Sustainability
How to Review Water Risk for an AI Data Center
A site-diligence checklist for water sources, cooling design, permits, consumption, and operating limits.
What the EU AI Act Says About Model Energy Use
The EU AI Act does address model energy use, but the scope is narrower than a blanket rule for every company that deploys a high-risk AI system.
The requirement in the enacted text
Article 53 requires providers of general-purpose AI models to keep technical documentation with the information listed in Annex XI. Annex XI includes the known or estimated energy consumption of the model. If that value is unknown, the provider may estimate it from the computational resources used.
Read the enacted regulation on EUR-Lex. Article 53 and Annex XI provide the controlling text.
What Article 40 does
Article 40 directs standardization work. That work includes reporting and documentation methods for resource performance and energy-efficient model development. Article 40 does not create the per-deployment reporting list that the earlier version of this page described.
Scope matters
Provider, deployer, general-purpose model, and high-risk system are different roles and categories under the Act. Do not copy the Annex XI requirement into a compliance plan until counsel confirms the organization's role, the model category, and any applicable exception.
Application dates
The European Commission states that the obligations for providers of general-purpose AI models began to apply on August 2, 2025. The Commission's enforcement powers apply from August 2, 2026. Providers of covered models placed on the market before August 2, 2025 have until August 2, 2027 for the relevant obligations.
Check the Commission's current guidance for general-purpose AI model providers before relying on those dates.
A useful technical record
An organization that needs to document model energy use should retain the model and version, training boundary, computational resources, measurement or estimation method, result, source data, assumptions, and review date. This record supports review. It does not by itself establish legal compliance.
Legal notice: This page summarizes selected provisions for research. It is not legal advice.
More Insights
AI Architecture
Why AI Analysis Still Needs Explicit Human Approval
How specialist analysis, review records, approval authority, and external action stay separate.
Technical
Reduce Agent Context Cost by Loading Tools on Demand
A method for measuring tool-schema overhead and loading only the instructions a task needs.
Industry
What Tapestry Can and Cannot Tell a Data Center Investor
What the public Tapestry and PJM record establishes, and what site power diligence still requires.
Investment
How to Use Prediction-Market Odds Without Treating Them as Facts
A review method for contract terms, liquidity, timestamps, resolution risk, and source reporting.
AI Architecture
Why Infrastructure Review Uses Specialist Lenses
How GreenCIO routes a question, records its sources, and leaves trade-offs with the human reviewer.
Finance
What LCOE Shows, and What It Leaves Out
How LCOE normalizes project cost, which assumptions drive it, and which risks need separate review.
Sustainability
When Carbon-Aware Scheduling Helps, and When It Does Not
How to review movable workloads, grid data, deadlines, baselines, and reporting before claiming a saving.
Technical
How a News URL Becomes a Feed Record
The bounded extraction, scoring, validation, and storage path behind each submitted source URL.
Investment
How to Review Power Risk in AI Data Center Investments
A source-based framework for power delivery, upgrades, tariffs, cooling, and schedule risk.
Finance
Who Finances Hyperscale Data Centers? A Diligence Guide
Review the contracts, funding layers, and operating dependencies behind a hyperscale project.
Sustainability
A Practical Sustainability Screen for AI Infrastructure
A review checklist for workload energy, power sources, cooling, water, hardware use, and evidence.