Countries approach artificial intelligence differently. Europe primarily views artificial intelligence as a regulatory challenge. China treats it as an industrial transition. The United States treats it as a matter of capital deployment.
That difference matters.
China’s recent inclusion of domestic artificial intelligence chips in public-sector procurement is often read through the lens of the US–China technology rivalry. That reading is incomplete. The deeper issue is not only chips, export controls, or national champions. It is how states use infrastructure, demand, and procurement to shape long-term strategic capacity.
The key question is not whether Europe should copy China’s model. It should not.
The more useful question is whether Europe can build comparable mechanisms for coordinated demand, infrastructure learning, and strategic optionality.
Europe’s AI weakness is not fragmentation itself. It is fragmentation without coordinated demand.
Procurement as an Expression of Power
Public procurement is often treated as a compliance-heavy purchasing function: risk-averse, procedural, and focused on cost control.
China’s approach shows that procurement can also become a coordination mechanism.
Through the Xinchuang initiative, Beijing has used public-sector demand to accelerate domestic technology adoption across hardware, software, and infrastructure. By adding domestic artificial intelligence processors from companies such as Huawei and Cambricon to procurement frameworks, the state turns government demand into industrial momentum.
The key insight is not what China purchases.
It is how procurement is used to shape the market.
Public agencies and state-owned enterprises create guaranteed demand. That demand gives domestic providers usage, feedback loops, integration pressure, and time to improve. It bypasses part of the uncertainty that normally slows infrastructure transitions.
This does not mean Europe should imitate China’s political model. It should not. Europe operates under different legal, institutional, and industrial constraints.
But Europe should not ignore the mechanism.
Europe has purchasing power. The problem is that this demand is fragmented across institutions, member states, procurement rules, compliance regimes, and budget cycles.
When demand signals are incoherent, ecosystems struggle to scale.
Europe pays at scale, but often decides in isolation.
This fragmentation has direct consequences for artificial intelligence infrastructure.
Industrial systems need predictable demand. Suppliers need credible adoption signals. Developers need stable platforms. Operators need confidence that the ecosystem will still exist after the first procurement cycle.
When each institution optimizes separately, the market receives noise instead of direction.
Europe’s weakness is therefore not spending power. It is the absence of coordinated demand.
Accepting short-term inefficiency can preserve long-term control
European organizations are trained to optimize for efficiency: cost per unit, energy consumption, short-term return on investment, operational convenience, and vendor maturity.
That logic is rational in stable conditions.
It becomes dangerous during strategic technology transitions.
China’s approach to domestic artificial intelligence chips shows a different logic. Beijing appears willing to tolerate a temporary performance penalty in exchange for long-term control, ecosystem learning, and reduced dependency.
The point is not that inefficiency should be subsidized indefinitely. The point is that efficiency is not always the highest-order objective.
In early infrastructure transitions, optimizing too soon can lock organizations into dependencies that later become expensive to reverse.
Cheap compute can become the most expensive dependency if it removes future optionality.
Cheap compute can become the most expensive dependency.
For Europe, the lesson is not to subsidize inefficiency blindly.
Europe faces higher energy costs, political constraints on subsidies, and a more fragmented institutional environment. It cannot absorb inefficiency in the same way China can. But it can learn from the mechanism.
The strategic question is whether organizations are willing to budget for controlled inefficiency in order to build capability.
Organizations should distinguish between production-grade efficiency and strategic experimentation. The former optimizes current operations. The latter preserves future options.
Both are necessary. The risk begins when short-term efficiency is allowed to eliminate long-term optionality.
Artificial intelligence workloads scale non-linearly. Training and inference costs increase with data gravity, tooling lock-in, talent specialization, and operational integration. Once these forces are embedded, reversing course becomes much harder, even if alternatives later become technically attractive.
Once that happens, the cost of transition is no longer only financial. It is embedded in architecture, tooling, workflows, and skills.
The Real Lock-In Comes from Software Dependencies
Hardware dependency is visible. Software dependency is more durable.
This is where many artificial intelligence sovereignty debates remain too shallow. Chip production matters, but chips alone do not determine strategic autonomy.
The deeper lock-in often comes from software: toolchains, framework bindings, proprietary extensions, MLOps pipelines, observability stacks, deployment patterns, internal workflows, and developer habits.
Artificial intelligence lock-in happens long before the chip is installed.
A processor can be purchased. A data center can be funded. A cloud contract can be renegotiated.
But a production artificial intelligence workload is not just a model running on hardware. It is an operational stack. It depends on runtime environments, drivers, frameworks, custom kernels, monitoring, automation, data pipelines, security controls, incident procedures, and the skills of the teams maintaining it.
That dependency becomes harder to reverse every time engineers build around a dominant ecosystem.
From an operational standpoint, artificial intelligence dependency accumulates across layers: runtimes and drivers optimized for dominant hardware platforms, framework bindings and custom kernels tuned to specific architectures, MLOps pipelines designed around proprietary APIs, observability and monitoring stacks embedded into platform workflows, and talent specialization shaped by what engineers use every day.
By the time an organization realizes it depends on a platform, the dependency is already embedded in code, workflows, hiring profiles, and operational routines.
For European enterprises, this risk is amplified by fragmentation. Unlike hyperscalers, most organizations do not control the full stack. They inherit dependencies through cloud services, managed platforms, third-party tools, vendor defaults, and procurement decisions made under time pressure.
What looks convenient today becomes an exit cost tomorrow.
The Chinese example is useful as a stress test, not as a model to copy. When organizations are forced to change architectures, the bottleneck is not always procurement, energy, or capital. It is often software portability.
Europe faces the same bottleneck; it has simply not been exposed under real scale constraints yet.
Most boards closely track cloud spending. Few can measure how many months of architectural inertia they have already accumulated.
A useful metric is the Time-to-Port Ratio: the time required to migrate a production artificial intelligence workload to an alternative stack. A low ratio indicates architectural flexibility. A high ratio signals strategic lock-in.
Europe’s Real Asymmetry
Europe’s challenge in artificial intelligence is often misdiagnosed as a gap in innovation, funding, or ambition.
The deeper issue is structural asymmetry.
Europe is neither China nor the United States. Trying to copy either model would fail for political and operational reasons.
China can absorb inefficiency through state-directed demand and long-term planning. The United States can rely on capital abundance, hyperscaler scale, and venture-market depth. Europe operates under neither condition.
It cannot out-subsidize China.
It cannot out-scale US hyperscalers.
But Europe does have a different potential advantage: the ability to coordinate across sovereign systems without full centralization.
That is a harder model to execute. It is also the only realistic one.
Europe’s weakness is not fragmentation itself. It is uncoordinated fragmentation.
Public procurement, energy policy, digital regulation, industrial incentives, cloud dependency, and artificial intelligence strategy are often managed as separate domains, by different institutions, on different timelines.
The result is strategic drift.
Fragmentation without coordination is inertia.
The competitive unit is no longer only the company or the nation-state. It is the coordinated system.
For executives, this changes the question. The question is not simply which artificial intelligence platform to select. It is how quickly the organization can adapt if assumptions change.
Can workloads be moved?
Can models be ported?
Can suppliers be replaced?
Can procurement be coordinated?
Can the organization absorb temporary inefficiency to preserve long-term control?
This managed optionality is the advantage Europe can still develop.
Operationally, that means recognizing that some inefficiencies, overlaps, and redundancies can function as insurance. It means using regulation not only to constrain supply, but also to shape demand. It means treating software portability and skills diversity as strategic assets.
Europe does not need to win the AI race by dominating compute.
It needs to avoid losing the race by locking itself into irreversible paths.
Festina Lente
Artificial intelligence sovereignty rarely fails because technology is unavailable.
It fails because dependency remains invisible until it becomes irreversible.
China has chosen to absorb inefficiency in order to gain time, experience, and control. The United States offsets dependency through scale, capital depth, and market dominance. Europe, by contrast, often optimizes for short-term efficiency while assuming that strategic optionality will remain available later.
That assumption is no longer defensible.
For European organizations, the central question is not which artificial intelligence platform will win. It is how quickly they can disengage from a dominant platform when economic, regulatory, or geopolitical conditions change.
In this context, sovereignty is not only a political objective.
It is an operational property.
It can be designed.
It can be tested.
It can be measured.
Organizations that remain competitive in the artificial intelligence era will not necessarily be those that bet early on the right technology. They will be those that maintain the ability to adapt without disruption.
Optionality, portability, and coordination are now executive responsibilities.
In artificial intelligence, resilience is fundamentally architectural.