In an op‑ed, artificial intelligence is presented as an engaging strategic decision, whose lasting value is created as close to the data and operations as possible, within a controlled, distributed, and adapted architecture. According to the text, AI can no longer be limited to experimentation if companies want to derive a lasting advantage.
- AI creates lasting value only when it runs close to data and operations.
- A controlled, hybrid, distributed architecture is needed, since a single centralized cloud cannot deliver that advantage.
- Digital sovereignty depends on precise technical choices, such as data location, auditability, inference placement, and reversibility.
- The main obstacle is alignment and governance, not technology, so governance must become a pillar of performance.
The author notes that executives no longer ask whether AI works, but where to deploy it, how to control it, and how to scale it without creating irreversible dependencies. In this view, AI that provides a lasting competitive advantage cannot be centralized in a single cloud, i.e., a set of remotely hosted IT services. Its future would be distributed.
Value would not come from a generic model, “no matter how performant it is,” but from the ability to connect AI to clean data, real business processes, and concrete operational constraints. These assets—sensitive, often regulated, and rarely movable—live as close to operations as possible, where AI must run to be useful.
A distributed architecture
Large centralized computing resources remain essential for training the most complex models, the text acknowledges. For adapting to proprietary data, governance, and operational execution, hybrid platforms would be needed, capable of bringing compute closer to data when latency (response time), security, or resilience become critical.
This logic would become more pronounced with the emergence of agentic AI, which refers to agents capable of acting across complete operational chains, rather than isolated assistants. At this level, the central question is no longer model selection but architectural coherence and governance quality.
Sovereignty: an architecture matter
The author refuses to treat digital sovereignty as an ideological stance or a binary choice between openness and retreat. It would be built through precise technical decisions: control over the location of sensitive data, ability to monitor and audit usage, choice of where inference runs—that is, the generation of a response by a model—and anticipation of reversibility.
A distributed AI would, according to the text, provide this level of control without sacrificing performance or scalability. Europe would have “often underestimated” assets: industrial and regulated sectors where trust is a structural condition, a solid scientific ecosystem and, in France, a largely decarbonized energy base, which would become an advantage for sustainably industrializing AI.
From experimentation to industrialization
The main obstacle to value creation would not be technological, but would stem from alignment. The text points to accumulated experiments lacking clear governance, without an explicit transformation roadmap for processes, and without real convergence among IT, business units, and partners. A platform, even an advanced one, does not on its own transform a business model.
For decision‑makers, the author advises focusing efforts on uses with measurable impact, designing an explicitly governed hybrid and distributed architecture, and relying on partnerships capable of large‑scale deployment. Governance, according to the text, must become “a pillar in its own right of performance.” For Europe, distributed AI would represent an opportunity to build pragmatic sovereignty founded on architecture.
