source of pressure on Donald Trump, stemming from the global economy and the U.S. economy. So he’s under this urgency.

      But today, he’s also facing internal divisions. Personally, he quickly realized the disaster that the U.S.-Israeli campaign against Iran had become, especially since he’d been promised it would only last a few days.

      On the American side, he’s dealing with tensions from parts of the Republican Party, particularly the neoconservatives, who oppose any diplomatically realistic agreement. And then there’s the pressure from Israel, which—at least from Benjamin Netanyahu’s perspective—has no interest in reaching a deal with Iran, let alone one with regional repercussions, especially regarding the war against Lebanon. So the situation remains highly uncertain.

      For the Americans, the core issue is still nuclear. Yet we also see the weight of Iran’s position and the immense military challenges—on top of the economic difficulties the U.S. faces—having to concede to Iran’s main demand: delaying a nuclear agreement. Let’s not forget that under Obama, such negotiations took years. Today, the discussion is about Iran committing not to develop nuclear weapons—but that’s already the case. The negotiations focus on civilian nuclear programs and enrichment levels to prevent Iran from being in a technical position to develop military nuclear capabilities.

      We’re seeing a situation of extreme difficulty for the United States. Trump has recognized the deadlock he’s in, but he’s facing massive opposition and pressure, with all this back-and-forth and incredible uncertainty. And the negotiation process itself is quite surprising: he’s being asked to approve something that, in theory, he himself is supposed to negotiate—because the negotiators have a very limited mandate.

      So the parties are indeed trying to agree on reopening the Strait of Hormuz, which was fully open before this conflict. But beyond this partial agreement, tensions remain extremely high with the U.S., which is struggling to acknowledge this strategic defeat.

      Rémi Bourgeot, you’re still with us. Are we talking about war, potential solutions to the conflict, or not? And are we mainly discussing nuclear and oil issues?

      Yes, I think it’s fair to say that even if an agreement is reached, it would be a very partial one. In reality, the situation would look much more like a frozen conflict than a peace deal. The goal is to reopen the Strait of Hormuz and, for now, set aside the nuclear issue for a potential future agreement—because such negotiations take a very long time.

      What’s happening right now is that Donald Trump feels the need to include the word “nuclear” in his announcement. But in reality, this isn’t a genuine concession, since Iran isn’t developing—and did not intend to develop—a military nuclear program. That wasn’t the issue.

      As for enrichment, even for civilian or experimental purposes, concessions were already on the table. For Iran, the key is maintaining the ability to enrich uranium. So we’re seeing the situation unfold as it has over the past three months.

      This is a strategic debacle for the United States and for Israel as well. Trump is facing immense difficulty in reaching an agreement, even though he’s been trying for two months to extricate himself from this situation.

      The most significant concession from Iran seems to be over control of the Strait of Hormuz. Even if they adjust access or transit conditions, or reopen the strait in coordination with the Americans, they’ve demonstrated what some have called their own “nuclear weapon”: the ability to control the strait, albeit with very limited means compared to, say, their ballistic missile program.

      So we’re looking at a situation that leans toward a frozen conflict, with the prospect of a nuclear deal coming later. The positive side is that there’s a shared will to reopen the Strait of Hormuz and make mutual concessions. But there’s still a major roadblock on the U.S. side, which Trump must overcome to realistically achieve any kind of breakthrough.

      Thank you, Rémi Bourgeot, for your analysis.

      This automatic transcript has been edited for the sake of clarity.

    • AI Financial Mechanics under Real-World Constraints

      AI Financial Mechanics under Real-World Constraints

      This piece is based on my research, The Global AI Race amid Asset Bubble Dynamics, which I will present on Friday, 15 May 2026, at King’s College London. .

      Deescalation in the Iran war has unleashed a market rebound led by AI and semiconductor stocks. Its strength in the face of major challenges actually raises questions about the sustainability of the broader boom. Physical disruptions persist, from energy shortages to critical materials, and the de-escalation remains fragile, with Donald Trump unwilling to resume hostilities yet unable to consolidate a lasting and realistic agreement. Long periods of strong performance, abundant liquidity and dominance by a few large firms have trained markets to treat disruptions as temporary and manageable.

      Despite AI’s many promises, circular funding and loss-making business models underpin the financial dynamic. The rise of passive investing supports this mechanism. Large tech firms receive a major share of these investments and in turn keep fueling circular financing within the sector, helping maintain their own earnings. This echoes various theoretical frameworks on asset bubbles.

      The global bet on the scalability of LLMs to reach human-level intelligence fits perfectly within this financial logic. Meanwhile, essential developments in physical AI, particularly for robotics, suggest a different technological path and financial structure.

      The Financial Mechanic and Its Technological Impact

      The rally in AI-related stocks reflects a market downplaying industrial effects from disruptions in the Gulf. Markets quickly returned to the dominant narrative of rising AI investment, continued infrastructure expansion and strong demand for advanced semiconductors. Nvidia, its suppliers, memory manufacturers and cloud infrastructure firms led the rebound. So far, there have been no major visible disruptions to semiconductor production, reinforcing investor positioning in the sector despite geopolitical uncertainty. Investors largely treated the situation as a temporary geopolitical disturbance rather than systemic stress.

      Semiconductor stocks are led by AI infrastructure growth expectations. Strong earnings guidance reinforces the belief that compute demand remains effectively unconstrained. As long as digital giants maintain investment plans and financing conditions remain supportive, markets tend to interpret geopolitical shocks as disturbance rather than turning points. The current rally reflects confidence in AI-driven earnings growth, continued hyperscaler investment and passive flows reinforcing index concentration in a small group of large technology firms. The so-called Magnificent Seven now represent 30–45% of S&P 500 capitalization and Nvidia’s market cap has exceeded $5 trillion.

      Part of this strength reflects the increasingly interconnected structure of AI-related investment. Amazon, Microsoft, Alphabet and Meta are projected to spend well over $600 billion on AI infrastructure in 2026, with some estimates approaching or exceeding $700 billion. Microsoft has invested approximately $13 billion in OpenAI, whose workloads reinforce Azure demand and revenue. Alphabet has invested heavily in Anthropic, with reported financing commitments potentially reaching $40 billion. Anthropic has reportedly agreed to spend roughly $200 billion on Google Cloud infrastructure over five years. Amazon has also expanded its investment in Anthropic beyond the original $8 billion commitment, alongside reported long-term AWS procurement agreements exceeding $100 billion over a decade.

      The prospect of usage-based pricing models (per token) introduce uncertainty over whether the observed expansion in usage will translate into durable willingness to pay once investors « subsidies », credits, or strategic cross-subsidisation are reduced. This creates a gap between measured demand growth and underlying monetisation capacity.

      Obsolescence add to the risk. Semiconductor progress is rapid enough that data centers built on current GPUs may become less efficient within 1.5–3 years, while remaining useful for inference and secondary applications. Depreciation schedules of 3–5 years stretch asset lifetimes beyond their peak economic usefulness, potentially masking underutilized infrastructure. Off-balance-sheet leasing structures can further delay visibility of long-term obligations.

      Furthermore capital concentration shapes technological choice. Large language models fit the current financial structure well. They scale on GPU infrastructure and align with subscription-based revenue models. Physical AI, particularly robotics, operates differently. It depends on physical systems, long development cycles and data that cannot be easily replicated in cloud environments.

      Theoretical frameworks explain these dynamics. The notion of reflexivity highlights how rising valuations attract further capital. Minsky’s instability hypothesis describes how stability encourages risk-taking. Shiller’s irrational exuberance emphasizes the role of narratives in driving capital flows. The unabated rise of passive investing reinforces these effects through index-linked momentum.

      The US–China chip competition adds complexity. US export controls on H100 and H200 GPUs force China to rely on domestic alternatives, though performance gaps vary by workload as local accelerators improve in certain use cases. State-backed funding reduces political constraints but does not remove inefficiencies. Chinese firms such as Alibaba and Tencent have significantly increased AI and cloud-related spending, while hardware constraints force higher spending for equivalent compute output. The result is a more fragmented semiconductor ecosystem, with duplicated investment.

      Physical Supply Chains and Semiconductor Production

      Financial markets and physical supply systems are not fully aligned. The Strait of Hormuz remains a critical energy and logistics corridor. It also affects flows of industrial inputs that are difficult to substitute quickly. Semiconductor production depends on LNG, helium, specialty gases, copper, cooling systems and stable electricity supply. Several of these inputs remain exposed to maritime risk in the Gulf region, though current data suggest limited constraint at this point.

      Helium illustrates this dependence clearly. Qatar is one of the world’s largest helium exporters and any disruption to transport routes can quickly affect semiconductor manufacturers in East Asia. Industry groups in Taiwan have raised concerns about LNG and helium security. The issue is the gradual reduction of redundancy in already tight supply systems.

      Market pricing assumes that conditions will normalize before supply constraints affect production or expansion. This assumption may prove correct. The semiconductor industry has repeatedly adapted to logistics shocks through inventory management, supplier diversification and government coordination. Markets are not ignoring geopolitical risk entirely. Oil prices, shipping costs and insurance premiums have all reacted. But equity investors—especially in AI-linked sectors—are still assigning relatively low probability to prolonged industrial disruption compared with current earnings momentum and investment trends.

      Physical Constraints, Valuation and Fragility

      The AI investment cycle has created a tight link between financial expectations and physical infrastructure. Unlike earlier software cycles, frontier AI requires continuous expansion in electricity use, data center capacity, semiconductor fabrication and cooling systems. This exposes the sector to physical constraints as well as financial conditions. Transformer shortages, copper constraints and grid delays are already emerging in several regions. Hyperscale expansion increasingly competes with broader industrial and public electricity demand.

      These constraints are only partially reflected in market pricing, which remains focused on growth narratives and future monetization potential. Index composition also plays a structural role. Semiconductor and hyperscaler firms now account for a large share of global equity indices. As their valuations rise, passive investment flows reinforce their dominance, strengthening momentum regardless of changes in underlying risk conditions. Capital concentration therefore becomes self-reinforcing during periods of uncertainty.

      Financial resilience and industrial resilience are no longer moving in lockstep. This does not imply current valuations are irrational or that a correction is imminent. The AI buildout is producing real revenue growth, infrastructure expansion and rising demand for compute. The issue is that markets increasingly extrapolate this trajectory as sustainable or unlimited, while assuming geopolitical and industrial disruptions remain temporary.

      Historically, strong technological cycles often produce similar assumptions. During periods of rapid capital concentration, markets tend to prioritize scale and dominance Fragility tends to appear only when constraints persist longer than expected or when financing tightens alongside operational stress. The current environment contains elements of both outcomes. Semiconductor demand remains strong, but its supporting infrastructure is increasingly exposed to geopolitical concentration. Taiwan remains central to advanced manufacturing. Gulf shipping routes remain important for energy and industrial inputs. Electricity systems are under pressure from AI-driven demand growth. Much of this expansion still depends on sustained capital expenditure and favorable financial conditions.

      The recent market rally may reflect less a resolution of geopolitical risk than the growing dominance of the AI investment framework in global markets. Investors still appear to assume that the importance of AI infrastructure will continue to justify large-scale capital deployment despite visible physical constraints and the lack of a viable business model.