Artificial intelligence is no longer one industry among others. It is becoming the cognitive layer of the world economy: the layer that organizes information, automates intellectual work, accelerates scientific research, drives infrastructure and, before long, takes part directly in industrial, military and political decisions.
The competition now under way is therefore not only about software. It is about the future distribution of power.
Two blocs have already set their doctrine. The United States mobilizes its capital, its platforms, its universities and its entrepreneurial capacity. China coordinates its infrastructure, its industrial base, its laboratories and its state apparatus around a single objective: technological sovereignty.
Europe is only beginning to convert its regulatory power into industrial policy. It keeps major scientific, industrial and institutional assets. But it carries a deficit of capital, of compute, of global platforms and, above all, of execution speed.
The question is no longer whether Europe will be able to govern artificial intelligence. It is whether it will still own a meaningful share of the infrastructure that produces it.
A note on the register. I write this one in the first person, as the founder of an operating firm that builds and runs AI systems for regulated companies on both sides of the Atlantic. It is an argument, not a briefing. What follows is my read of where the contest stands in July 2026, and what a European answer would actually require.
This is no longer a digital revolution
The vocabulary of digital transformation has become too small for what is happening.
A digital transformation equips an organization with software. Artificial intelligence changes the nature of software itself. The program no longer only executes a pre-written instruction. It interprets, reasons, generates, plans and acts with a growing degree of autonomy.
That shift makes AI a general-purpose technology, comparable in its possible effects to the steam engine, to electricity, to the internet. It will not necessarily replace the infrastructure that came before it, but it will raise its efficiency, change its architecture and redistribute value between those who master it and those who merely consume it.
The Stanford AI Index 2026 describes a simultaneous acceleration of technical capability, investment and economic adoption. It also notes that the institutions charged with measuring, governing and securing this evolution are advancing more slowly than the technology itself (Stanford HAI, 2026).
The real rupture sits there. Intelligence is gradually becoming an industrializable infrastructure.
Frontier models, autonomous agents, data centers, advanced semiconductors and cloud platforms now form a single chain of cognitive production. Whoever controls that chain will be able to sell not only tools, but a capacity for analysis, optimization and decision to the entire world economy.
Sam Altman, Demis Hassabis, Jensen Huang, Dario Amodei, Mark Zuckerberg and Elon Musk are therefore no longer building only technology companies. They are building the cognitive infrastructure that will decide part of the productivity, the security and the scientific power of the coming decades.
The world map of AI is forming around two centers of gravity
This is not a fight between companies. It is a fight between complete ecosystems.
The United States holds the strongest financing machine
The United States assembles almost every critical layer of the value chain:
- OpenAI, Anthropic, Google DeepMind, Meta, xAI and several leading labs for the models;
- NVIDIA, AMD and a deep semiconductor-design ecosystem for compute;
- Microsoft, Amazon, Google and Oracle for the cloud;
- Palantir and other specialists for operational deployment;
- Stanford, MIT, Carnegie Mellon, Berkeley and the large private labs for talent;
- Silicon Valley, the US capital markets and the venture funds for financing.
This concentration produces a compounding effect. Capital funds compute. Compute attracts researchers. Researchers improve the models. The models create products. The products generate revenue and data that fund the next generation.
In 2025, US private investment in artificial intelligence reached 285.9 billion dollars, more than twenty-three times the amount recorded in China. The Stanford AI Index notes, however, that private figures probably understate the Chinese effort, given the weight of public funding and state-guided funds. The United States also counted 1,953 newly funded AI companies over the year (Stanford HAI, 2026).
This lead does not mean the United States will hold an absolute technical advantage forever. It means it holds, today, the sturdiest capacity to finance research, infrastructure and commercialization at the same time.
The main American advantage is not any single model. It is the machine that keeps producing new ones.
China is building an integrated sovereignty
The Chinese strategy follows a different logic.
As early as 2017, the national plan titled New Generation Artificial Intelligence Development Plan set the goal of making China a major world center of AI innovation by 2030. That plan organized fundamental research, industrial applications, training, infrastructure and support for national champions as one program (DigiChina, State Council of the PRC, 2017).
Alibaba, Tencent, Baidu, Huawei, ByteDance, Moonshot AI, Zhipu AI and DeepSeek operate in an environment where industrial policy, university research, compute infrastructure and sovereignty objectives are tightly articulated.
The results are starting to move the global balance. According to the Stanford AI Index 2026, US and Chinese models traded the top position several times through 2025. By March 2026 the measured gap between the best US model and its main Chinese competitor had narrowed to 2.7 points on the composite index used. The United States still produces more top-tier models, while China now leads in the volume of publications, citations, patents and industrial-robot installations (Stanford HAI, 2026).
The arrival of open or low-cost Chinese models reveals something else that matters. The competition is not only about peak performance. It is about the cost of inference, the openness of the models, their ability to be hosted locally and their speed of diffusion.
China is trying less to reproduce Silicon Valley than to build a technology chain it cannot be forced to abandon by a foreign decision.
Europe does not lack intelligence. It lacks concentration.
Describing Europe as a technological desert would be factually false.
The continent holds excellent universities, world-recognized researchers, mathematicians, engineers, advanced industrial companies and several credible AI players. Mistral AI, Aleph Alpha, Helsing, Black Forest Labs, DeepL, Synthesia and many specialists show that a European capacity exists.
Europe also keeps major positions in the sectors where AI will meet the physical world: aerospace, automotive, energy, pharmaceuticals, defense, industrial robotics, luxury, engineering and precision manufacturing.
ASML holds a singular position in the global semiconductor industry through its extreme-ultraviolet lithography equipment. That position is a strategic asset, even if it is not enough on its own to create a complete European chain for advanced compute.
The European problem is therefore not a shortage of skill.
It is the chronic inability to aggregate that skill fast enough around companies able to reach global scale.
The United States is a relatively integrated financial, commercial and technological market. China can mobilize national resources around industrial priorities set by the state. The European Union has to coordinate twenty-seven member states, several capital markets, heterogeneous energy policies, distinct administrations and industrial interests that sometimes contradict each other.
That fragmentation slows the passage from scientific discovery to global company.
Europe knows how to produce researchers. It struggles more to keep the companies they found once those companies need several billion euros, massive access to compute and an immediately unified market.
Dependence begins beneath the models
Public debate tends to fixate on ChatGPT, Claude, Gemini, Grok, Le Chat or DeepSeek. But the model the user sees is only the top of the architecture.
European dependence sits at several layers.
- It depends on processors designed mainly in the United States and manufactured largely in Asia.
- It depends on advanced semiconductor fabrication located outside its territory.
- It depends on cloud platforms that are mostly American.
- It depends on frontier models developed mainly in the United States and China.
- It depends, finally, on software ecosystems, libraries, developer tools and marketplaces controlled by foreign companies.
In 2024, European providers held only about 15 percent of the European cloud infrastructure market, down from 29 percent in 2017. Over the same period, Amazon, Microsoft and Google captured most of the growth. Worldwide, those three companies represented 63 percent of spending on cloud infrastructure services in the third quarter of 2025 (Synergy Research Group, 2025).
This dependence is not theoretical. The cloud is the environment where data is stored, models are trained, agents are deployed and industrial processes are orchestrated.
A company can own its software and still depend on the infrastructure that runs it.
A state can write its rules and still depend on the companies those rules apply to.
European regulation is not the problem. Its isolation is.
The debate over the AI Act is too often reduced to a caricature: safety against innovation.
A powerful technology has to be governed. Systems used in health, hiring, justice, security, finance or critical infrastructure cannot be deployed without duties of traceability, resilience and accountability.
Europe is right to treat certain uses of artificial intelligence as high-risk.
The AI Act, in force since August 1, 2024, sets a graduated framework by nature of risk. Some provisions, notably those on prohibited practices and AI literacy, already apply. Other obligations phase in over time. The calendar has been debated and adjusted, partly because the technical standards needed to apply it arrived late (European Commission, Shaping Europe's digital future).
The strategic problem appears when the norm runs far ahead of industrial capacity.
Regulation can be an advantage when it builds trust, sets exportable standards and raises product quality. The GDPR shaped many international data-protection frameworks that way.
But a norm does not guarantee that the economic value attached to it will be created on the territory that wrote it.
Europe can perfectly well impose transparency duties on American and Chinese models while transferring to those same players its cloud spending, its software subscriptions, its operational data and part of its productivity gains.
It would then have earned the right to set the terms of use for an infrastructure it does not own.
So the relevant question is not: should we scrap regulation?
It is: how do we make sure every new obligation comes with the industrial, financial and technological capacity that lets European companies stay competitive?
Security without power produces dependence. Power without security produces instability. A strategic policy has to build both at once. Franck Ohrel · Hikari Blue
The real danger is regulatory asymmetry
Compliance costs do not fall on every company the same way.
A global platform can spread its legal, technical and documentary spending across hundreds of millions of users. A European startup has to fund it before it has even found its market.
A large company keeps lawyers, cybersecurity specialists, data-governance teams and compliance officers. A small company or a young one often has to choose between those functions and hiring more engineers.
A rule that is identical in law can therefore produce deeply unequal effects in economics.
The risk is that Europe unintentionally protects the players already on top by raising the fixed cost of entering the market.
This does not mean that all regulation reduces innovation. A clear framework can lower uncertainty, ease investment and speed adoption in sensitive sectors.
But that outcome depends entirely on execution: simple procedures, harmonization across member states, standards that are actually available, real support for smaller companies, access to regulatory sandboxes, and no contradictory overlap between the AI Act, the GDPR, the Data Act, the Digital Services Act, NIS2, the Cyber Resilience Act and sector rules.
Europe is not only threatened by an excess of norms.
It is threatened by their uncoordinated accumulation.
The geopolitical risk can no longer be dismissed
For a long time, European companies treated access to American cloud, semiconductors and software as an ordinary commercial relationship.
That assumption is becoming fragile.
US restrictions on the export of advanced semiconductors show that compute is now treated as a national-security asset. Governments can limit the sale of processors, impose licenses, restrict certain technology transfers or control access to fabrication chains.
As models gain cyber, scientific or military capability, it is reasonable to expect that they too could be subject to diffusion controls.
Such a move would not necessarily target Europe as a whole. It could concern specific models, sectors, countries, users or capabilities.
But a strategic vulnerability is not measured only by the probability of a total cut-off. It is measured by the possibility that a foreign decision abruptly changes the cost, the terms of access or the capabilities available to a national industry.
The hypothesis of a future restriction on certain models remains prospective. It is not an established fact.
It is, however, consistent with the recent turn of semiconductors, cloud and artificial intelligence into instruments of foreign policy.
Sovereignty does not mean that every country must rebuild the whole technology chain alone.
It means that no essential economic function should depend on a single political point of failure.
European companies run carrying extra weight
European leaders already face a tighter environment than many of their competitors.
- They have to fund their own transition to artificial intelligence.
- They have to reach data that is usable and legally exploitable.
- They have to hire from a global talent market.
- They have to absorb energy and administrative costs that are often higher.
- They have to integrate several regulatory layers at once.
- They have to compete with American firms that sit on deeper capital markets, and with Chinese firms that benefit from more direct industrial coordination.
The image of the marathon runner carrying a heavier pack is accurate, but incomplete.
The problem is not only the load. It is that the other runners are already building their own highways.
Artificial intelligence amplifies gaps in productivity. A company that has performant models earlier, data properly organized and agents integrated into its operations learns faster than its competitors. It cuts costs, improves its products, captures more data and reinvests those gains.
The lag then becomes cumulative.
An initial six-month difference can turn into a structural gap once systems learn from real usage, teams reorganize how they work, and productivity gains fund the next generation of investment.
Europe has begun to react
It would be unfair to claim that the European Union now only regulates.
In February 2025, the Commission announced the InvestAI initiative, with the objective of mobilizing 200 billion euros of investment, of which 20 billion is meant to fund up to five AI gigafactories. The European program also includes 19 AI Factories intended to give startups, small companies, researchers and industrial players access to compute and specialized services (European Commission, InvestAI and AI Continent Action Plan).
In January 2026, the Council of the European Union opened the legal path for developing gigafactories within EuroHPC. A new European program was launched to support the training of a frontier model (Council of the EU, 2026).
These initiatives are necessary.
They show that Brussels has understood that artificial intelligence cannot be treated only as a regulatory object.
But a funding announcement is not yet an industrial capacity.
Everything will depend on how fast the funds are actually committed, the infrastructure connected to competitive energy, the processors obtained, the researchers hired and the companies allowed to experiment.
In AI, timing is itself a strategic factor.
Infrastructure available in 2030 does not necessarily offset dependence accumulated between 2026 and 2029.
The historical precedent is less reassuring than it looks
Europe has already lived through several technology waves it helped invent without capturing their industrial value.
The Web was created at CERN, but the main internet platforms became American.
Europe once held a dominant position in mobile telecommunications, then lost much of its influence to the makers of smartphones, operating systems and application platforms in the United States and Asia.
It owns leading automotive industries, yet now has to face the car turning into a software platform, the Chinese advance in electric vehicles and American dominance over certain digital layers.
The pattern recurs: scientific excellence, industrial fragmentation, underfunded scaling, then dependence on the platforms that managed to impose themselves.
AI would make this cycle more dangerous.
The internet was an infrastructure of communication. Artificial intelligence is becoming an infrastructure of decision.
Depending on a foreign search engine is an economic weakness.
Depending on foreign systems to optimize medicine, defense, energy, public administration, research and industrial production is a sovereign vulnerability.
The European choice is not about building a "European ChatGPT"
The strategic answer is not to fund a national imitation of every American product.
The contest has several layers.
Europe has to keep a credible capacity in general-purpose frontier models, if only to understand how they work, train its researchers and avoid absolute dependence.
But it can also concentrate its resources on the domains where its industrial base gives it a specific advantage:
- artificial intelligence for industry;
- robotics and autonomous systems;
- health, pharmaceuticals and life sciences;
- energy and critical infrastructure;
- defense and cybersecurity;
- regulated financial services;
- engineering, materials and precision manufacturing.
The next phase of artificial intelligence will not play out only in conversational interfaces. It will play out in factories, laboratories, power grids, hospitals, aircraft, vehicles and defense systems.
Europe holds precisely part of these physical assets and sector competencies.
Its potential advantage is not a late reproduction of Silicon Valley.
It is the connection of artificial intelligence to the real industrial world.
A European doctrine of power
A credible answer would need at least six structural decisions.
1. Treat compute as critical infrastructure
Europe has to secure massive, durable access to advanced processors, data centers, electricity and the networks that keep them running. The AI Factories are a first step. Their governance should favor real usage, speed of access and the needs of companies over a mainly administrative logic.
2. Build a real European market for technology capital
An AI company cannot be financed with the instruments of a traditional small business. Frontier models, cloud infrastructure and industrial platforms need several billion euros before they reach stable profitability. Europe has to complete the capital-markets union, ease institutional investment in technology, and create the conditions that let companies stay European after they reach scale.
3. Reserve part of public procurement for strategic capabilities
European states are, together, a considerable customer. Public procurement could serve as a first market for European players in defense, health, education, public administration, cybersecurity and infrastructure. The United States long used its federal agencies and defense budgets to support technologies that later became commercial. Europe cannot claim to build sovereignty while systematically buying abroad the very systems its own companies need in order to grow.
4. Radically simplify regulatory execution
The goal is not blind deregulation. It is unification. A company that meets a validated European standard should be able to deploy its system across the single market without restarting twenty-seven separate procedures. The obligations that apply to smaller companies must be proportionate to their resources and to the real risk created.
5. Organize the circulation of industrial data
Europe holds considerable volumes of data from health, industry, energy, mobility and research. But that data stays fragmented, incompatible or legally hard to use. Data sovereignty is not about locking data away. It is about allowing its use in secured, traceable, governed environments. Data that is protected but unusable creates neither innovation nor sovereignty.
6. Rebuild a culture of risk
The deepest question is cultural. Europe often penalizes entrepreneurial failure more heavily than institutional inertia. Yet major technology ruptures necessarily produce errors, lost investments and companies that disappear. An innovation policy that demands every result be known in advance does not fund innovation. It funds conformity to the past.
Three trajectories to 2035
The scenarios below are reasoned estimates, not certain forecasts.
Scenario 1 · The industrial recovery
Europe deploys its gigafactories quickly, consolidates its capital markets, uses public procurement, simplifies its regulatory framework and concentrates resources on industrial and scientific AI.
It does not necessarily dominate the global general-purpose models, but it gains a credible autonomy in several strategic sectors. It stays dependent on certain foreign components while holding fallback options and European players able to negotiate at global scale.
Estimated probability: 30 percent. This path is possible. It assumes a break in the speed of decision and in coordination between member states.
Scenario 2 · Prosperous dependence
Europe adopts American, and in some cases Chinese, artificial intelligence at scale.
Its companies gain productivity and some industrial leaders keep global positions. The continent stays wealthy, regulated and relatively stable.
But most of the value created by the models, the cloud and the infrastructure flows back to foreign platforms. European companies become integrators, distributors or advanced users of technologies they do not control.
Estimated probability: 50 percent. This is, today, the most plausible path. It would not produce an immediate collapse, but a gradual erosion of economic autonomy.
Scenario 3 · Strategic decoupling downward
The gaps in compute, capital and productivity keep widening.
The most promising European companies are acquired or move their decision centers. Talent follows the infrastructure and the pay. Technology restrictions or regulatory conflicts reduce access to the most advanced systems.
Europe keeps a legal and commercial power but gradually loses the ability to define the technologies its economy rests on.
Estimated probability: 20 percent. This scenario is less likely than organized dependence. But its consequences would be severe enough to be treated as a major strategic risk.
Sovereignty is not autarky
The word sovereignty is often misunderstood.
It does not mean producing every chip, every model and every line of code inside European borders.
No modern economic bloc is fully self-sufficient.
Sovereignty is a capacity for choice.
It means being able to change supplier, host certain systems locally, audit critical technologies, guarantee the continuity of essential services and keep the skills needed to rebuild or adapt a system.
A sovereign company is not an isolated company.
It is a company that cannot be paralyzed by the unilateral decision of a partner.
A sovereign Europe would therefore not cut its ties with the United States. It would seek, on the contrary, a more balanced technological alliance, one where interdependence replaces dependence.
The choice is now political
Europe still holds the science, the industry, the talent, the savings and the market it needs to remain a technological power.
What it lacks is not, mainly, another statement of intent.
What it lacks is an explicit hierarchy of priorities.
Should artificial intelligence be treated as one digital program among others, subject to the ordinary administrative cycles?
Or should it be treated as a strategic infrastructure, on the same footing as energy, defense, telecommunications and semiconductors?
From that answer follow the budgets, the timelines, the procurement rules, the competition policy, the organization of capital and the degree of acceptable risk.
Europe does not have to choose between its values and power.
It has to understand that its values will not survive in the world order unless it keeps the material means to defend them.
Conclusion: the continent that regulates, or the continent that builds
Artificial intelligence is no longer a technology file.
It has become a matter of economic, scientific, industrial and military power.
The decisive question is therefore not who will write the most sophisticated rules.
It is who will own the models, the compute centers, the platforms, the data, the skills and the companies on which tomorrow's economies will run.
Europe is right to want an AI that is safe, transparent and compatible with the rule of law.
But it would make a historic error if it treated regulation as a substitute for industrial strategy.
A continent that no longer builds critical technologies ends up regulating the terms of its own dependence. Franck Ohrel · Hikari Blue
Europe keeps a window of action. It has a considerable industrial base, high-level research, abundant savings and a market of nearly 450 million people.
But that window is closing fast.
In technology revolutions, a lag is not measured only in years. It is measured in infrastructure installed, talent relocated, data accumulated, companies financed and habits of use that have become irreversible.
History does not condemn powers for wanting to protect their citizens.
It condemns them when they mistake protection for inertia.
Europe should not give up regulating. It should start building again.
Sources
- Stanford Institute for Human-Centered Artificial Intelligence (2025 and 2026). AI Index Report. Figures on US private AI investment (285.9 billion dollars in 2025, more than twenty-three times China), 1,953 newly funded US AI companies, and the US-China frontier gap narrowing to 2.7 points by March 2026. hai.stanford.edu/ai-index
- Mario Draghi (2024). The Future of European Competitiveness. European Commission. On Europe's investment gap and the coordination deficit in technology. commission.europa.eu/topics/eu-competitiveness/draghi-report_en
- European Commission. AI Continent Action Plan, InvestAI, and the AI Act. InvestAI (February 2025): up to 200 billion euros mobilized, 20 billion for up to five gigafactories, 19 AI Factories. AI Act in force since August 1, 2024. digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
- Council of the European Union (2026). Legal path opened for AI gigafactories within EuroHPC, and support for the training of a frontier model. consilium.europa.eu
- State Council of the People's Republic of China (2017). New Generation Artificial Intelligence Development Plan. Goal of making China a major world center of AI innovation by 2030. Translation via DigiChina. digichina.stanford.edu · full translation, 2017
- Synergy Research Group (2025). Global and European cloud infrastructure market. European providers at about 15 percent of the European market in 2024 (down from 29 percent in 2017); AWS, Microsoft and Google at 63 percent of global cloud infrastructure spending in Q3 2025. srgresearch.com
- Daron Acemoglu and Simon Johnson. Work on technology, power and the distribution of productivity gains.
- Erik Brynjolfsson and associated work on artificial intelligence, productivity and organizational change.
- International Monetary Fund. Work on generative artificial intelligence and the future of work. imf.org/en/Topics/artificial-intelligence
- OECD. Publications on artificial intelligence, productivity and innovation policy. oecd.ai
Franck Ohrel · Founder and CEO, Hikari Blue · Austin, July 2026