Behind the most publicized high-tech trends, concrete constraints are reshaping the landscape: strengthened European regulations, tension on the energy supply for data centers, and a growing gap between the promises of artificial intelligence models and their actual operating costs.
Data Centers and AI: When Electricity Hinders Innovation
Infrastructure projects dedicated to artificial intelligence face an obstacle that product presentations rarely mention. Reliable access to electricity now determines where data centers are located, more than fiber connectivity or proximity to markets.
The consumption of data centers dedicated to AI has been rising sharply since 2025. At the same time, the energy consumed per AI task is decreasing rapidly due to hardware and software optimizations. These two trends coexist, complicating any linear projections about future demand.
Electrical grid connection delays are becoming a geographical selection factor for companies. In France, as elsewhere in Europe, the waiting lists to obtain a high-power connection are growing. A data center project may be technically ready but remain stalled for several months due to a lack of available network capacity.
Cloud and AI players who regularly publish their advancements on the Next Generation site document this on-the-ground reality, where energy constraints weigh as heavily as computing power in investment decisions.
Energy Cost per AI Task: Disparities Hidden by Benchmarks
Not all uses of artificial intelligence require resources in the same way. A simple text query consumes a fraction of the energy needed for video generation or multi-step reasoning.
The most resource-intensive tasks (video generation, complex reasoning, orchestration of multi-agent systems) can mobilize significantly higher resources per query. Improvements in efficiency per task do not necessarily offset the multiplication of intensive uses.
This disparity raises a question rarely addressed in AI model comparisons: the actual energy cost of a service depends less on the underlying technology than on the type of task performed. Companies deploying large-scale generative AI solutions are discovering energy bills that are difficult to anticipate, as standard benchmarks do not reflect the diversity of production uses.
Cybersecurity in Europe: The Cyber Resilience Act Changes the Rules of the Game
European regulations regarding cybersecurity for digital products are entering a phase of concrete application. The Cyber Resilience Act (CRA) imposes new obligations on manufacturers of connected products marketed in the European Union.
Three points deserve the attention of tech professionals:
- The obligations do not only concern new products. Devices already marketed in the EU are also targeted, which implies retroactive compliance updates for some manufacturers.
- Companies established outside the European Union are not exempt. As soon as a digital product is accessible in the European market, its manufacturer must comply with the regulatory framework, regardless of its country of origin.
- Vulnerability reporting deadlines are tightening, with obligations to notify the competent authorities within shorter timeframes than before.
For tech companies, regulatory compliance is becoming a budget line comparable to R&D for certain segments of connected products. It directly conditions access to the European market.
AI Act and Implementation Timeline: What Announcements Don’t Specify
The European AI Act has received extensive media coverage, often focused on the most visible prohibitions (social scoring, facial recognition in public spaces). The actual implementation timeline is more nuanced.
Not all obligations of the AI Act come into effect simultaneously. The rollout follows a staggered timeline, with different dates depending on the risk category of the AI system concerned. Systems classified as high-risk face later deadlines than the general prohibitions, which apply first.
This gradual implementation creates a zone of uncertainty for AI model developers and integrators. Some companies are delaying their deployments in Europe while waiting for clarifications, while others are anticipating the strictest obligations to avoid costly redesigns.
What Is Still Missing in the Regulatory Framework
Detailed guidelines for assessing the risk level of an AI system are still being developed. The precise classification criteria are not all stabilized, leaving room for interpretation for both companies and regulatory authorities. For French SMEs integrating AI into their products, this situation makes regulatory planning particularly complex.
Cloud Technologies and Data Sovereignty in France
Hybrid and sovereign cloud remains a structuring trend for French companies. The issue is no longer just about performance or cost, but about the location and control of data.
The question of data sovereignty takes on an additional dimension with AI models, which require substantial volumes of training and computing data. Hosting on national territory and governance compliant with European law are becoming selection criteria for an increasing number of organizations.
Adoption remains gradual. Large companies and administrations are moving faster than SMEs, for whom the additional cost of a sovereign cloud compared to standard offerings is a measurable barrier. The trend is real, but a massive shift to sovereign cloud has not yet occurred in France.
The tech trends of 2026 are reflected as much in regulatory and energy constraints as in product announcements. Access to electricity, the actual cost per AI task, the Cyber Resilience Act, and the timeline of the AI Act outline a landscape where innovation is negotiated with very concrete limits.
