Every week, Cleo News gives you 5 minutes to turn the latest research papers into action.

This week, we’re stepping out of code and prompts. AI isn’t just about algorithms; it’s also about servers, energy, and cooling. Two realities collide in 2025. On one hand, French regulations demand transparency and efficiency from data centers. On the other, AI is pushing traditional infrastructures to their limits and demanding a complete overhaul. For businesses, the message is clear: integrating AI also means rethinking your infrastructure. 

This week, we decode how the unseen becomes strategic.

1. DDADUE Law: New energy performance requirements for data centers (Gide Real Estate, May 2025 — 10 min)

France is imposing a new regulatory framework on data centers. Since May 2025, any data center over 500 kW must transmit its energy data to a European platform. Centers over 1 MW will have to utilize their waste heat starting October 2025. This residual heat holds enormous potential: 1 MW recovered can heat 500 homes. Penalties go up to 50,000 euros per site for non-compliance. A cost-benefit analysis is mandatory for every new project or significant modification. The directive aims to transpose the European directive on energy efficiency. Banks, companies, and research centers must now follow the same rules as telecom operators. The regulations come into force gradually, with application decrees expected in the coming months.

Key takeaway: French regulations impose transparency and energy utilization on data centers with real penalties to enforce efficiency.

2. AI reshapes the datacenter landscape (CBRE - 20 min)

AI is radically transforming the global digital infrastructure. France is investing 109 billion euros in AI, with 55% allocated to physical infrastructure. Hyperscalers (AWS, Microsoft, Google) account for 70% of infrastructure spending. A GPU server costs 3 to 4 million dollars and consumes 100 to 300 kW, compared to 5 to 10 kW for a standard server. Air cooling is reaching its limits. Google and Meta aim for 80% liquid cooling by 2027. Data centers could increase from 1.5% to 4% of global electricity consumption by 2030. The trend: companies are bringing their AI back to private infrastructures for security and sovereignty. Edge computing is finally developing in 2025, driven by the need for local inference. The market is consolidating with a 20% increase in mergers and acquisitions. AI is returning to data centers not as a workload, but as a tool for energy optimization.

Key takeaway: AI demands a technical and energy revolution in data centers, with a shift towards hybrid and sovereign infrastructures.

Key points for you – summary of both studies

Infrastructure becomes strategic. AI doesn’t deploy in some abstract cloud; it requires massive, energy-intensive, and costly physical data centers.

Sovereignty is gaining ground. After 20 years of migrating to the American cloud, companies are bringing back their critical AIs.

Energy efficiency is no longer optional. Between legal requirements and skyrocketing costs, optimizing is vital.