Category: AI

  • 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.

  • War Deepens the Industrial and Social Crisis

    War Deepens the Industrial and Social Crisis

    Interview on France 24 in French with journalist and novelist Aude Lechrist and in English with William Hildebrandt on how the Middle East war derails the West’s economic, industrial and social model further. Translation of the French interview below the English video.

    Aude Lechrist: In France, as elsewhere in the world, the closure of the Strait of Hormuz is making itself felt. Trade unions are pushing for wage increases, particularly as inflation makes a comeback. To help us understand how workers are being affected by today’s upheavals — the geopolitical situation, climate pressures, and the dizzying pace of advances in artificial intelligence — we are joined by Rémi Bourgeot. Thank you for being here. First, are workers facing the same pressures the world over?

    Rémi Bourgeot: Extreme globalization has taken hold, creating significant transmission belts running through industrial models — but alongside that, vastly different policies have been pursued on either side of the divide. The fast-developing countries of Asia have pushed industrial policies, import substitution strategies, and drives toward productive self-sufficiency. The West, by contrast, has undergone rapid deindustrialization over the past few decades.

    And yet Asia is heavily affected today on the energy front, even though China in particular had put anticipatory policies in place. This crisis feeds directly through to workers, to job opportunities, and to cost pressures stemming from globalized supply chains — though that globalization is now somewhat in retreat, as countries seek greater autonomy and resilience.

    So workers are immediately more exposed — that much is clear from the geopolitical context. What knock-on effects are you observing?

    The economic consequences are immediate and concrete — the energy crisis, for instance, has brought production lines to a standstill. Economists point to fractions of a percentage point being shaved off overall GDP, but the real issue is a crisis of the real economy, the physical economy, of supply chains. For many countries around the world, that is precisely what drives economic and industrial development.

    And just about everywhere, questions of industrial development, genuine development, educational development are back at the centre of the debate — because these are the factors that determine long-term growth prospects and the opportunities open to workers.

    A major fault line has opened up between countries that believed growth could rest indefinitely on services — particularly financial services — and others that have followed a more traditional development path, reminiscent of postwar Europe: industrial development, educational development — which opens up more opportunities for workers, even if working conditions are sometimes very tough.

    But right now, an inflection point has clearly been reached: developed countries no longer have a functioning growth model.

    Artificial intelligence, which you mentioned at the outset, is also reshuffling the deck — particularly through its applications in robotics, which will increasingly affect manual workers, in addition to office jobs. And again, that fault line is visible, with the development model unraveling across much of the Western world.

    The United States has managed to stay ahead on the digital front and now in AI. How do you read that, especially against the backdrop of the Strait of Hormuz crisis — given that the key investors are the Gulf states?

    The development model has genuinely unraveled right across the Western world. The United States holds the high ground technologically, but on the premise that it can keep pushing indefinitely down a path heavily dependent on financial flows and foreign capital — particularly from the Gulf.

    The announcements from Sam Altman and OpenAI have been staggering — trillions supposedly raised in the Gulf to fund data center infrastructure in the United States and beyond. And the financial structures taking shape among players in this sector have all the hallmarks of a bubble — customers being financed by their own suppliers like Nvidia, investments completely untethered from economic and industrial reality.

    And yet genuine innovations do exist, and there is extraordinary talent out there, even from a purely technological standpoint. AI researchers like Yann LeCun argue that generative AI and LLMs are running into a dead end because of their intrinsic errors — something anyone who uses these tools day to day can see for themselves. Other technologies need to be developed, and that is already happening, particularly for robotic applications in the real world.

    But the moment an innovation appears, vast financial edifices get built up around it that have little to do with actual economic, industrial, or human development.

    And then there is the fear among workers — Americans in particular — who see an economic crisis on the horizon. Trade unions are clearly gearing up for major action. Labor Day in the United States is separate from International Workers’ Day, but significant mobilization is expected today all the same. Donald Trump has clearly done very little to address the concerns of American workers.

    Yes, and that is the great paradox. All the wavering, the U-turns, the chaos surrounding Donald Trump tell the story — he was supposed to upend the system in favor of reindustrialization from his very first term. Efforts were made in that direction, but the personal competence simply was not there, nor were the right people around him, to deliver a genuine industrial policy — not even on the tariff front, when it came to applying duties where they were actually needed, where domestic production could realistically be substituted or rebuilt at an acceptable cost.

    And yet that question did get put on the table — one that recurs throughout American economic history, as it does in the history of any country pursuing industrial development.

    When the Democrats returned to power, they largely continued in the same vein of industrial realism, of attempted reindustrialization — more through subsidies than tariffs, but still within a broadly protectionist logic.

    And now, with this new Trump term, the result is a bizarre and catastrophic world of blunt-instrument measures that get walked back almost immediately, with no strategic underpinning and utterly chaotic trade negotiations. The negotiators — on trade, but also on geopolitical, diplomatic, even military matters — have no idea what they are doing. Some of them can barely find the countries in question on a map.

    The chaos that has ensued points to a very deep systemic crisis — a crisis of American society and of Western society more broadly — an inability to bring about political renewal, or even basic reform, that would reconcile human and industrial development with the realities of globalization. That can only deepen the anxiety of workers who already see a vast gulf between the uncertainty generated by outside forces — conflicts, tensions, climate risks, artificial intelligence — and their governments’ capacity to respond, compounded by the interdependence between all these countries.

    That is genuinely alarming — because beyond all the political divides, the different countries and currents of opinion, there actually is a broad shared diagnosis: reindustrialization is needed. And yet nothing happens. Promises are made and forgotten.

    You said as much about the United States, but Europe is no different — if anything it is worse, having missed every significant technology wave over the past twenty or thirty years. The engineering expertise is still there for now, but it is eroding. And has the appetite for innovation gone with it? Is it no longer what drives students who dream of building a better world? Do you share that concern?

    What keeps me from losing hope entirely is that talking to young people — students in engineering schools, in other fields, in the humanities — one still finds that curiosity, that intellectual energy. Despite the decline of the education system, a wealth of tools exists online, countless ways of accessing knowledge — with their strengths and their limitations — that still allow people to learn, to catch up, to make discoveries. The curiosity is very much alive.

    The problem lies in the economic, political, and industrial system as it stands, which crushes that creativity. Entrepreneurship is a case in point — starting a business is an uphill struggle in Europe and in France especially. And at the level of larger companies and public bodies, reindustrialization is talked about endlessly but always in the vaguest of terms.

    Looking back, what has been the real impact of Emmanuel Macron’s two terms on workers in France?

    There has been a genuine slippage. A commitment to entrepreneurship was at least proclaimed, but it was mostly rhetorical from the outset. The occasional junior minister had a genuine grounding in the real economy, but overall, a headlong rush toward deindustrialization has unfolded, dressed up in rhetoric pointing in the opposite direction — toward rebuilding France’s industrial fabric. The means simply have not been there: the human resources, the investment decisions at the national level, the European coordination.

    Then there is the energy pricing system, which is extremely damaging for the French economy. France should enjoy a competitive advantage thanks to nuclear power, but that advantage is largely neutralized by the European pricing mechanism — a trap the country remains locked in.

    On top of that, the strategy of kicking the can down the road goes back to the introduction of the euro. The trade balance has been deteriorating and in the red since the start of the eurozone. This ongoing decline has been masked by the illusion of monetary stability — but with debt soaring and interest rates rising, that cannot go on indefinitely.

    What is really lacking is a technological, industrial, economic, and human understanding — including in terms of skills — to get an industrial development agenda back on track. That is exactly what other countries are doing, not that their models should be copied wholesale — China being the obvious example. A genuine boom in industrial development and technological expertise is underway in China today, comparable to Japan’s spectacular catch-up across every technological front forty or fifty years ago.

    France has extremely strong expertise — pockets of world-class engineers, outstanding skills, including in mathematics — and none of it is being properly put to use.

    Rémi Bourgeot, thank you very much for joining us — a fascinating conversation. Thank you.

  • Beyond the Iran Fiasco, an Abysmal Strategic Vacuum

    Beyond the Iran Fiasco, an Abysmal Strategic Vacuum

    Op-ed published by Les Echos on 24 March 2026. As Donald Trump seeks a way out of the Iranian quagmire—to suspend hostilities without any real prospect of peace—I invite you to consider a broader reflection on the strategic void that accompanies this situation:

    The war with Iran reveals a structural failure within the American decision-making apparatus, marked by a difficulty in aligning immediate tactical actions with long-term political objectives. This misalignment extends beyond the military sphere. Defense, trade, finance, and technology policies interact in a chaotic manner. The conduct of the trade war has already illustrated this: the legitimate goal of reindustrialization has been overshadowed by geopolitical coercion.

    In this very real war, the inability to anticipate the consequences of a failed regime change or the closure of the Strait of Hormuz further demonstrates a loss of overall vision. The military instrument is wielded without a political framework capable of setting a clear direction.

    These strategic and material flaws, already evident in the war of attrition in Ukraine, suggest a system grappling with internal contradictions and a reliance on a form of virtual thinking. The strategic framework appears frozen in patterns inherited from the era of the Iraqi adventure, even as industrial and geopolitical realities have radically transformed, reshaping the balance of power. This latest crisis calls on Europe to undertake a difficult reorientation.

    Retreat of Monetary Hegemony

    Economic sanctions have become a central tool of diplomacy. Yet their use generates side effects that are beginning to reshape the global financial architecture. Initially designed to isolate specific actors without direct military engagement, these measures have accelerated the search for alternatives. Beyond the surge in gold, we are witnessing a proliferation of bilateral agreements in local currencies and the development of parallel clearing systems, which are undermining one of the pillars of American power.

    The Iranian conflict acts as a catalyst here. The paralysis of the Strait of Hormuz underscores how power depends not only on dematerialized flows but even more on complex material systems: energy and industrial infrastructures. The West finds itself in a position where its instruments of pressure are losing effectiveness as regional powers adapt, coordinate outside traditional frameworks, and are prepared to escalate.

    Industrial Wars of Attrition

    Above all, the evolution of recent operational theaters, particularly in Ukraine, has forced a belated rediscovery of the importance of the industrial base. Technological superiority and the development of financial markets may have created the illusion that mass production capacity was secondary. The reality of a war of attrition has shown that economies with much more modest GDPs, but equipped with resilient production systems supported by China, can stand up to technological powers whose production chains are fragmented or optimized for peacetime.

    This situation reveals a divide between nominal wealth, driven by services and intangible assets, and the actual ability to mobilize material resources in prolonged crises. Tensions over ammunition stocks and delays in reactivating defense industries illustrate this lack of industrial depth. Although deindustrialization is recognized as a risk to social cohesion and strategic autonomy, the response has remained superficial. Tariff policies are often employed erratically, serving more as diplomatic tools than as genuine levers for rebuilding an integrated productive fabric.

    Misalignment of Capital and Educational Erosion

    Meanwhile, financial markets continue to channel capital toward high-visibility sectors, to the detriment of fundamental infrastructure. The AI bubble absorbs a disproportionate share of investments, while heavy industry and industrial transformation struggle to attract the necessary funding. This imbalance creates a two-speed economy, where digital innovation advances without an industrial infrastructure capable of withstanding geopolitical shocks.

    This crisis of strategic thinking is rooted in the weakening of educational structures, particularly in scientific culture and the classical humanities. The decline in science education reduces the ability to grasp the physical and technical constraints of the real world, fostering a virtual vision where it is believed that large language models can replace versatile engineers. At the same time, the retreat of the humanities deprives decision-makers of the historical intuition needed to understand the long term.

    Europe particularly embodies this tension. The continent’s industrial catch-up is hampered by regulatory complexity, compounded by a shift in decision-making power from the technical to the administrative, reducing the capacity for long-term planning. The management of contemporary crises highlights the need for a transition toward a systemic approach, integrating energy security, industrial resilience, monetary stability, and technological innovation within a strategic framework. This transformation cannot occur without questioning decision-making and educational mechanisms. The reallocation of resources must be accompanied by a renewed emphasis on fundamental knowledge, capable of restoring a long-term vision.

    This piece was originally published on Les Echos website in French.

  • After the Bubble: AI Can Serve Industrial Power Instead of Draining It

    After the Bubble: AI Can Serve Industrial Power Instead of Draining It

    This op-ed has originally been published by Les Echos(fr).

    The generative AI bubble is built on circular funding between sector players, valuations disconnected from economic realities, and an extreme concentration of resources on large language models (LLMs). What should be alarming is not so much the scale of these investments as their stark contrast with the disintegration of Western industrial capacities. The war in Ukraine exposed this structural flaw, revealing the inability to produce sufficient quantities of essential military equipment—the result of decades of deindustrialization and skewed capital allocation. Beyond its strategic dimension, this paradox calls into question how we measure economic power.

    On the AI front itself, the success of more frugal players like Mistral or DeepSeek demonstrates that innovation does not depend solely on a relentless race to build ever-larger models. Billions continue to pour into colossal physical infrastructures—energy-hungry data centers, specialized chips, computing networks—without questioning the fundamental limits of LLMs. These massive investments stand in sharp contrast to the chronic underfunding of industry, and paradoxically, of automation.

    Beyond the fantasy of a dematerialized digital world, data centers are infrastructures that consume vast material resources: energy, rare metals, electronic components. Their proliferation highlights the current paradox: we are exponentially increasing computing power, while the productive sectors that could benefit from these technologies lack funding and orders. Many of these sectors launch AI projects merely to tick a box and make announcements to attract investors. In the military domain, autonomous drones, intelligent combat systems, and predictive maintenance represent concrete applications where AI will make a difference—but only if integrated into a solid industrial base, rather than betting everything on unreliable models.

    The production chains for ammunition, armored vehicles, and electronic components, weakened by years of underinvestment, struggle to meet demand. Factories have closed, skills have dwindled, and revival attempts are hampered by the absence of long-term strategic planning. The United States, despite its own contradictions, is trying to correct this imbalance by relocating some strategic production. Europe, however, remains on the sidelines, locked in extreme technological dependence that undermines its sovereignty.

    The core issue lies in this skewed allocation of resources. Capital and talent are concentrated on speculative technologies, while industrial applications of AI—advanced robotics, autonomous systems, production process optimization—remain underfunded. Above all, they lack commercial guarantees in the form of orders. This creates a vicious cycle: the more investments flow into LLMs and their infrastructure, the fewer resources remain to modernize the real productive apparatus.

    Yet AI could be a major lever for reindustrialization if approached differently. A more balanced strategy would involve redirecting some investments toward industrial automation, developing practical applications embedded in production processes, and fostering hybrid skills that combine digital expertise with industrial know-how, rather than chasing publicity stunts.

    Without this strategic shift, the gap will widen between an oversized digital sector and an industrial base unable to meet material challenges. The war in Ukraine served as a wake-up call. Power is not measured solely by the ability to develop sophisticated algorithms but also by the capacity to produce essential equipment. The challenge is not to reject AI but to reintegrate it into an industrial logic, where digital innovation finally serves material production rather than replacing it. Without this rebalancing, the West risks ending up with an economy where computing power soars, but factories continue to close.

  • Behind DeepSeek: France’s Path to AI Excellence

    Behind DeepSeek: France’s Path to AI Excellence

    By leveraging its mathematical expertise and open-source innovation, Europe can compete with the United States and China—not just through massive investments, but above all by keeping scientific culture at the heart of its strategic vision.

    As China’s DeepSeek reshuffles the global AI competition, France is also seeking to highlight its cutting-edge capabilities, announcing major investment projects in digital infrastructure at the Paris Global Summit. The rapid success of Mistral AI has demonstrated France’s potential, with its researchers and engineers defying the educational crisis through their mathematical talent. Yet a gap persists between this scientific excellence and public action, as seen in recent missteps—most notably the premature launch of the open-source AI model Lucie. The state must redeploy its scientific expertise to ensure strategic cohesion in these investments and prevent Europe’s digital ecosystem from being systematically overshadowed by Silicon Valley.

    This moment is all the more critical as the notion that cutting-edge AI is an exclusively American domain fades, given the proven capabilities of countries like China—and France, with its strong mathematical tradition perfectly aligned with the challenges of neural networks. DeepSeek has shown the world that, with just a few million dollars and limited graphics cards, it’s possible to achieve results that rival those of American giants. Barely a year ago, Mistral also unveiled a model that competed with OpenAI’s, developed in a matter of months by a team of just a few dozen people. France’s AI expertise is undeniable. This talent is also evident within U.S. tech giants: Yann LeCun, Meta’s chief AI scientist, has inspired an entire generation. His company’s open-source model, LLaMA, was initially developed by a Paris-based team.

    Many of us already recognized in 2023 the rise of a more efficient and refined AI than that of California’s giants. French minds often find opportunities in Big Tech to apply their mathematical brilliance. Several of Mistral’s founders, in fact, honed their skills in these companies. However, if every European success is ultimately absorbed by American giants—as Mistral nearly was—the benefits for Europe will remain minimal. Given the economic upheaval AI brings, such a trend would lock us into dangerous dependency. Transhumanist visionaries have no real plan for Europe beyond its picturesque landscapes.
    The development of infrastructure and data centers, backed by massive investments, is essential for our autonomy. While France’s efforts in this direction are commendable—assuming they materialize fully—they must avoid hiding behind convoluted consortia reminiscent of Airbus-era strategies. Yet we cannot overlook the need for a deeper reflection on funding sources, decision-making balance with international partners, and the long-term viability of these projects.

    This also requires addressing the persistent technological deficit in public administration, despite the renewed focus on industrial policy. Scattered funding, insufficient analysis, and the excessive event-driven communication of “France 2030,” along with the overhyped “hydrogen revolution” and reindustrialization statistics skewed by self-employment, demand a more fundamental effort from the state. This is especially urgent as global political shifts threaten to disrupt the open-source ecosystem, which is central to Europe’s AI catch-up strategy.

    Open source represents a remarkable opportunity for technological knowledge sharing. Yann LeCun is a vocal advocate, and he seems receptive to the idea of his home country reclaiming its rightful place in scientific tradition. However, given U.S. officials’ outcry against DeepSeek and calls for stricter restrictions, there is a risk that Big Tech’s dominance could tighten further, leaving only China as a credible counterbalance. Governments will now have to address the circulation of AI models and open-source frameworks as a key issue in trade negotiations.

    Europe will not match the scale of American investments. Yet DeepSeek, Mistral, and others worldwide have proven that we can reposition ourselves in the digital landscape—by relying on open source for now, but above all by placing engineering culture, with all its versatility, back at the core of our strategic decisions. This path, neglected by Europe over the past three decades, is the one being followed by BRICS nations that are effectively positioning themselves in the tech race. We will not succeed by focusing solely on regulatory questions, but by restoring scientific culture to the heart of our choices.

    This text was originally published on the website of Les Echos.

  • Semiconductors Are the Achilles’ Heel of the AI Giants

    Semiconductors Are the Achilles’ Heel of the AI Giants

    The ultra-concentration in the design and production of semiconductors for AI, centred around Nvidia and TSMC, is fuelling the interest of digital giants, which are highly dependent in this regard. However, catching up looks to be a difficult task, despite the mobilisation of state actors.

    The explosion of artificial intelligence rests on two pillars of a different nature: on the one hand, the development of large language models such as GPT, and on the other, spectacular computing power with dedicated processors. These are designed in particular by the omnipresent giant Nvidia, and manufactured by a tight handful of actors, especially the Taiwanese company TSMC. AI models and semiconductors both require gigantic investments and cutting-edge expertise. However, these are two worlds that, although they cooperate closely, respond to very different requirements.

    In terms of model development, American digital giants such as Microsoft, Meta and Google have all the technological, economic and political resources to dominate the sector, both internally and through acquisitions/partnerships. This latter aspect even enables them to domesticate the diversification seen with the explosion of open source, i.e., models that are freely distributed and reusable by anyone. Although open source allows an entire AI ecosystem to exist, it cannot exactly be seen as David’s weapon against Goliath, as the giants themselves are deeply invested in it. Meta’s LLaMa language models are, for example, open source. Moreover, the financial weight and grip of Big Tech are such that we are seeing independent actors being drawn into their orbit one after the other. The French gem Mistral recently announced it was joining Microsoft’s fold, entrusting it with the distribution of its most advanced model, which will therefore be closed. The giants thus have ample means to maintain control over model development.

    Nevertheless, behind the domination of these behemoths, the importance of the processors that enable the training of these AI models should not be underestimated. It is in fact the crux of today’s technological warfare and lies in the hands of industrial giants of a different kind. The entire AI scene remains highly dependent on a semiconductor design and production chain that is incredibly concentrated, revolving around Nvidia and TSMC.

    A boom in demand for semiconductors dedicated to AI, and very few suppliers

    For digital giants, autonomy in terms of semiconductors remains a challenge in which it is difficult to position oneself. After years of investment, Nvidia holds a near-monopoly position in the design of semiconductors dedicated to AI. The American company designed 80% of this type of semiconductor worldwide last year.

    Once the design is completed, Nvidia outsources their manufacturing to Taiwan’s TSMC, one of only ten companies in the world capable of producing them. Nvidia is said to be “fabless”. In this industry, manufacturing a semiconductor requires a production line with specific characteristics (manufacturing equipment, testing and packaging). These new production lines are extremely costly. A brand-new factory (or foundry in the sector’s terminology) requires between 15 and 20 billion dollars and a minimum of two years of construction. Very few economic players can invest such colossal sums and overcome the entry barriers to the foundry market (“Fabs”).

    States are seizing the issue in the name of technological sovereignty

    Despite the enormity of the investments, some actors are entering or returning to this market, such as the American Intel or the Japanese Rapidus. Manufacturers already in the race, like TSMC or South Korea’s Samsung, are continuing to invest in an attempt to maintain their market shares. After the Covid-19 crisis and the subsequent semiconductor shortage, several states decided to relaunch their financial support for the sector. “Chip Acts” have multiplied to increase national semiconductor manufacturing capacity, bolster economic security and guarantee supplies for military use even in times of crisis. Among these countries are the United States in 2022 with the Chips and Science Act ($39 billion), the European Union with the Chip Act (€43 billion in 2023), Japan with the creation of the Rapidus conglomerate and a support plan ($100 billion for Rapidus and new TSMC factories over 2023–2027), China with the launch in 2023 of phase 3 of the Chinese government’s semiconductor fund ($46 billion for 2023–2027), and South Korea with a government plan of $7.3 billion. In the United States, the leverage effect of public subsidies in the sector is noteworthy. The $39 billion of the Chips and Science Act encouraged a wave of private investments amounting to $200 billion, spent by American and foreign companies on American soil.

    New entrants and a new scale of financing

    Until new factories produce more chips, supply will not be able to meet global demand for AI-dedicated chips. Hence a significant rise in prices. A Nvidia GPU (H100) can cost up to $40,000 per unit. Its availability is limited, because even with increased production volumes, the company still cannot meet market demand. Some users and buyers of Nvidia chips are concerned about being dependent on a single supplier. This is the case for Sam Altman, CEO of OpenAI, because the lack of AI chips risks hindering the development of his own company. Why not try to create one’s own industrial tool to restore this supply-demand imbalance? This is the logic of every new entrant in a booming sector. Sam Altman has been holding numerous meetings with manufacturers and investment funds over the past few months. In his initial estimates, he mentioned a (staggering) investment goal of $7 trillion to build a new segment of the semiconductor industry. The project is still ongoing. And Altman is not alone. Initiatives are springing up. Apple is working with TSMC to manufacture AI chips. The head of the Japanese group SoftBank, Masayoshi Son, wants to turn his group into an AI powerhouse. His latest project is to enable its subsidiary ARM to create a new AI chip division. A prototype will be tested in spring 2025, and mass production should begin in autumn 2025. For its part, Nvidia is maintaining its technological lead in a rapidly growing market. According to the Canadian research centre Precedence Research, the global market is expected to reach $100 billion by 2029 and $200 billion by 2032.

    This new type of shortage is prompting digital giants to position themselves in the segment, each in their own way. Faced with these ambitions, Nvidia continues tirelessly to position itself to do even better, notably better than what the giants will probably be able to achieve in designing AI-dedicated processors. The digital giants find themselves caught in an industrial vice that will be difficult to overcome. The prospect of balanced global competition in which all major regions manage to position themselves remains distant and uncertain. Beyond their own interests, the ultra-concentration of AI-dedicated semiconductors highlights a very real risk to industrial resilience across the entire chain, down to end users. In this regard, diversification is a major political issue.

    This piece was originally published by the French Institute for International and Strategic Affairs – IRIS.

  • Mistral Under Microsoft: Europe’s AI Catch-Up Challenge Remains Unresolved

    Mistral Under Microsoft: Europe’s AI Catch-Up Challenge Remains Unresolved

    Mistral AI’s move into Microsoft’s sphere has sparked political criticism in Europe. As a champion of open source, the company had recently advocated for a more flexible AI Act before announcing its shift to a closed model. Nevertheless, its technical success in developing foundational models with limited resources demonstrates Europe’s—and other global players’—potential to catch up. However, achieving true autonomy would still require overcoming a difficult economic equation that pushes the most promising startups into the arms of Big Tech.

    Mistral’s Success Highlights Europe’s Technical Potential in the AI Race

    Many observers had assumed Europe was destined to remain merely a user of American AI models for developing various applications. Technically, Mistral’s success confirms the opportunity for a relatively resource-efficient AI compared to Big Tech’s massive data usage and financial and human resources.

    In just a few months, Mistral managed to develop AI models that rival OpenAI, Google, and Meta in performance, with significant but far more limited resources than those of the American giants. This is particularly striking in terms of workforce, with its team of around thirty employees. This achievement not only showcases the team’s prowess but also sheds light on the nature of the technology driving the generative AI boom.

    Beyond new neural network architectures (like transformers), the spectacular progress in AI over the past decade has largely been due to the use of enormous amounts of data and computing power. While riding this wave of quantitative explosion, Mistral has also carved out a path for more refined AI engineering, allowing it to establish itself on the global stage in record time.

    Even amid an educational crisis and severe deindustrialization, it remains possible to mobilize skills from top-tier training programs to compete with global tech giants. Beyond the issue of European autonomy, this technical reality offers valuable lessons about the global AI race. Catching up and competing in AI is possible, provided there is sustained funding and market opportunities.

    Mistral’s Move into Microsoft’s Sphere Illustrates the Economic Challenge of Independent and Open AI

    After positioning themselves as champions of open, reusable models, Mistral’s leaders decided that their new, most advanced model would be closed—distributed through an agreement with Microsoft, which is also taking a stake in the company. The open-source approach had boosted Mistral’s appeal among developers, alongside other open models like Meta’s LLaMA, in contrast to the now radically closed model of the misnamed OpenAI.

    In fact, it was precisely this shift that led Elon Musk, who had been involved in OpenAI’s launch, to recently announce legal action against Sam Altman’s company. Beyond the irony of the billionaire’s outbursts, it is true that OpenAI, with its labyrinthine structure, reflects a gap between its original open-source and research-focused mission and its current purely commercial purpose. The issue of Big Tech’s grip on AI is particularly sensitive for Europe but is also relevant in the United States.

    Like OpenAI, Mistral’s agreement with Microsoft confirms its technical success and popularity. The French company is also launching a chatbot called “Le Chat,” modeled after ChatGPT. However, this partnership, for now, buries the dream of an independent, open-source European AI.

    Beyond the recent virulent attacks on the company’s leadership, we must question the European economic environment. The core issue remains the prospects for development, funding, and commercial opportunities needed to maintain a leading position in the digital sector. These challenges and the financial power of tech giants inevitably draw successful startups into their orbit. It is this economic aspect that has turned Mistral’s technical feat, which could have marked a turning point toward autonomy, into a strategic setback for Europe.

    Beyond Distrust of Lobbying, a Flexible Approach to AI Regulation Remains Essential

    The AI Act addresses an obvious need for regulation and risk management in AI. However, its complicated development has resulted in particularly convoluted agreement terms. Its creators had missed the generative AI revolution and embarked on a titanic adaptation effort last year.

    The idea of positioning Europe as the world’s digital regulator, with too little concern for the continent’s technological offerings, poses an existential risk to the European economy and its competitive autonomy. Moreover, with its difficult application to future technical developments, the AI Act risks serving the interests of Big Tech, which has the means to navigate these regulatory labyrinths. Mistral’s move into Microsoft’s orbit seems to confirm this.

    Mistral had strongly advocated at the end of last year for a loosening of the AI Act, particularly regarding open-source foundational models of generative AI. It is natural to think that the company had already considered its shift to a closed model in partnership with Microsoft. Nevertheless, the concessions made in response to objections from the French and German governments, defending their national companies like Mistral and Aleph Alpha, mainly concerned open source, which will thus benefit from greater flexibility. While Mistral’s reversal may be regrettable, its lobbying primarily resulted in a loosening of the AI Act that could, under certain economic conditions, encourage the emergence of future open-source competitors.

    This piece has initially been published by the French Institute for International and Strategic Affairs – IRIS.

  • AI Act: What Compromise Will Enable the Rise of European Artificial Intelligence?

    AI Act: What Compromise Will Enable the Rise of European Artificial Intelligence?

    the European Union reached a political agreement to regulate the development of artificial intelligence (AI). What does this unprecedented global framework entail, and what are its implications for EU member states and tech industry lobbies? What were the main sticking points in negotiations between EU institutions and certain countries? What do these disputes reveal about the solidity of the agreement? Why is AI a critical issue for Europe, and what would be the economic consequences for the continent? Where does France stand in this debate?

    What Does This Unprecedented Global Agreement Entail, and What Are Its Implications?
    The AI Act, in development since 2021, has faced numerous challenges, particularly due to the explosion of generative AI, which disrupted its original risk-based approach. Initially designed to classify applications—from harmless spam filters to unacceptable uses of facial recognition in daily life—the regulation had to be hastily revised to address generative AI’s unexpected rise. While the need for regulation is undeniable, the last-minute additions risk stifling European startups just as they begin to close the gap, burdening them with complex rules that ironically favor more advanced U.S. giants.
    The rapid progress of large language models, built on neural networks with billions of parameters trained on opaque datasets, raises concerns about privacy, copyright, and security risks tied to their unpredictable behavior. Given AI’s breakneck evolution, a flexible, adaptive regulatory approach is essential. Yet the current framework—hundreds of pages of self-referential legal considerations—risks quick obsolescence.
    Beyond mere exemptions, flexibility is crucial, especially given the growing role of open-source AI, which European startups are leveraging. Many repurpose existing models from tech giants, while some are now developing their own foundational models. An open regulatory approach is needed to address emerging risks while fostering innovative, homegrown European AI capable of competing with U.S. and Chinese dominance.

    Key Disputes in Negotiations: What Do They Reveal About the Agreement’s Strength?
    EU lawmakers initially sought to replicate the GDPR’s success—a global gold standard for data regulation—but AI presents a different challenge. Europe already lags behind U.S. tech giants, and the AI Act introduces uncertainty just as European startups like Mistral (France) and Aleph Alpha (Germany) begin gaining traction.
    In recent weeks, France, Germany, and Italy pushed back, creating a cacophony over two issues:

    State use of facial recognition (some member states refuse to fully abandon it).
    Preserving the potential of startups working on foundational models (the backbone of generative AI).
    These governments proposed self-regulation and codes of conduct for such models, clashing with EU institutions rushing to finalize the agreement amid pressure from NGOs advocating for strict adoption. The compromise includes broad exemptions for open-source developers, central to Europe’s AI foundation models.
    French Digital Minister Jean-Noël Barrot claimed the deal would preserve Europe’s ability to develop its own AI technologies and strategic autonomy. But why is AI such a critical issue for Europe, and what are the economic stakes? Where does France fit in?

    Why AI Is a Major Stake for Europe—and What’s at Risk for Its Economy?
    Europe is falling behind not only the U.S. but also China—a situation that was not inevitable. Neural networks owe much to European pioneers, whether they stayed on the continent or moved abroad. Geoffrey Hinton, the “godfather of AI,” left Silicon Valley for Toronto to distance himself from U.S. military influence, while Yann Le Cun (Meta’s Chief AI Scientist) and Sepp Hochreiter (who introduced long-term memory in neural networks in 1991) laid the groundwork for today’s transformer-based language models, the core of generative AI since 2017.
    Despite educational crises and declining math proficiency, Europe—particularly France—still hosts pockets of excellence that could drive a distinctive AI approach. The idea that Europe should settle for a regulatory role, dependent on U.S. and Chinese tech, is economically and strategically suicidal, given AI’s pivotal role in technological development. Historically, mastering cutting-edge technology has been key to catching up, growing, and projecting power. Yet for decades, the EU focused on competition policy over industrial strategy, treating citizens more as consumers than producers—a trend the AI Act risks perpetuating.
    While Thierry Breton’s leadership marks a shift toward industrial sovereignty, the task remains monumental. The AI Act must balance regulation with innovation, ensuring Europe doesn’t just consume AI but develops it. The alternative—a future where Europe remains a rule-maker but not a tech-maker—would be a strategic failure.

    This piece was originally published by the French Institute for International and Strategic Affairs – IRIS.