The sustainability risks threatening AI’s resilience narrative

This is a thought leadership report issued by S&P Global. This report does not constitute a rating action, neither was it discussed by a rating committee.

The driving force behind markets, corporate strategy and the sustainability landscape in 2026 can be spelled with two letters: AI. The keywords “AI” and “data centers” have soared across US listed companies’ earnings calls and other communications since 2022. The combined market capitalization of just four of the hyperscalers pushing the technology forward now represents roughly 18% of the S&P 500. (On Aug. 1, 2026, the sum of the market capitalizations of Alphabet ($4.35T), Microsoft ($3.45T), Amazon ($2.93T) and Meta ($1.42T) was about $12.15T, or 18% of the combined S&P 500 market capitalization of $69.5T.)

The need for a massive increase in computing power to turn the expectations around AI into reality has pushed spending on data centers to unprecedented heights, with the top five hyperscalers expected to invest nearly $600 billion by year-end 2027, according to 451 Research by S&P Global. Electricity demand from these facilities is reshaping energy markets and appears poised to lock in higher fossil fuel use, particularly gas-powered generation.

Initial AI enthusiasm among investors and business leaders is meeting resistance in 2026, however. In this research, we investigate three growing sources of pressure on the AI adoption narrative, all speaking to the sustainability implications of the technology and the infrastructure it relies on:

• Corporate governance adoption is lagging rapid AI tech development and use.
• AI infrastructure, namely data centers, is increasing pressure on energy systems.
• Communities are challenging the local costs and benefits of data center expansion, primarily in the US.

These pressures arise at different levels of the AI ecosystem: within companies through governance practices, within infrastructure through energy and resource requirements, and within communities through opposition to data center development. Together, they influence whether AI can continue scaling while maintaining trust and access to essential resources. These forces are coinciding at a time when AI-related spending and development shows little sign of stopping. Nevertheless, they highlight real risks that are beginning to test the resilience of AI expansion — or the ability of AI investment and adoption to continue while maintaining corporate accountability, access to reliable infrastructure and public buy-in.

AI governance is a growing concern

Generative AI hit the mainstream this year, with a June 2026 Pew Research Center survey finding that nearly half of American adults use chatbots, including about one-in-four who use them on a daily basis. Furthermore, 38% of employed adults say they use gen AI tools at work. Widespread use is also raising awareness of the technology’s limitations. Large language models have the potential to reinforce biases based on their training data, and AI-generated responses sometimes present false conclusions with a tone of certainty.

These issues can present challenges for the companies moving to enhance their products and solutions with AI and those seeking to unlock worker productivity by using AI tools in-house. Reputational or financial risk is a possibility if a product presents inaccurate AI-generated information, and headlines are full of examples of AI-generated errors or hallucinations. Firms protesting that such errors are the responsibility of the AI tool can still lose face in the public eye.

Cybersecurity and advanced AI models’ disregard for the restrictions their developers place on them are also becoming greater risks. In recent weeks, OpenAI LLC and Anthropic both disclosed that their AI agents breached other companies’ cyber defenses, prompting the US government to propose voluntary tests for measuring advanced AI models’ hacking abilities.

These incidents have turned a spotlight onto the need for safeguards and governance of AI at all levels, including at companies. AI governance is key to adopting the technology ethically and responsibly. It ensures AI use is aligned with societal values and human rights, as exemplified by regulatory frameworks like the EU AI Act adopted in June 2024. For companies, a governance policy is the foundation of responsible AI use.

Corporate adoption of AI governance is rising, but remains weak

To evaluate the state of AI governance at companies, the 2024 and 2025 cycles of the S&P Global Corporate Sustainability Assessment (CSA) asked voluntary questions about whether companies have an AI governance policy in place and what those policies say. From the full CSA universe of about 13,000 companies, 939 responded about AI policies in both years.

The CSA defines an AI policy as a dedicated policy or commitment with the purpose of managing AI-related risks and opportunities and the governance system the company has implemented. It should cover at least one of four main areas: data privacy, cybersecurity, mitigation of potential biases and identification of AI-generated content.

The share of respondents that have an AI policy rose from 40% in 2024 to 47% in 2025. This includes companies with a dedicated standalone policy or one that is integrated into other governance documentation. About 41% of companies in our analysis do not have a policy, and another 13% said they do not currently have a policy but plan to implement one in the next two years. In other words, 54%, or more than half, of responding firms have not put AI governance into practice.

The communication services and information technology sectors, which include many of the hyperscalers and tech firms driving AI development, stand out as having the highest percentage of companies with dedicated policies. Their experience with building AI models gives them direct knowledge of potential risks. Sectors with notable year-over-year improvement include materials, energy and real estate, all of which saw the share of companies without a policy decrease by at least 5 percentage points.

From a geographic perspective, governance is relatively strong in North America and Europe but weaker in Asia-Pacific, where only 36% of CSA respondents have some kind of AI policy. That signals the potential for greater risk embedded in the technology’s use in the region.

However, our data shows that identifying AI-generated content is a clear governance gap: Only 40% of companies across sectors enforce this practice in their policies. The consumer discretionary, real estate and communication services sectors have the lowest share of companies recognizing this risk in their policies, ranging from 30% to 33% of companies. A lack of guardrails around gen AI use may pose reputational risk or lower trust in a company’s products and services if incorrect AI-generated content is presented as true, or if companies produce low-quality AI-generated content. Policies can address this by explicitly requiring authentication, such as by following human-in-the-loop best practices, and by implementing watermarking or disclaimers that allow users or customers to recognize AI-created or -assisted content.

Growing data center energy demands complicate decarbonization plans

As data center demand continues to grow, the sector’s sustainability challenge will depend increasingly on whether operators can align infrastructure expansion with resilient, credible and climate-aware energy strategies. While not the sole driver of data center growth, increasing demand for AI training and inference is contributing to the sector’s rising need for computing capacity and electricity. Scrutiny of the AI boom is extending to whether data center developers can build hyperscale facilities responsibly — particularly given these projects’ massive electricity demands. A single hyperscale facility might require hundreds of megawatts of power with annual consumption on the order of terawatt-hours.

The question of whether AI tools can have a net positive impact on addressing climate change starts with how much electricity they use and where it comes from. Data from the 2025 CSA shows that, for a sample of companies operating or leasing data centers for the last eight years, power consumption has almost tripled. Over the same period, renewable energy use has risen alongside this growth: Companies reported that less than one-quarter of their total energy consumption came from renewables in 2017, soaring to 84% in 2024.

This high share of clean energy usage suggests that many operators have sought to reduce the carbon intensity of their footprint as their consumption soared. However, it remains to be seen how much higher the renewables share can go. Fossil fuel generation, particularly natural gas, is in high demand among developers looking to build on-site power for facilities. Orders for natural gas turbines have hit a 25-year high, driven by demand from data centers, according to S&P Global Energy. Even if the clean energy share of data center power continues to rise, the immense increase in overall electricity consumption means that electricity provided by fossil fuels will grow in absolute terms.

Alongside the source of electricity, another central question is how efficiently that electricity is used. The industry measures efficiency with a metric called power usage effectiveness (PUE), which is the ratio of total facility energy use to the energy consumed by IT equipment. An ideal, theoretical PUE of 1.0 would represent a perfectly efficient data center that uses all its power for computation, with none used for cooling, lighting or other functions. In reality, PUE of 1.1 to 1.2 is considered highly efficient, and modern hyperscale facilities fall in this range.

Average weighted PUE has improved consistently over the past decade, falling from 1.54 in 2017 to 1.23 in 2024 among the sample of CSA companies used in this analysis. In the past, efficiency gains were able to offset demand that was increasing at a relatively slow pace. But now, as data centers approach the physical limit of maximum efficiency and power demand accelerates at an unprecedented pace, future improvements in efficiency cannot offset the additional power the industry expects to draw.

AI advances sustainability performance — sometimes

On the other side of the climate ledger, it is becoming more common for companies to use AI tools to improve some aspects of corporate sustainability. In the 2025 CSA, about half of assessed companies in five sectors said they are using AI to measure or enhance their sustainability efforts.

Further listening:

As a result of pushback across the US, more companies are making commitments to pay for infrastructure upgrades to include electricity, water and roads. However, ongoing local opposition indicates that developers and operators still have significant ground to cover before fully addressing the concerns of residents and local governments.

Looking forward

As AI tool adoption spreads throughout the working and social lives of people around the world, scrutiny of its potential drawbacks is growing. Cybersecurity breaches caused by AI agents or models and examples of reputational damage born from gaps in AI governance are making headlines. Academic studies of AI’s impact on education and cognition, concern about an AI-driven future that automates jobs and questions about the resources data centers use are contributing to a rethink of the technology’s risks.

As these concerns become increasingly mainstream, we expect to see companies devoting more resources to AI governance — and data from the S&P Global CSA suggests there is room for improvement, as most of the companies in our analysis still lack an AI policy. We likewise expect to see stronger scrutiny of how growing data center energy demands impact corporate decarbonization plans, and of the resource constraints AI infrastructure places on the communities where hyperscale data centers operate. Governance, controls and policies around key areas of AI risk will be key to the technology’s resilience going forward.