Behind the debates over AI models, another battle is being waged: the battle over infrastructure. During this discussion featuring Octave Klaba, Catherine MacGregor, and André Kudelski, the conclusion was clear: AI is no longer just a competition in software, but a competition centered on energy, data centers, and digital sovereignty.
The real shortage is no longer AI, but energy
The figure that stood out in the debate came from Engie:
- Electricity demand from data centers has increased by 17% over the past year.
- Global electricity consumption, on the other hand, increased by only about 3%.
In other words, AI infrastructure is growing nearly six times faster than overall energy demand.
The challenge is no longer just to build high-performance models, but to connect data centers to the power grid quickly enough. According to Catherine MacGregor, the time it takes to bring energy infrastructure online is now becoming a strategic factor just as important as the availability of chips.
Energy Is Becoming a Competitive Advantage
According to Octave Klaba, the situation puts France in a relatively favorable position.
Thanks to its energy mix, which relies heavily on nuclear power, France offers:
- a more stable power supply;
- more predictable costs;
- available connection capacity.
Conversely, several European countries are already facing difficulties in securing new electricity capacity for future data centers.
One of the most striking points concerns Germany. According to Klaba, obtaining a clear answer regarding future connection capacity can already be problematic for some projects.
European digital sovereignty could thus depend more on energy geography than on technological excellence alone.
The European Paradox: Sovereignty or Efficiency?
One of the most interesting points in the debate concerned European fragmentation.
For OVHcloud, sovereignty often involves deploying infrastructure in many countries to ensure that data remains localized.
Consequence:
- more small data centers;
- fewer economies of scale;
- greater operational complexity.
André Kudelski was particularly blunt:
“Either Europe thinks on a European scale, or it will remain less competitive.”
His assessment is grim: AI is advancing at a pace that is incompatible with traditional European decision-making cycles. Each additional layer of fragmentation reduces the continent’s ability to compete with the United States or China.
The Illusion of the “Green” Cloud
The debate also touched on a topic that is rarely discussed: how to accurately measure the carbon footprint of AI.
Octave Klaba criticized the current metrics used by the industry.
According to him, most suppliers focus on:
- operational emissions;
- direct electricity consumption.
But many overlook Scope 3, that is, the footprint associated with manufacturing:
- GPUs;
- servers;
- theses;
- electronic components.
In other words, a significant portion of the environmental cost of AI is currently not reflected in the indicators highlighted by the market.
This issue is likely to quickly become a key concern for large corporations subject to decarbonization targets.
The Next Metric: CO₂ per Token
Catherine MacGregor’s remarks hinted at a major shift.
Today, companies compare models based on:
- their cost;
- their speed;
- their accuracy.
Tomorrow, they could also compare:
- their carbon footprint;
- their emissions per token generated.
Engie already plans to require its suppliers to provide not only cost metrics (“FinOps”), but also environmental metrics (“CO₂ Ops”).
For those in the AI field, this could become a new selection criterion just as important as technical performance.
Sovereignty is not self-sufficiency
On the issue of cybersecurity, the discussion was more pragmatic than ideological.
Neither Engie nor Kudelski advocates an approach that involves systematically replacing U.S. models.
The goal is rather:
- avoid any single dependency;
- maintain multiple suppliers;
- be able to switch quickly from one solution to another.
Catherine MacGregor explained, among other things, that Engie provides access to several models within its internal AI Studio, including those from Mistral AI.
The reasoning is simple: sovereignty does not necessarily mean producing everything yourself, but rather retaining the freedom to choose.
Three Visions for Accelerating Europe
During the final roundtable discussion, the speakers proposed three priorities:
André Kudelski
- Reserve a greater share of public procurement contracts for European entities.
- Reform the way Europe regulates emerging technologies.
Catherine MacGregor
- Moving from 27 national regulators to a much more integrated approach.
- Build a true European market for energy and AI.
Octave Klaba
- Continue to invest heavily in energy efficiency.
- Develop infrastructure that requires less computing power.
What This Roundtable Really Reveals
The debate showed that the question of AI sovereignty is often framed incorrectly.
The problem in Europe is not just a matter of models.
This is a coordination issue between:
- energy;
- cloud;
- cybersecurity;
- infrastructure;
- industry.
The United States has hyperscalers. China has a centralized national strategy.
Europe, for its part, has considerable strengths—low-carbon energy, industry, research, and companies such as OVHcloud and Mistral AI—but continues to function as a collection of national markets.
Perhaps the most revealing statement from the panel came from André Kudelski: “AI moves at the speed of light.”
The question, then, is no longer whether Europe has the necessary technologies. It is whether its infrastructure, regulations, and governance can evolve at the same pace as AI itself.
Cette publication est également disponible en :
