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 3: After Mistral: Who Will Rule in the New AI World Order?

Introduction: Moving Beyond the Binary Definition of Sovereignty

For a long time, we have thought of sovereignty in terms of traditional industrial categories: whoever produces has control, and whoever buys is dependent.

This logic remains valid for many critical technologies. However, it falls short when applied to artificial intelligence. A model can be designed in one country, trained in another, hosted elsewhere, adapted locally, combined with other models, connected to proprietary data, and replaced in a matter of weeks. Its weights may be open-source while its computing infrastructure remains foreign; its data may be sovereign while its architecture is not; its deployment may be local while the bulk of the economic value flows back to an external provider. In such a system, asking whether an AI is “sovereign” or “non-sovereign” is almost too simplistic a question.

Academic literature on digital sovereignty had already identified this challenge. Daniel Mügge shows that sovereignty in AI encompasses different—and sometimes contradictory—objectives, ranging from industrial autonomy to security, political control, and the protection of citizens. Other studies on European digital sovereignty also emphasize that it has no single definition and must be understood in terms of the concrete interdependencies between infrastructure, markets, technologies, and institutions (Mügge, 2024; Adler-Nissen & Eggeling, 2024; Blancato, 2024).

Recent developments in European policy are, in fact, revealing. In 2026, the Commission’s work on digital sovereignty placed less emphasis on the idea of replicating everything within Europe and more on the ability to reduce critical dependencies while remaining open and connected. In particular, it highlighted interoperability, openness, resilience, and the ability to regain control when a dependency becomes problematic (European Commission, 2026).

The Mistral case illustrates why this shift in framework has become necessary. A Chinese model can be hosted on European infrastructure, used with French data, and integrated into the systems of a local company, yet still remain the product of artificial intelligence designed and trained elsewhere. Depending on the criteria used, this situation can therefore appear to be both sovereign and dependent. It is precisely this apparent contradiction that we must overcome.

Perhaps AI sovereignty should no longer be thought of as a state that one possesses or loses, but rather as a set of capabilities that one masters to a greater or lesser extent.

The crucial question then becomes much more challenging: no longer “Is this AI French, European, American, or Chinese?” but rather, in what areas are we truly capable of choosing, understanding, controlling, replacing, and creating?

It is by starting with this question that we can propose an alternative mapping of sovereignty in artificial intelligence.

I. The Six Domains of Artificial Intelligence

Recent work on digital sovereignty converges on one point: technological autonomy can no longer be reduced to national ownership of infrastructure or a company. Daniel Mügge thus shows that sovereignty in AI encompasses several objectives that may come into conflict, while Huw Roberts emphasizes the need to distinguish between technological control and the political capacity to steer systems in accordance with collective goals. The European Joint Research Center also proposes a multidimensional approach that integrates governance, infrastructure, data, markets, and human capabilities (Mügge, 2024; Roberts, 2024; Di Marco et al., 2025).

Building on this research, but taking into account the specific characteristics of generative models, I propose distinguishing six forms of sovereignty in artificial intelligence. These do not constitute six independent states, but rather six dimensions across which a country, a company, or an institution may have very different degrees of autonomy.

The first is cognitive sovereignty. It concerns the ability to design, train, and evolve the models themselves. Who possesses the scientific expertise needed to create the intelligence we use? Who understands its architecture and functioning well enough not to be entirely dependent on decisions made elsewhere? This is the aspect that most closely resembles our traditional conception of technological sovereignty, but it is only one part of it.

The second is data sovereignty. Models have practical value only when they are combined with data. It is therefore essential to know who owns the data, where it is stored, under what legal basis it is processed, who can reuse it, and whether it is used to improve an external system. A company can certainly use an American or Chinese model while retaining a high degree of control over its data; conversely, it can use a European model while relinquishing a significant portion of that control through its cloud architecture or contractual terms.

The third is computational sovereignty. It is the most tangible and, paradoxically, one of the least visible to the end user. It depends on access to GPUs, data centers, the cloud, networks, semiconductors, and now the energy needed to power these infrastructures. The European Commission specifically identifies computing, data, and infrastructure as central pillars of its current technological sovereignty policy, thereby recognizing that digital autonomy also relies on industrial and physical resources (European Commission, 2026).

The fourth is operational sovereignty. It raises a very concrete question: What can I actually do without asking my supplier for permission? Can I install the model on my infrastructure, modify it, audit it, change its rules, stop using it, or replace it with another one without rebuilding my entire system? This dimension introduces a fundamental distinction between ownership and control. It is possible to have not created a technology and still retain a high degree of autonomy in its use; conversely, it is possible to use a European technology while becoming heavily dependent on its provider.

The fifth is normative and cultural sovereignty. It concerns languages, values, categories, worldviews, prohibitions, and trade-offs embedded in the models. Research on digital sovereignty still devotes relatively little attention to this dimension, even though it becomes central once AI systems no longer merely process data but also participate in the production of texts, knowledge, advice, and decisions. Sovereignty, then, also means being able to understand the norms embodied in a model, discuss them, and, if necessary, modify them.

Finally, the sixth is economic and industrial sovereignty. It raises perhaps the simplest question: where does the value remain? Where are the jobs, intellectual property, revenues, infrastructure, investments, and industrial capacity created by AI? A country can be very protective of its data from a legal standpoint while transferring the bulk of the economic value to a handful of foreign platforms. Conversely, it can integrate technologies from elsewhere while building businesses, skills, and markets around them that are capable of retaining a significant share of that value.

This framework fundamentally changes how the Mistral case is interpreted.

By offering GLM-5.2 on a European infrastructure, Mistral can strengthen its customers’ data sovereignty, their operational sovereignty, and—if its infrastructure truly expands across Europe—part of their computing sovereignty. It can also contribute to a form of economic sovereignty by keeping the integration, hosting, and part of the value created around the model within Europe.

At the same time, using a model designed by Z.ai obviously does not strengthen European cognitive sovereignty over that particular model. Nor does it guarantee normative and cultural sovereignty, since the fundamental choices made during its training and initial alignment were made elsewhere.

The same technology can therefore enhance some forms of sovereignty while diminishing others.

That is why binary judgments are no longer very useful. It is insufficient to say that a system is “sovereign” simply because it is hosted in Europe. It is equally insufficient to say that it is not sovereign simply because its model was developed in China or the United States.

The real question then becomes: In what areas do we want to exercise sovereignty, to what extent, and for what purposes?

A hospital, a bank, a defense agency, an industrial SME, or a university will not all require the same level of autonomy across each of these six dimensions. Sovereignty thus ceases to be a national slogan and becomes a strategic choice that can be evaluated, weighed, and, above all, measured.

This approach also helps us move beyond a false dilemma that often poisons the European debate. We do not necessarily have to choose between producing everything ourselves and buying everything from others. We must determine what we cannot afford to lose control over.

II. Sovereignty Is Not Self-Sufficiency

Recognizing the multiple dimensions of sovereignty leads us to abandon another idea inherited from the industrial age: the notion that a country is truly sovereign only if it produces all the technologies it uses itself.

Such an ambition would be largely illusory today. No European country has sole control over the entire chain, from semiconductors to data centers, and from fundamental models to industry-specific software. Even the United States remains dependent on international value chains for certain critical technologies. Sovereignty, therefore, cannot mean the absence of dependence. Every advanced economy is interdependent.

The problem begins when interdependence becomes irreversible.

This is where a more realistic definition can be proposed: sovereignty may no longer be the ability to produce everything oneself, but rather the ability to never become a prisoner of what one does not produce. This approach is based on three properties.

The first is reversibility: the ability to switch away from a technology, platform, or provider without economic and technical costs making such a switch practically impossible. This issue is already at the heart of the European Data Act, which imposes obligations designed to facilitate switching cloud service providers and reduce the contractual, commercial, and technical barriers responsible for vendor lock-in (European Union, 2023). The European Commission has now explicitly made this a key element of its digital sovereignty policy.

The second is substitutability. Reversibility allows you to walk away; substitutability allows you to replace. For an AI architecture, this means avoiding building an entire system around the specific characteristics of a single model to the point where replacing it would require rebuilding applications, data, processes, and interfaces. Interoperability then becomes an attribute of sovereignty. The European Commission also recognizes this when it explicitly links portability and interoperability to users’ ability to choose their providers and combine multiple cloud services (European Commission, 2026).

The third is plurality. A truly autonomous organization should not necessarily seek a single, fully sovereign provider, but rather avoid allowing any one provider to become irreplaceable. In AI, an architecture can combine multiple models based on their performance, costs, level of confidentiality, or the sensitivity of their uses. A bank could, for example, use a local model for certain critical data, a European model for regulated applications, and an American or Chinese model for other, less sensitive tasks. Sovereignty would then lie less in the purity of origin than in the ability to maintain multiple options.

This approach is beginning to emerge in European public policy. In 2026, the Commission awarded its sovereign cloud contracts to four different providers, explicitly stating that this diversification was intended to strengthen resilience and avoid lock-in with a single provider. Its new Cloud Sovereignty Framework evaluates sovereignty based on 48 criteria across eight dimensions, including legal, operational, technological, data-related, supply chain, and security aspects (European Commission, 2026).

Thus, a less spectacular but undoubtedly more robust conception of sovereignty emerges. It does not promise absolute independence; rather, it fosters the capacity for change.

For a company or a government, the right question would therefore no longer be just: “Who developed this model?” It would also become: Can I run it elsewhere? Can I retrieve my data? Can I audit how it operates? Can I switch providers? Can I replace it with another model without having to rebuild my entire system? And have I retained the necessary expertise to regain control if geopolitical, economic, or regulatory conditions change?

A controlled dependency can remain an interdependence. An irreversible dependency becomes a vulnerability.

It may be around this distinction that credible sovereignty for the 2030s can be built. But this definition itself carries a danger. By insisting that it is enough to be able to choose, host, audit, and replace others’ technologies, Europe could end up becoming extraordinarily sovereign over their use while gradually ceasing to be sovereign over their creation.

III. The Risk of “Empty Sovereignty”

The previous definition has its merits, but it also carries a risk. By placing so much emphasis on reversibility, interoperability, and the ability to choose among multiple providers, Europe could end up believing that simply regulating other countries’ technologies effectively is enough to ensure its sovereignty. It would then become highly skilled at hosting, regulating, certifying, securing, and auditing technologies developed elsewhere.

This scenario is by no means absurd. Part of Europe’s digital policy has already been built around this logic: reducing critical dependencies while dealing with the dominance of foreign providers. Academic research on the cloud highlights precisely this tension. Blancato points out that for several years now, European data sovereignty policy has sought to regain control over a market largely dominated by American hyperscalers, without, however, having European alternatives capable of replacing them on a large scale (Blancato, 2024). Research on Gaia-X reaches a similar conclusion: Europe has gradually shifted its focus from eliminating dependence to managing it through interoperability, standards, and compliance (Adler-Nissen, 2024).

This strategy may be realistic. It becomes problematic if it turns into a long-term goal. Because controlling the rules for accessing a technology is not the same as controlling its trajectory. Being able to host a model locally does not give you the ability to invent its successor. Being able to audit a system does not necessarily give you the scientific expertise to rebuild another one. And being able to switch providers offers protection only as long as there are multiple providers to choose from.

Controlling the walls of the data center does not mean controlling the intelligence that flows through it. The message is important. True sovereignty cannot merely manage dependence; it must retain a credible capacity for creation.

This does not mean that Europe must have a European equivalent of every American or Chinese model. Such a race would likely be ineffective and, at times, impossible. But it must retain sufficient research, engineering, computational capabilities, and industrial expertise to understand the technological frontier, contribute to it, and—if a dependency becomes critical—have a real ability to rebuild.

This distinction is fundamental. There is a difference between choosing to use someone else’s technology because it is better and being forced to use it because we no longer know how to produce it ourselves. That is where the line between interdependence and dependence lies.

In June 2026, the European Commission further strengthened its policy on semiconductors, the cloud, open source, and AI, with the stated goal of developing more comprehensive technological sovereignty and strengthening European alternatives in critical technologies (European Commission, 2026).

The real challenge for Europe, therefore, is not to choose between two extremes—manufacturing everything or integrating everything. It is to determine which technologies we can accept depending on, and which ones we must absolutely retain the ability to understand, innovate, and produce.

Without this capability, sovereignty could become a perfectly regulated, secure, and interoperable shell—but one whose intelligence would continue to be generated elsewhere.

IV. Toward a More Stratified Global AI Order

We often continue to describe the geopolitics of artificial intelligence as a single race: several powers are said to be competing on the same track, with the goal of producing the most powerful model. This view is becoming too simplistic.

The rapid convergence in performance provides an initial indication of this. In March 2026, Stanford observed several U.S. companies clustered at the top of the rankings, with Alibaba and DeepSeek close behind, while the gap between the top U.S. and Chinese models had narrowed considerably. Stanford also noted that the competition is gradually shifting toward other criteria, including cost, reliability, and specialized performance (Stanford HAI, 2026).

If this trend continues, the world of AI may come to resemble less a pyramid dominated by a single model and more a stratified system in which multiple forms of power coexist.

The United States currently holds an exceptional advantage in proprietary frontier models, software platforms, cloud computing, capital, and ecosystems capable of rapidly transforming an innovation into a global product. China is developing a different combination: cutting-edge research, industrial power, and a proliferation of effective models, many of which are distributed as open-source software and can be deployed outside of Chinese infrastructure. These categories are obviously not absolute: the United States also produces open-source software, and China develops proprietary systems. Nevertheless, they help us understand different strategic directions. Stanford confirms both U.S. leadership in the production of notable models and China’s significant rise in cutting-edge research and models (Stanford HAI, 2026).

Faced with these two powers, Europe would have little reason to choose between imitation and renunciation.

Another approach is possible: becoming a space where artificial intelligence is controllable, interchangeable, auditable, and integrable into even the most sensitive environments.

This proposal is not about turning an industrial weakness into a regulatory strength. Europe will not become an AI powerhouse simply because it has the AI Act or because it knows how to certify models produced elsewhere. It should combine this regulatory capacity with infrastructure, its own models, open technologies, computing power, and an industrial ecosystem robust enough to ensure that the control it claims is technically feasible.

Europe’s strategic choice could then be phrased differently. It would not be a matter of becoming a third version of the United States or China, but rather of building an ecosystem in which a bank, a hospital, a government agency, an industrial company, or a critical infrastructure could combine multiple models while retaining control over its data, computing power, rules, and ability to switch technologies.

Of course, this is still only a strategic possibility, not a guaranteed advantage for Europe.

But it could offer a way out of a competition defined exclusively by others. In a world where the best models are becoming more numerous and increasingly comparable in performance, the advantage could also lie with whoever ensures that these intelligences can be combined without creating lock-in, used without compromising the most sensitive data, and replaced without losing the entire system built around them.

Europe will not succeed by becoming a belated copy of Silicon Valley or Shenzhen. It can still seek to define a different kind of power: one that allows for the use of multiple forms of intelligence without relinquishing control over what is entrusted to them.

This possibility becomes all the more plausible if a hypothesis—which remains uncertain at this point—is confirmed: that the model itself will gradually cease to be the sole economic center of gravity for artificial intelligence.

V. What the History of Computing Teaches Us: Value Often Ends Up Shifting

At the dawn of a technological revolution, we tend to confuse the technology that makes disruption possible with the source of sustainable value creation. The history of computing should make us cautious.

When personal computers became mainstream in the 1980s, attention naturally focused on the machine itself. However, the decisive transformation of the industry would gradually come from its modularization. Computers became assemblies of components that could evolve independently: processors, operating systems, memory, peripherals, and applications. The work of Baldwin and Clark has shown how this modular architecture fostered innovation by allowing different players to specialize in specific building blocks while building on common standards (Baldwin & Clark, 2000).

This development profoundly reshaped the structure of the industry. IBM had helped establish the PC, but a considerable share of economic power subsequently became concentrated among those who controlled certain layers that had become strategic, notably Intel for microprocessors and Microsoft for operating systems. Bresnahan and Greenstein have shown that competition in the computer industry gradually shifted from being primarily between complete computers to being between technology platforms capable of attracting developers, manufacturers, and users (Bresnahan & Greenstein, 1999).

The Internet has undergone a similar shift.

The protocols that enable the network to function have, of course, remained of fundamental importance. However, the bulk of the economic value created by the Internet has not been concentrated among those who invented TCP/IP. It has gradually shifted toward web browsers, search engines, e-commerce, platforms, the cloud, social media, and the countless services built on top of this infrastructure. Shane Greenstein has shown how the commercialization of the Internet relied precisely on this ability of entrepreneurs and companies to build new uses on top of a network whose fundamental building blocks were already widely shared (Greenstein, 2015).

Research on platforms has since expanded on this insight: in the digital industries, power does not necessarily lie solely with those who own a technology, but also with those who succeed in becoming the foundation upon which others innovate (Gawer & Cusumano, 2014).

I am willing to bet that artificial intelligence will undergo a similar evolution.

Large-scale models will, of course, remain strategic. They will continue to account for a significant portion of research, investment, and technological power. But they will gradually cease to be the sole focal point of value.

The first signs of this shift are already visible. The Stanford AI Index 2026 notes a growing convergence among several leading models and shows that competition is gradually shifting toward other dimensions: cost, reliability, specialized performance, and adaptability to specific use cases (Stanford HAI, 2026).

As the differences between models narrow, companies will place less importance on whether they use GPT, Claude, Gemini, Qwen, Mistral, or any other single model. They will focus above all on determining which system enables them to sell more effectively, produce more efficiently, provide better care, make better decisions, learn more effectively, or serve their customers better.

Value will then shift toward the layers that transform general capabilities into concrete advantages: proprietary data to which competitors do not have access, the business knowledge accumulated within the organization, agents capable of performing complex tasks, workflows that link AI to real-world processes, security and control mechanisms, interfaces, traceability, and, above all, the company’s ability to integrate these technologies into its day-to-day operations.

This trend could even be accelerated by the widespread adoption of multi-model architectures. In the future, an agent will be able to automatically choose one model for reasoning, another for coding, a third for processing an image, and a local model for handling sensitive data. In such an environment, the end user will sometimes no longer even know which model produced each step of the response. Models will remain indispensable, but they will gradually become a component of a larger system.

That is why I do not believe that the artificial intelligence industry will, in the long run, boil down to a competition between a few major foundational models. Just as computer science was not limited to the processor and the Internet was not limited to its protocols, AI will not be limited to the model.

The next battle will hinge on the ability to assemble these building blocks to create systems that are useful, reliable, and deeply integrated into business operations. This is where an increasing share of economic value—but also a vital part of sovereignty—will lie.

The history of computing thus reminds us to avoid one mistake: confusing the most spectacular component of a technological revolution with the entire economy it will create.

The processor has always remained strategic because software has become essential. Internet protocols have not ceased to be valuable because Google, Amazon, and social media platforms have been built on top of them. Similarly, foundation models could remain a critical technology while becoming just one layer among many in AI’s value creation. This hypothesis profoundly changes the European debate.

If the future of AI also hinges on data, personnel, business software, security, integration, and work organization, then Europe must obviously not give up on developing models. But conversely, it would be wrong to assume that its technological destiny depends exclusively on its ability to have the number-one model in a global ranking.

The history of technology shows that power rarely lies with those who master a single layer. It more often lies with those who understand early on how the different layers will come together to create the applications of tomorrow.

General Conclusion. Sovereignty will be a capability, not a border

We started with a French paradox. Mistral, which in just a few years had become a symbol of our determination not to leave artificial intelligence in the hands of major American platforms, has now chosen to incorporate models developed in China into its sovereign offering.

At first glance, the story might seem like one of giving up.

In fact, it tells us something much more important: sovereignty in artificial intelligence is changing in nature.

We had envisioned a relatively simple world. The United States would produce cutting-edge technologies; Europe would have to develop its own to regain its autonomy. China has turned this picture upside down. It is no longer content merely to close the technological gap. It produces architectures, high-performance models, and open technologies—or “open weight” technologies—that are now circulating throughout the rest of the world and that even European players are beginning to adopt.

So the question is no longer whether to choose between American AI and European AI. The world that is taking shape will be much more intertwined.

Our companies will use American, Chinese, and European models. Some applications will run in the cloud, while others will run locally. Open models will coexist with proprietary systems. Agents will be able to call upon multiple intelligence systems in succession to accomplish a single task. Data will sometimes remain within the company even though the model was designed on the other side of the world. And, most likely, users will be increasingly unaware of which specific model is driving each decision made by their system.

In this world, continuing to measure sovereignty solely based on the national origin of the model will no longer make much sense.

That is why I have proposed looking at it from a different perspective, through six dimensions: cognitive, data, computational, operational, normative, and cultural, economic, and industrial. This framework leads to a conclusion that is less comforting than the usual slogans: we may be highly sovereign in some dimensions and deeply dependent in others.

The political question then becomes one of trade-offs. What must we absolutely be able to produce? What can we buy? Which dependencies can we accept? Which ones become dangerous? And above all, what must we preserve today so that we will still be able to choose tomorrow?

Because not all dependencies are problematic. Our economies have always thrived on trade, specialization, and interdependence. The danger arises when we lose the ability to switch technologies, understand what we’re using, switch from one supplier to another, or rebuild a capability that has become critical on our own.

That is why sovereignty in the 2030s will be measured less by our ability to own everything than by our ability to never become captives.

But this new approach must not become an excuse for us to abandon our industrial ambitions. Europe cannot be content with being the continent that regulates American models, hosts Chinese models, secures data, and certifies technologies invented by others. We can draw on intellectual talent from around the world, but we must continue to produce enough of our own to remain capable of understanding the cutting edge, influencing it, and, when necessary, building our own alternatives.

The history of computing also teaches us that the battle will not stop at models. Just as the processor did not define the computing economy and protocols did not define the Internet, large models will not define the AI economy. Value will also shift toward data, agents, software, domain expertise, security, interfaces, and organizations’ ability to transform available intelligence into useful action.

Europe therefore still has a strong hand, provided it doesn't repeat yesterday's mistakes.

It will not succeed by trying to become a belated copy of the United States or China. Instead, it can build its own form of power, based on research that remains at the cutting edge, infrastructure it controls, European models when they are strategic, openness when it reduces dependencies, a diversity of suppliers, and an industrial capacity to integrate AI into our businesses and institutions.

True sovereignty, then, will consist neither in closing technological borders nor in dreaming of an impossible digital self-sufficiency. It will consist in remaining powerful enough so that openness remains a choice and never becomes an obligation.

Perhaps that is, after all, what the Mistral paradox has taught us. Sovereignty does not lie in knowing where all the intelligence we use comes from. It lies in knowing what we entrust to it, what we control, and what we are still capable of building without it.

Perhaps the challenge for Europe is no longer to develop artificial intelligence that is independent of anyone. Rather, it is to build a Europe capable of harnessing everyone’s intelligence without ever abandoning its own.