What environmental information must large EU data centers report?

Data centers with at least 500 kW of installed IT power must report information including energy consumption, water use, renewable energy, waste-heat reuse and other performance indicators. The European database publishes information in aggregated form, while the extent of public access to site-level figures is now being challenged.

Why are data center efficiency scores not enough?

Efficiency scores show how effectively a facility uses energy or water, but they do not reveal its total demand. A very efficient, very large campus can still consume more electricity and water than a smaller, less efficient one. Local planning needs both ratios and absolute figures.

What did Amazon promise for data center transparency?

Amazon says it no longer uses nondisclosure agreements with government agencies on its data center projects and will publish energy use, energy efficiency, water use, water efficiency and carbon-free energy information annually. Its October announcement does not clearly specify whether all of those figures will be disclosed site by site.

What should communities ask a data center operator to disclose?

Communities should ask for annual and peak electricity demand, water withdrawal and consumption by source, cooling method, generator emissions and operating hours, required utility upgrades, who pays for them, tax incentives, permanent jobs and the measurement period behind every claim.

Data Center Transparency Is the New AI Environmental Fight

The most important data center news this week is not a new campus or a bigger chip order. It is a fight over disclosure. Europe now collects detailed energy and water information from large data centers, but the public mainly sees aggregated results. In the United States, Amazon has responded to mounting opposition by dropping nondisclosure agreements with government agencies and promising more environmental reporting.

Those stories landed within days of each other. Together, they show where the AI infrastructure debate is heading. Efficiency claims and corporate sustainability totals are no longer enough. Communities want to know what a specific project will require from their grid, water system and public finances before it is approved.

The short version

The next standard for responsible AI infrastructure is site-level disclosure: annual electricity and water use, peak power demand, cooling method, grid-upgrade costs, backup-generator emissions and the basis for any claimed community benefit.

Europe already has the data

Under the EU Energy Efficiency Directive, data centers with at least 500 kilowatts of installed IT power must report information including energy consumption, water use, renewable energy and waste-heat reuse. The European Commission describes transparency as essential and says the EU aims to triple its data center capacity by 2035.

The problem is what happens after the figures are submitted. The Commission's public database presents information at country and EU level, grouped by size and type. The regulation explicitly makes the European database public only in aggregated form. That can show the direction of the sector, but it cannot tell residents how one proposed campus compares with the local water supply or grid.

On September 30, the investigative newsroom Lighthouse Reports announced a filing with the Aarhus Convention Compliance Committee. It argues that the Commission is withholding site-level figures on electricity, water, emissions and waste heat that people need to participate in environmental decisions. The filing is a legal claim, not a ruling. But it identifies the central gap accurately: Europe has created a reporting system without settling how much of the underlying data the public should be able to inspect.

Why an efficiency score is not the same as environmental impact

A data center can become more efficient and still use more electricity and water in total. Power usage effectiveness, or PUE, compares a facility's total energy with the energy delivered to its computing equipment. It is useful for spotting overhead from cooling and power systems. It does not show the absolute size of the load.

The same distinction applies to water-use efficiency. A better ratio can coexist with rising total consumption when a campus adds servers or runs them more intensively. Location matters too. A cubic meter of water in a water-secure region does not carry the same local risk as a cubic meter taken during drought or from a stressed watershed.

Aggregated ratios are valuable for comparing a sector. Site totals are what let a town test whether a particular project fits its resources. Both are needed; one cannot substitute for the other.

Amazon is changing its public posture

On October 2, Amazon announced a new Data Center Commitment and a US community program called Built Together. The company says it will add more than $1 billion over five years for education, job training, household and public-building efficiency upgrades, water projects and priorities chosen through local partners.

The more consequential promise may be the one that costs less: Amazon says it no longer uses nondisclosure agreements with government agencies on its projects. It also says it will publish energy use, energy efficiency, water use, water efficiency and the share of carbon-free energy every year.

These are meaningful commitments, especially if they make negotiations and resource demands visible early. Important details remain open, however. Amazon's announcement does not clearly say whether annual environmental figures will be available for each facility or only as company-wide totals. It also does not define a common format that would let the public compare Amazon sites with projects from other operators.

Community funding cannot replace project data

Funding for lower household energy bills, schools, training and water restoration can produce real benefits. It should be evaluated on its own results. But a community grant is not evidence that a data center's peak electricity demand will be manageable, that its water source is resilient or that ordinary ratepayers are protected from infrastructure costs.

The clean test is whether the benefit and the burden are reported separately. Residents should be able to see what a company funds, what the facility consumes, what the utility must build and who pays for each part. Combining those questions into one broad claim about economic impact makes public scrutiny harder.

Transparency also has to arrive before the decisive permit, tax agreement or utility contract. A detailed sustainability report published after construction does not give a community the same bargaining power as project-level information disclosed during planning.

What useful site-level disclosure would include

A practical public record does not need to reveal customer data or server configurations. It should answer the resource questions that affect people outside the fence:

Annual electricity consumption and expected peak grid demand, with measured and projected figures clearly separated.
Annual water withdrawal and consumption by source, plus peak-day demand and drought restrictions.
Cooling method, use of potable or reclaimed water, and how water use changes with weather.
On-site generation and backup-generator fuel, permitted emissions and actual annual operating hours.
Grid, substation and water-system upgrades required for the project, their cost and who pays.
Waste-heat reuse, renewable-energy claims and whether clean electricity is matched annually or hour by hour.
Tax incentives, community payments, permanent jobs and the reporting period behind each claim.

Operators can add efficiency ratios, replenishment projects and avoided-emissions estimates. Those are useful context, but they should sit beside the absolute local figures rather than replacing them.

The real test comes next

Europe's reporting system could become a global model if it gives the public enough detail to understand individual projects while protecting genuinely sensitive information. The legal challenge will test whether its current balance meets environmental-information rights.

Amazon, meanwhile, has made promises that can be judged against future disclosures. Dropping government NDAs should make local agreements easier to scrutinize. Annual publication becomes much more valuable if it is site-specific, consistent and tied to the places absorbing the demand.

The direction is clear. The AI boom has moved data centers from the background of the internet into the center of public infrastructure policy. Companies that want communities to trust the buildout will need to show the local numbers, not only the global narrative.

Check the evidence behind a local project

Our AI Infrastructure Atlas links project records to their sources and includes a draft request for water-use information.

Open the infrastructure atlas How to use the map