Artificial intelligence: sustainability reporting

Artificial intelligence (AI) needs a large amount of computing power, and each unit of computing power consumes significant amounts of electricity. This computing largely takes place in data centres which are rapidly being constructed around the world to keep up with the growth in AI. In addition to the energy needed for computing AI queries, data centres also consume power and water to run cooling and ventilation systems. AI is thereby increasing demand for scarce natural resources, and also putting pressure on electricity grids, which are still in-part dependent on fossil fuel plants to meet rising needs. The result: increased carbon emissions and concern about the long-term environmental impact of AI.

In response, there has been an increase in regulations requiring the reporting of energy and resource use by AI. This article compares the US, UK and EU reporting regimes, highlighting the key regulations and their implications.

UK reporting requirements

The UK's primary regulation that governs the recording and reporting of energy usage by AI is the Streamlined Energy and Carbon Reporting framework (SECR) ^[1]. Introduced in April 2019, under SECR "large" companies must report their energy use, greenhouse gas emissions, and energy efficiency actions in their annual reports. A "large" company or LLP is one that meets two of the following requirements:

• It has at least 250 employees
• It has an annual turnover of more than £36 million, or
• It has annual balance sheet of more than £18 million.

Many companies offering AI services (or providing the data centres in which they are hosted) are likely to meet these requirements.

Companies caught under SECR are required to disclose the following information:

• "Scope 1 and 2" greenhouse gas emissions, being (1) emissions directly generated by the company and (2) indirect emissions from purchased energy
• Reporting "Scope 3" emissions (being indirect emissions generated along a company's supply chain) is voluntary but strongly encouraged, and many organisations do disclose this information
• At least one emissions intensity ratio. These ratios compare emissions data with a relevant business metric or financial indicator, such as emissions per sales revenue or square metre of floor space
• The underlying global energy use for the reporting year
• Figures for energy use and greenhouse gas emissions from the previous year
• The methodology used for reporting as outlined in the SECR guidelines, and
• Actions taken to improve energy efficiency, including a narrative description of the main measures implemented during the relevant financial year.

In their annual report, companies must explain the actual and potential impacts of climate-related risks and opportunities on their businesses, strategy, and financial planning. This involves considering different climate scenarios and how they might affect the company's operations and financial performance. The risk management section requires companies to disclose how they identify, assess, and manage climate-related risks, integrating these processes into their overall risk management framework.

Larger providers of AI services may also need to comply with the Companies (Strategic Report) (Climate-related Financial Disclosure) Regulations 2022 (CRFD Regulations). These regulations apply to companies with more than 500 employees and are either (i) a listed company or (ii) have a turnover of more than £500m.

Under the CRFD regulations, companies must disclose information aligned with the recommendations of the Task Force on Climate-related Financial Disclosures (TCFD). This includes details on governance, strategy, risk management, and metrics and targets related to climate change. Companies are required to describe the board's oversight of climate-related risks and opportunities, as well as management's role in assessing and managing these risks.

Both the SECR and CRFD regimes apply to companies as a whole and in any sector, rather than specifically to AI services or individual data centres, but given the public scrutiny about AI's environmental impact the reporting under these regimes by AI providers will likely be more heavily scrutinised.

EU reporting requirements

The EU has gone further, with reporting requirements specifically for data centres. Given that nearly all AI services are hosted within data centres, the EU regime will indirectly, but certainly, impact AI companies.

The EU's approach to energy usage reporting in data centres is primarily governed by the Energy Efficiency Directive (EED) ^[2]. The EED requires data centres to report their energy performance and sustainability metrics to a central European database. Data centres were first required to comply with these requirements by 15 September 2024, and then provide an annual report by 15 May of each following year. The EED applies to data centres that are sited in the EU with an installed IT power demand of at least 500 K, which will apply to nearly all AI-focused data centres. It applies to both own-use data centres (for a company's own systems or "as-a-service" solutions provided to customers) and to data centres where space is provided for a customer's systems (co-location).

Data centres must report various key performance metrics, including energy consumption, power utilisation, temperature set points, waste heat utilisation, water usage, and the use of renewable energy. They must also report the compute capacity of the data centre so that the EU can calculate the energy efficiency.

The EU plans to publish aggregated data at both the Member State and EU levels, promoting transparency and accountability, but data on individual sites will be kept confidential.

There is currently no UK regime equivalent to the EED that specifically targets data centres in the UK, but there is growing pressure for the UK to institute its own data centre reporting so that energy providers and water companies can properly assess future demands, and ensure that the demands of data centres are balanced against other priorities. 

United States AI energy usage reporting requirements

While the United States does not yet have a unified, mandatory national/international framework equivalent to the UK’s and EU's regimes, several federal and state-level initiatives are emerging to address the sustainability of AI infrastructure – although there are contradictions of approach at different levels of government.

1. Congress: Artificial Intelligence Environmental Impacts Act of 2024

The most significant federal development is the proposed Artificial Intelligence Environmental Impacts Act of 2024 (S.3732) (AI Environmental Impacts Act). The AI Environmental Impacts Act was introduced into the U.S. Senate on February 1, 2024, and was referred to the Committee on Commerce, Science, and Transportation. The Act has not passed into law, but its introduction signals growing federal interest in the sustainability of AI technologies.

More specifically, the AI Environmental Impacts Act aims to establish a reporting system for entities involved in AI development and deployment. Importantly, however, the Act and its “requirements” are voluntary. Under this Act, the U.S. Environmental Protection Agency (EPA), the U.S. Department of Energy, and the National Institute of Standards and Technology (NIST) are tasked with convening a consortium of stakeholders, comprised of industry, academia, and government, to evaluate potential impacts and recommended best practices. The consortium would then create guidelines for reporting energy consumption, water usage, pollution, and electronic waste associated with the full lifecycle of AI models and hardware. NIST would also develop a voluntary framework for companies to report the environmental impacts of their AI systems.

Although voluntary, this framework lays the groundwork for future mandatory disclosures and reflects increasing federal attention to AI’s environmental footprint. The proposed Act also encourages transparency and accountability in how AI systems affect the environment, which will undoubtedly be questions asked by investors or pursued in diligence in the context of a transaction.

2. Governmental: EPA Guidance and Executive Orders

In contrast, in January 2025, the EPA issued clarifying guidance under the National Emission Standards for Hazardous Air Pollutants (NESHAP) supporting the use of Reciprocating Internal Combustion Engines (RICE). This guidance allows certain fossil-fuel powered backup engines at data centres to operate up to 50 hours per year in non-emergency conditions to support grid reliability. While not a reporting mandate, this interpretation acknowledges the critical role of data centres in AI infrastructure and signals regulatory flexibility to use fossil fuels to ensure uninterrupted power supply. This development has also been welcomed by AI and data centre developers who have been heavily involved/invested in acquiring RICE generation units to support data centre development.

In a similar vein, Executive Order 14318, signed by President Donald Trump on July 23, 2025, is aimed at accelerating federal permitting and reducing regulatory hurdles for the construction of data centres that support AI and includes no environmental reporting requirements. The order is part of a broader America's AI Action Plan and focuses on things like: encouraging “qualifying projects”, streamlining environmental reviews, efficient permitting, access to development on federal lands. It also includes prioritising dispatchable baseload energy sources, meaning the use of fossil-fuel and other non-renewable energy.

4. State-level legislation

Several U.S. states have however gone in the opposite direction and begun enacting AI governance laws that indirectly touch on energy usage through broader environmental and consumer protection frameworks. By way of a few examples:

• In October 2023, California passed two bills comprising its Climate Accountability Package: SB 253 (Climate Corporate Data Accountability Act) and SB 261 (Climate-Related Financial Risk Act). SB 253 requires large public and private companies above certain revenue thresholds and that “do business” in California to report greenhouse gas Scope 1, Scope 2, and Scope 3 emissions. SB 261 requires companies to publish climate-related financial risk reports on publicly accessible websites. These bills have been passed into law, and the first reporting deadline is under SB 261, on January 1, 2026. While California’s Climate Accountability Package is not targeted at data centre operations, certain data centres will certainly be captured within the scope of reporting.
• California's Climate Leadership and Community Protection Act (CLCPA) sets ambitious statewide goals for greenhouse gas reductions, including a 40% reduction by 2030 and an 85% reduction by 2050. It also requires 70% of energy production to come from renewable sources by 2030 and zero emissions for electricity by 2040. These provisions directly impact data centres by encouraging renewable energy adoption and reducing reliance on fossil fuels.
• New York has introduced two bills that mirror the two California Climate Accountability bills, one of which would require reporting of Scope 1, Scope 2, and Scope 3 emissions, and one that would require reporting of climate-related financial risks. Both bills are currently under review in the New York Senate Environmental Conservation Committee as of mid-2025, and not yet passed into law.
• New York has proposed/introduced a swath of bills (e.g., Senate Bill S6394A) aimed at regulating energy consumption by data centres; requiring annual disclosure reporting; prohibiting incentives in fossil fuel power purchase agreements with utilities; and directing the public service commission to establish a data centre surcharge and discount plan. While this bill has not been passed, it serves as a good indication of state legislation seeking to address data centre resource demands and infrastructure costs.
• Oregon’s POWER Act (HB 3546) provisions regarding data centre utility costs and consumption align with broader state-level environmental and utility regulations. The POWER Act appears to address the taxation and assessment of data centres and their associated utility costs, as reflected in Oregon Revised Statutes §§ 308.516 and 308.519. These statutes provide specific guidelines for the assessment of data centre property and utility-related assets, including electricity generation and consumption.

U.S. regulatory landscape: conclusion

In the current U.S. political landscape, characterised by a somewhat adversarial structure, competing legislative efforts regarding data centre development have emerged. On one hand, states like California are enacting laws that promote renewable energy development, directly impacting data centres by pushing for increased sustainability and environmental responsibility. These state-level initiatives align with broader goals of reducing carbon footprints and enhancing energy efficiency. On the other hand, federal actions, such as the EPA guidance and the President’s Executive Order 14318, may seem to discourage reliance on renewable energy sources by prioritising expedited infrastructure development without explicitly emphasising renewable energy integration. This dichotomy reflects a broader national debate on how best to balance economic growth and technological advancement with environmental stewardship, leading to a complex regulatory environment for AI providers to navigate.

Conclusion

With no unified global approach, an AI provider could be subject to multiple reporting regimes depending on the location of their business, customers, and underlying data centre infrastructure. The UK and EU are generally aligned in moving towards mandatory environmental reporting rules whilst the US approach is fragmented with widely different approaches being adopted across federal and state levels. The US landscape currently appears much like the UK and European one from nearly a decade ago. That began with fragmented voluntary reporting (under the TCFD) which was only adopted by some businesses, and then only later crystallised into mandatory reporting at a company level, and now in the EU at an individual data centre level. 

Even without legal regulation, more AI providers are expected to voluntarily report their energy usage and emissions. Where a US-based AI is providing an international service involving the cross-border processing of data of UK and EU citizens, the AI provider may need to provide this information to satisfy international customer expectations as this becomes the norm.

There may be other commercial benefits too. Understanding the full supply chain energy usage can help improve energy efficiency which allows AI providers to cut operational costs and boost profits. Energy / emissions reporting might enhance a business's reputation and appeal to environmentally conscious customers. Transparent reporting also opens up investor opportunities, making businesses more attractive to those investors focused on sustainability. Additionally, institutional lenders sometimes favour companies with clear, measurable improvements in energy and utility usage, and this could grant access to ring-fenced sustainable finance funds.

Looking forward, AI providers will need to closely monitor the development of sustainability reporting regulations and standard market practice to ensure that they remain both legally compliant and commercially competitive.

Our team

Womble Bond Dickinson's transatlantic team of digital lawyers understands how to deliver artificial intelligence (AI) services across a global footprint. This article is one in a series comparing the US, UK and EU legal regimes around AI – find the other articles on our dedicated AI Hub.

Sources

[1] The legal basis for SECR can be found in section 10 of The Large and Medium-sized Companies and Groups (Accounts and Reports) Regulations 2008 (as amended by The Companies (Directors’ Report) and Limited Liability Partnerships (Energy and Carbon Report) Regulations 2018).

[2] As an EU Directive, it has to be implemented by local laws in each member state. The exact reporting requirements and method of reporting may vary between states.