Europe Has Two Years to Avoid Becoming an AI “Vassal State,” Warns Mistral CEO

In practical terms, that means European AI developers might depend on foreign providers for computing capacity, pricing, supply access, and even policy decisions such as export controls or platform restrictions.

Why the Next Two Years Matter

Mensch’s timeline is tied to how quickly AI infrastructure is expanding globally. Massive investments are already being made in computing clusters, semiconductor supply chains, and energy-intensive data centres. Once these systems are built and supply contracts are locked in, they tend to create long‑term market dominance.

His argument is that the next two years represent the decisive window for Europe to build sufficient infrastructure before the global supply of compute becomes effectively locked up by existing hyperscale players. If that happens, late entrants could struggle to access the resources needed to train and operate frontier AI systems.

AI as Infrastructure: “Turning Electrons Into Tokens”

A key part of Mensch’s framing is that AI should be viewed as an industrial infrastructure technology rather than just software innovation.

He describes the process as converting “electrons into tokens” — meaning that electricity powers computing hardware that generates the outputs of AI models. In this view, success depends less on clever algorithms alone and more on access to physical resources: energy, chips, and massive computing facilities.

This shift in perspective places AI alongside sectors such as energy, telecommunications, and transportation infrastructure.

Why Chips, Energy, and Data Centres Are Critical

The infrastructure behind AI involves several tightly linked components.

Chips

Advanced AI models require specialized processors such as GPUs or AI accelerators. Control over the supply and availability of these chips determines who can train the largest and most capable models. If European firms rely mainly on external supply chains or foreign cloud providers, their ability to scale AI systems could be constrained.

Energy

Running large AI clusters requires enormous electricity capacity. Training and operating frontier models demands reliable, affordable power at scale. Mensch’s “electrons into tokens” metaphor reflects the idea that energy availability is becoming a competitive factor in AI development.

Data Centres

Chips and electricity must be combined inside high‑capacity data centres designed for AI workloads. These facilities host the computing clusters used to train and run models. To expand Europe’s independent capacity, companies like Mistral have begun investing directly in infrastructure — including a reported €1.2 billion ($1.4 billion) data‑centre project in Sweden aimed at strengthening local compute capabilities.

The Role of Digital Sovereignty

The broader debate surrounding Mensch’s comments is about digital sovereignty — the ability of a region to control its own digital infrastructure and technological capabilities.

For AI, sovereignty means more than building local startups. It also requires control over the underlying stack: computing infrastructure, energy supply, cloud platforms, and access to data.

Without those elements, European companies could still build applications, but the foundational layers of AI would remain controlled elsewhere.

Regulation Alone Won’t Be Enough

Europe has been active in shaping global AI regulation, including comprehensive rules around safety and governance. But Mensch’s argument is that regulation by itself cannot create technological independence.

Rules can shape markets, but they do not build GPUs, power grids, or computing clusters. To compete with the scale of infrastructure being deployed elsewhere, Europe also needs substantial investment in compute capacity, energy systems, and data‑centre construction.

Building an Independent European AI Ecosystem

The long‑term vision described by Mensch involves creating a full AI ecosystem in Europe that includes:

• Local AI model developers and research talent
• Access to high‑performance chips and compute clusters
• Large-scale data centres connected to abundant energy
• Investment capital and public procurement to support adoption

Only by combining these elements, he argues, can Europe avoid becoming merely a consumer of AI technologies built and operated elsewhere.

The next few years will likely determine whether Europe develops that full stack — or relies on infrastructure controlled by global tech giants.