Enhancing Corporate Governance Through Responsible AI Management

Key insights:

• • • Elevate AI governance to the board — Companies should tie their AI deployment to enterprise risk management with explicit KPIs for energy intensity, water withdrawals and consumption, and supply‑chain human rights.

    • Make transparency a competitive asset — Implement auditable disclosures on AI workload footprints, water stewardship, and supplier traceability, and then link executive compensation and vendor contracts to measurable efficiency and resiliency outcomes.

    • Demand transparency despite practical challenges — Although demanding transparency from suppliers may not be practical now due to current challenges, collectively asking for detailed information sends a notable requirement to AI infrastructure providers that the company is seeking to drive change and preserve trust in an AI-driven economy.

AI now sits at the center of corporate sustainability governance as it supercharges data gathering, analytics, and reporting. Indeed, there is a clear upside for AI is areas of energy optimization, emissions monitoring, land‑use assessment, and climate scenario analysis.

At the same time, AI’s rise is colliding with sharply growing electricity and water demands from data centers and concerns over geopolitically exposed supply chains. The governance challenge for companies therefore is to manage risk at this intersection. This means treating AI as a capital‑intensive, cross‑border infrastructure program whose environmental footprint and supply dependencies must be actively governed.

Why electricity and water are now board‑level AI risks

AI has turned electricity and water from background utilities into constraints that should be dealt with on the board level. Indeed, AI magnifies water risk across cooling, power generation, and chip manufacturing. This makes sourcing and efficiency choices strategic imperatives for many organizations.

Electricity demand — AI use and the data centers that power the tools already account for a significant and rising share of electricity use in the United States. The latest national assessment finds roughly 4.4% of US electricity consumption now goes to data centers, a figure poised to grow as AI workloads scale. Forward‑looking projections from the U.S. Department of Energy indicate that by 2028 a significant portion of data center electricity could be attributed to AI workloads.

If you translate those projections into household‑equivalent consumption, you can get an idea of the potential magnitude of the problem. Together, these sources suggest that the fastest‑growing part of AI’s energy appetite is not just for training models, but the steady, pervasive inference capabilities required to power AI features in everyday products and operations.

Direct and indirect water use — Data centers powering AI also negatively impact local water footprints. It shows up in three places: i) data‑center cooling; ii) the electricity feeding those facilities, including thermoelectric and hydroelectric generation; and iii) AI’s own hardware supply chain. In regions already facing scarcity, these demands compound local stress. For example, the average per capita water withdrawal is 132 gallons per day; yet a large data center consumes water equivalent to that of 4,200 persons.

This makes data centers one of the top 10 of water-consuming industrial or commercial industries in the country, which incidentally is home to more than 5,300 data centers and counting. At the end of 2021, around 20% of these centers were drawing water from moderately to highly stressed watersheds in the western US. This is a common situation in countries all over the world as well.

Geopolitical exposure — The hardware that powers AI includes advanced logic and memory chips, which depend on concentrated manufacturing nodes and supply chains with access to critical minerals. Extraction and processing of inputs, such as lithium and cobalt, are often clustered in jurisdictions with elevated levels of human‑rights, environmental, or geopolitical risk. This potential amplifies exposure to export controls, sanctions, or resource nationalism, especially directly for companies’ supply chains and indirectly for those companies using AI.

Companies need to ensure their communication on legal and policy issues are pointing in the same direction in regard to these concerns. Indeed, companies need to deepen value‑chain due diligence while navigating evolving supply‑chain and AI‑specific regulatory regimes.

Recommended actions for companies

These intersections have clear implications for corporate governance. AI’s promise to accelerate decarbonization, improve transparency, and strengthen decision‑making will be realized only if leaders can properly manage the physical, political, and social realities underpinning the technology. Recommended actions to manage risk in areas in which AI and geopolitics converge include:

Demand transparency in electricity and water consumption of AI infrastructure — Companies building AI infrastructure need to conduct AI workload planning. Companies using AI can demand transparency of their suppliers’ 24- to 36-month forecast of training and inference by region with overlays in grid carbon and local water stress to better understand their indirect environmental impacts.

De‑risk impact by incentivizing clarity in supply chains — Companies using AI can begin asking AI infrastructure companies to provide due diligence in tier 2, 3, and 4 suppliers, all the way down to smelters, refiners, and miners to make sure that companies are not indirectly contributing to environmental and social harms.

The bottom line

While these recommendations generally align with evolving corporate practices in sustainability and risk management, the challenge of implementation will vary based on the company’s size, influence over suppliers, and existing governance structures. The most challenging aspect will likely be achieving transparency and clarity in supply chains, which requires cooperation from suppliers and the investment of potentially significant resources.

At the same time, however, if more companies collectively ask for this level of detailed information from their AI infrastructure providers, it will send a notable demand signal. Indeed, AI is both a sustainability tool and a sustainability liability, but its benefits will be realized only if leaders confront the physical and geopolitical constraints that make AI possible.

Those companies that begin asking for this level of transparency can preserve the trust that underwrites their license to navigate successfully in an AI‑driven economy.