THE MISTRAL PARADOX: 3 STEPS TOWARD RETHINKING SOVEREIGNTY IN THE AGE OF AI

By Dr. Tawhid CHTIOUI, Founding Presidentof aivancity School of AI & Data for Business & Society; selected by Keyrus as one of the 25 most influential global figures in the field of AI and data (January 2025).

---

ACT 2: France knows how to develop talent. Why does it struggle to build power?

Introduction: The French Paradox

France faces a paradox regarding artificial intelligence that it can no longer afford to accept. It has renowned researchers, rigorous scientific institutions, high-performing laboratories, and entrepreneurs capable of launching ambitious technology companies. Yet when it comes to transforming this excellence into global industrial power, the landscape changes dramatically.

The figures illustrate the extent of the gap. In 2025, U.S. companies attracted nearly $286 billion in private investment in AI, compared with $4.36 billion for France. In generative AI alone, the United States accounted for $163.6 billion in private investment, while China and Europe combined totaled $4.7 billion. The OECD reaches the same conclusion using a different method: U.S. companies captured approximately 75% of the global value of venture capital invested in AI in 2025, compared to 6% for the European Union as a whole (Stanford HAI, 2026; OECD, 2026).

These disparities do not mean that Europe has stopped innovating. They point to something even more troubling: in contemporary AI, scientific quality and technological power have become two distinct realities.

Mario Draghi had articulated the problem particularly clearly in his report on European competitiveness: Europe lacks neither ideas, nor researchers, nor entrepreneurs; rather, it far more often fails to turn innovation into global commercial success, particularly because its companies face difficulties with financing and scaling up in a market that remains fragmented (Draghi, 2024). Two years later, the European Commission itself acknowledged that the continent continues to lag behind in computing power, semiconductors, and certain infrastructure essential to the development of AI (European Commission, 2026).

This distinction is essential. For a long time, we thought of scientific power in terms of universities, laboratories, publications, and talent. Artificial intelligence forces us to look at the entire value chain. To turn a discovery into power, we must now be able to quickly mobilize billions in capital, access tens of thousands of advanced processors, secure the energy needed to run them, attract and retain the best researchers, have a market large enough to grow, and be willing to finance several years of uncertainty before a technology becomes profitable. The OECD also shows that, since 2023, IT and hosting infrastructure has become the leading sector for venture capital dedicated to AI, with $109 billion invested in 2025 alone (OECD, 2026).

In other words, the new technological frontier no longer rewards only those with the best ideas. It favors those who have the ecosystem capable of scaling them up.

Perhaps this is where the true vulnerability of France and Europe lies today. For a long time, we have expected our educational and scientific systems to produce knowledge and talent. We must now ask ourselves what becomes of this intellectual capital once it has been produced: who funds it, who provides the resources for it to be applied, who commercializes it, who opens up a market for it, and, ultimately, in which country it creates value.

For a nation can produce excellent researchers while failing to provide sufficient funding for their ambitions. It can produce discoveries that others will use to build businesses. It can even become a formidable talent-development system for the technological powers it then seeks to compete with.

So perhaps the challenge for France is not to produce more intelligence. It is to understand why we have such a hard time turning the intelligence we produce into power.

We now need to take this mechanism apart.

I. The Great Confusion: Talent Does Not Mean Sovereignty

For too long, we have confused two things: having talent and having technological power. The two are obviously linked, but they are not the same thing. A country can train excellent mathematicians, publish cutting-edge research, and host renowned laboratories without being able to foster the companies, infrastructure, and market positions that will enable it to sustainably master the technologies resulting from that research. One could almost summarize the problem with a deliberately provocative equation: talent + research ≠ technological sovereignty.

There is an entire processing chain missing between the two.

In today’s world of artificial intelligence, a scientific discovery must have access to sufficient capital, immediately available computing power, digital infrastructure, energy, high-quality data, engineers capable of turning research into a product, entrepreneurs capable of building an organization, and a customer base large enough to enable rapid scaling. It must then be able to finance this growth over several years, attract new talent, secure access to critical components, and compete against rivals who sometimes have resources far beyond those of a young company. Speed itself becomes a strategic resource.

An excellent innovation that takes five years to reach the market may be surpassed by a technology that is slightly less advanced but can be commercialized in eighteen months. At the cutting edge of AI, where new generations of models now follow one another within a matter of months, the time between research and commercialization is becoming almost as important as the quality of the research itself.

Data from the Stanford AI Index illustrates this new reality. Global computing capacity dedicated to AI is estimated to have grown by an average of 3.3 times per year since 2022. At the same time, more than 90% of the models considered notable in 2025 were produced by industry, while universities continue to play a major role in basic research and influential publications (Stanford HAI, 2026).

This shift is significant. It means that the frontier of innovation increasingly lies within organizations capable of bringing together science, capital, engineering, and infrastructure simultaneously. Knowledge remains essential, but it is no longer sufficient on its own to reach the technological frontier.

Yet the literature on national innovation systems had long identified this challenge. The work of Lundvall and Nelson has shown that innovation never results simply from the excellence of a single laboratory or company. It depends on interactions among universities, companies, funders, government agencies, markets, and institutions capable of transforming knowledge into economic activity (Lundvall, 1992; Nelson, 1993). AI is now taking this logic to an extreme, because the cost and complexity of certain infrastructures make this interdependence much more visible.

The European Commission itself seems to have gradually embraced this shift in perspective. Its plan for an “AI continent” is not based solely on training researchers. It explicitly combines computing power, access to data, funding, the adoption of AI in businesses, and skills development. In particular, the EU plans to establish up to five AI gigafactories, nineteen AI Factories, and aims to mobilize 200 billion euros in investments. This program implicitly acknowledges a shortfall: the Commission itself describes Europe’s lack of large-scale computing infrastructure as a bottleneck for the continent’s competitiveness and strategic autonomy (European Commission, 2025–2026).

We must add to this chain a player whose role is sometimes underestimated in Europe: the customer.

Governments can help create a market for their own technologies through government procurement, industrial policies, or major programs. Private companies can do the same when an economic ecosystem is large enough to immediately offer opportunities to new entrants. A technology does not become strategic simply because it is excellent; it becomes strategic when it quickly gains enough users to finance its improvement, accumulate data, attract investors, and gradually establish its standards.

Public procurement, in particular, is therefore not merely a budgetary tool. It can become an instrument of sovereignty when it enables emerging technologies to find their first major domestic markets. This is where the comparison with the United States and China becomes illuminating. Their advantage does not simply lie in having more researchers or entrepreneurs. They have developed—following very different approaches—systems capable of linking research, funding, infrastructure, industry, and the market.

For a long time, Europe has viewed innovation as a matter of knowledge. The major technological powers also treat it as a value chain.

Sovereignty, therefore, does not consist in possessing each of these links separately. It presupposes that capital can meet talent, that talent can access calculation, that calculation can become a product, that this product can find a market, and that the company that created it can grow without having to leave its ecosystem to find the means to achieve its ambitions.

In other words, the real question is no longer just how many talented people we are training. It is what collective system we have built to transform their intelligence into power.

It is on this point that the paths of the United States, China, and Europe diverge.

II. Three Different Power Generation Systems

The United States, China, and Europe do not build their technological power in the same way. Their economic, political, and institutional models differ profoundly. Yet a comparison reveals a key distinction: the first two powers have succeeded—each according to its own logic—in establishing a chain capable of rapidly transforming research into technologies, and then those technologies into markets.

In the United States, this ecosystem functions first and foremost as a capital-generation machine. Major universities produce research and attract international talent; venture capital funds extremely uncertain projects at a very early stage; and Big Tech companies then provide computing power, cloud infrastructure, data, distribution networks, and financial clout that are hard to match. Finally, acquisitions make it possible to quickly integrate emerging innovations into large platforms.

The financial scale of this system has become staggering. In 2025, U.S. private investment in AI reached $285.9 billion—more than twenty-three times the amount recorded in China—while 1,953 new U.S. AI startups received funding, more than ten times the number in the next-highest country (Stanford HAI, 2026). These figures do not mean that all this money necessarily leads to better innovations. They do, however, demonstrate the exceptional depth of an ecosystem capable of simultaneously funding thousands of experiments and a few massive industrial ventures.

China has built a different system that could be described as an “industrialization machine.” The government sets strategic priorities and invests in infrastructure, semiconductors, computing power, and the integration of AI into the economy. But it would be a mistake to imagine a system that is simply administered from above. Alongside giants like Alibaba, Tencent, Huawei, Baidu, and ByteDance, a particularly intense competition has emerged among DeepSeek, Z.ai, Moonshot, MiniMax, and many other players.

Recent research from the Stanford Institute for Human-Centered AI thus describes a Chinese ecosystem that is far more diverse than is often perceived in the West, in which more than a dozen organizations are developing high-performance open models or models with open weights. The authors highlight the role of public support but note that it does not, on its own, explain this dynamic: competition among companies, the pursuit of computational efficiency, and the desire to facilitate the deployment of models also play a major role (Meinhardt et al., 2025).

China’s official strategy, in fact, reflects this commitment to rapidly transition AI from the laboratory to the real economy. The “AI+” initiative explicitly calls for its integration into industry, research, services, and production chains in order to shorten innovation cycles and accelerate industrial modernization (State Council of the People’s Republic of China, 2025).

Europe presents a third face: that of fragmented excellence.

It boasts leading researchers, renowned universities, major industrial companies, a market of nearly 450 million consumers, and substantial savings. Yet these resources still do not flow easily between countries, financial markets, universities, and companies.

The Draghi report bluntly sums up this challenge: no company in the European Union founded in the last fifty years has achieved a market capitalization exceeding 100 billion euros, whereas the six U.S. companies valued at more than 1,000 billion were all founded during that period. Furthermore, nearly 30% of the unicorns founded in Europe between 2008 and 2021 have moved their headquarters abroad. Draghi emphasizes the core cause: Europe generates many ideas but struggles to turn them into global commercial successes, largely due to the fragmentation of the single market and the lack of financing for scaling up (Draghi, 2024).

This weakness is now widely recognized, to the point that the Union itself is seeking to address it through new policies on funding, AI infrastructure, and the simplification of the internal market. But the assessment remains harsh: Europe possesses many of the resources necessary for technological leadership but has not yet managed to ensure the same level of continuity as its competitors.

Europe may not be short on building blocks. What it still lacks is a mechanism capable of assembling them quickly and effectively.

It is this difference that changes the way we must now evaluate our educational and scientific systems. Developing more talent will remain essential, but it will not be enough if the ecosystem in which they operate does not then allow them to conduct research, launch ventures, secure funding, commercialize their innovations, and grow on the same scale as their American or Chinese competitors.

III. Rethinking Our Definition of Educational Excellence

This situation should prompt us to take a closer look at a metric that is particularly important to us: the number of talented individuals we train.

We rightly celebrate our engineers, mathematicians, researchers, and graduates when they join the world’s top laboratories. But if a growing proportion of these talented individuals go on to build America’s technology leaders, if their discoveries are commercialized elsewhere, or if the innovations coming out of our laboratories find the capital and markets needed for their development abroad, then our measure of excellence remains incomplete.

This is obviously not about calling international mobility into question. Science advances because researchers move around, compare their work, and join environments that allow them to push the boundaries further. A sovereignty policy that seeks to keep talent within national borders would ultimately weaken the very research it claims to protect.

The question is a different one: What remains in the ecosystem that helped shape these talented individuals once they have succeeded elsewhere? And, above all, is this ecosystem attractive enough for them to return, start ventures, share their discoveries, or build their businesses there?

This is where higher education’s role is shifting. While training outstanding scientists remains its primary mission, it must also learn to better connect research, entrepreneurship, industry, and society. The creation of spin-offs, collaborations with companies, technology transfer, joint patents, mobility between laboratories and industry, and even the ability of researchers to become entrepreneurs are no longer peripheral activities to the academic mission. They are now an integral part of the path that leads from knowledge to impact.

The European Commission itself appears to recognize this shift in perspective. In 2026, it proposed a new framework for measuring the “commercialization of knowledge” that goes beyond publications and patents. The proposed indicators now include joint patents between universities and companies, venture capital invested in spin-offs, and the number of startups founded by PhD holders. The stated goal is precisely to better measure research’s ability to generate economic and social value and to contribute to Europe’s strategic autonomy (European Commission, 2026).

Academic research on technology transfer shows, moreover, that this transformation cannot be imposed by decree. An analysis of more than 400 European universities by Wolszczak-Derlacz (2025) highlights, in particular, the importance of relationships between universities and businesses in the ability to generate and share intellectual property. In other words, scientific excellence has a greater impact when it is linked to an environment capable of extending it beyond the laboratory.

This leads us to broaden our definition of academic performance. A prestigious school or university should no longer be able to measure itself solely by how many graduates it produces, how many researchers it publishes, or how many of its alumni work at Google, Meta, or OpenAI. It should also be able to demonstrate how many companies have emerged from its labs, how many technologies have reached the market, how many researchers move between academia and industry, how many graduates become entrepreneurs, and what scientific, economic, or social value this intellectual capital generates within its community.

Educational sovereignty, therefore, is not measured solely by the number of talented individuals we educate, but by our collective ability to transform their talents into scientific, economic, and social power.

This shift in perspective is essential. It is no longer a matter of pitting universities against businesses, basic research against entrepreneurship, or international mobility against national interests. On the contrary, it is about building more bridges between these worlds. For the real failure would not be that our best researchers one day go to work at Stanford, at OpenAI, or in a Chinese laboratory. It would be that they never find the conditions in France or Europe that would inspire them, one day, to build something just as ambitious there.

IV. The Chinese Shock Reveals a European Mistake

For a long time, Europe viewed its technological lag almost exclusively through the lens of its comparison with the United States. The equation seemed simple: on one side, American platforms, their capital, their cloud infrastructure, and their proprietary models; on the other, a Europe seeking to preserve its autonomy without having the same resources at its disposal.

China's rise is changing this landscape.

China isn’t just pitting new champions against OpenAI, Google, or Anthropic. Part of its ecosystem offers a different combination: models that are now close to the technological frontier, often lower costs, and, above all, a much more proactive policy of distributing open-source models that can be downloaded, adapted, and run outside their developers’ platforms.

Stanford HAI thus considers that Chinese open-weight models have now caught up with—and in some applications even surpassed—several of their international competitors, and emphasizes that their global spread could alter the balance of technological dependence among nations (Meinhardt et al., 2025). Qwen is already widely used by developers outside of China, while DeepSeek has demonstrated how quickly a Chinese model can spread globally.

This trend is particularly significant for countries that lack both the capital and the infrastructure needed to train their own state-of-the-art models. The Carnegie Endowment thus highlights that another competition is emerging, particularly in the Global South: it is no longer just about who will build the most powerful model, but about who will offer the most accessible, adaptable, and easy-to-deploy technologies. After its launch, DeepSeek became one of the most downloaded apps in more than 140 markets, notably in Brazil and India (Sheehan et al., 2025).

The thinking of a government, a university, or a company can then change profoundly. Why rely exclusively on a closed U.S. API—where the provider retains control over the model, its development, and often its terms of use—if a sufficiently powerful Chinese model can be deployed on locally chosen infrastructure, tailored to the country’s needs, and, within certain limits, controlled by the entity operating it?

This obviously does not mean that the Chinese model has become “sovereign” simply because it is implemented locally. Its design, training, and certain technological dependencies remain external. Issues related to security, licensing, auditing, culture, and governance remain, as we have seen. But the nature of the dependency relationship has changed.

This may be where one of the most significant geopolitical innovations in China’s strategy lies. China can spread its technological influence without necessarily forcing the world to use a platform located in China. Its models can travel separately from their original infrastructure.

This strategy resonates with a particularly strong demand in emerging economies: to have access to high-performance tools without completely relinquishing control over their data, to be able to adapt models to their languages and customs, and to avoid exclusive dependence on major U.S. platforms. Carnegie also points out that the Sino-American competition for AI in the Global South is increasingly playing out in the areas of accessibility, cost, infrastructure, and local adaptation—and no longer solely on the absolute performance of the models (Lu & Winter-Levy, 2025).

Europe may have made an analytical error by viewing AI for too long as a race in which it simply had to catch up to the United States. China is gradually transforming the race itself.

It is not merely seeking to shift the global center of innovation to Beijing, Hangzhou, or Shenzhen. It may also be seeking to make its technologies the building blocks upon which other countries will build their own artificial intelligence.

If this strategy prevails, the geopolitics of AI will no longer be merely a competition over who has the best model. It will become a competition over whose models will become the cognitive infrastructure for others.

And it is this new landscape that calls for a more nuanced reexamination of Mistral's strategy.

V. What if Mistral made a rational choice ?

There is, however, one possibility that the debate on Europe’s “falling behind” tends to overlook: what if Mistral’s evolution were not merely a matter of reacting to circumstances, but also reflected a clear-eyed assessment of how the AI market is currently transforming?

Shahin Vallée, a researcher at the German Council on Foreign Relations, frames the hypothesis in a deliberately provocative way. In his view, Europe currently lacks the resources needed to win the race for cutting-edge models on its own and should prioritizethe adoption of AI over the pursuit ofcutting-edge innovation alone. He therefore views Mistral’s repositioning toward inference, infrastructure, and enterprise deployment as a potentially rational strategy, provided that high-performing open models continue to be available (Vallée, 2026).

The hypothesis is uncomfortable because it can be interpreted in two ways.

The first scenario is pessimistic: Mistral would realize that it is becoming extremely difficult to keep up, in the long term, with the financial and computational pace set by the American and Chinese giants, and would gradually shift its focus toward activities less exposed to this race.

The second is more strategic: the economic battle over AI may not be won simply by having the number-one-ranked model at all times.

As the performance of the top models converges and high-performing open-source models become available, an increasing share of the value may shift toward what companies build around them: the infrastructure on which they run, the ability to select and orchestrate multiple models, their adaptation to proprietary data, security, compliance, agents capable of acting within information systems, and, above all, their effective integration into the organization’s processes.

Mistral is clearly expanding its offerings in this direction today. The company no longer positions itself solely as a research lab for state-of-the-art models. It now offers a much broader stack combining models, computing infrastructure, agent creation tools, model customization and alignment, and applied AI applications and services. Its official strategy explicitly aims to enable organizations to build custom systems based on their own knowledge while retaining control over their infrastructure and their “learning loop” (Mistral AI, 2026a).

This shift could become significant if the underlying models follow a trajectory similar to that of other technologies that have gradually become components of a larger system. When multiple engines become sufficiently powerful, the competitive advantage no longer necessarily lies in the engine alone, but in the vehicle built around it, in its use, and in the way it integrates into a specific environment.

It would be premature, however, to conclude that the battle over models no longer matters. Those who control the frontier retain a capacity for innovation, negotiation, and independence that users of others’ models do not fully possess. Vallée himself acknowledges the strategic cost of his scenario: prioritizing adoption amounts, at least in part, to accepting that others will continue to push the technological frontier (Vallée, 2026).

That is exactly why it would be too simplistic to choose right now between two narratives: one in which Mistral is falling behind, and one in which Mistral has discovered before anyone else where value will lie in the future.

Perhaps we are witnessing both trends at once: a relative slowdown in the race for raw computing power, but an attempt to stake a claim on another front—that of their adoption by businesses, institutions, and governments.

If this hypothesis is confirmed, it will once again change the European landscape. The question will no longer be simply whether we can produce the world’s best model, but rather which stages of the intelligence chain we absolutely must control in order to avoid becoming dependent on those who produce it.

Conclusion. A rational strategy for Mistral is not necessarily a sufficient strategy for Europe

One might conclude that Europe must simply accept the new international division of labor: the United States and China would produce the most advanced models, while we would specialize in hosting, securing, orchestrating, and integrating them into businesses.

This strategy may make economic sense. In the short term, it may even significantly accelerate the adoption of AI in a Europe where parts of the economic fabric still lag behind in terms of AI usage. Shahin Vallée is right to point out that technology only has an impact when it is actually adopted, and that sovereignty is not built solely in laboratories (Vallée, 2026).

But turning this observation into an industrial doctrine would be dangerous.

For there is a fundamental difference between making smart use of others’ innovations and gradually losing the ability to produce them oneself.

If Europe abandons the development of its own models on the grounds that others can build them faster and more cheaply, it risks becoming extraordinarily proficient at integrating technologies whose future trajectories it will no longer control. It will be able to choose among several suppliers as long as several suppliers exist, benefit from open models as long as they remain open, and build its infrastructure around innovations whose broad directions will be decided elsewhere.

There would be something paradoxical about this sovereignty: we would be gaining ever greater control over the use of an intelligence that we would have ever less control over in its creation.

The issue, therefore, is not a choice between innovation and adoption. A sustainable technological power must be capable of doing both. Adoption without innovation leads to dependence; innovation without adoption leads to scientific excellence without economic impact. The real challenge lies in the ability to sustain the cycle that links research, industry, applications, and new research.

This is where the Mistral case takes on its full significance. If its repositioning enables the company to become a powerful European player in infrastructure, orchestration, and applied AI while maintaining a credible capacity for model research and development, it could represent a new form of technological power. If, on the other hand, the “model” aspect gradually becomes secondary because decisive innovations consistently come from elsewhere, we must have the honesty to acknowledge that this is no longer exactly the kind of sovereignty we thought we were building in 2023.

The debate obviously goes beyond any single company. It concerns the collective choice that France and Europe must make in the face of a historic shift in scale.

The Draghi report had already emphasized the need to bridge the gap between invention and commercialization, but the challenge posed by AI adds an additional complication: we must learn to industrialize more without giving up our role as a hub for creating the next generation of technologies (Draghi, 2024).

This means we must stop treating these policies as separate entities—policies that have all too often been developed in isolation: academic excellence, industrial policy, capital, energy, computing infrastructure, public procurement, entrepreneurship, and the development of technology companies must all become components of a single strategy.

France can continue to train outstanding researchers. Europe can expand its funding programs and infrastructure. Mistral can become a powerful platform for integrating global models. All of this is useful. But none of it fully addresses the issue of sovereignty.

For being sovereign does not mean producing everything ourselves, nor does it simply mean knowing how to use what others produce. Between self-sufficiency—which is impossible—and dependence—which is dangerous—we must determine what we absolutely must control, what we can share, what we can buy, and what we must never lose the ability to rebuild ourselves.

Perhaps that is where the real question begins.

Nor: “Can Europe still win the AI race?”

But: What forms of sovereignty do we really need to remain in control of our own choices in a world where no form of intelligence will be entirely national?