The AI Boom Is Becoming an ESG Test

• AI’s electricity and water demands can affect margins, capital spending, and returns.
• ESG analysis should identify resource constraints that change company economics and valuations.
• Portfolios across sectors may share exposure to the same data-center growth and power bottlenecks.

Artificial intelligence is making ESG exposure inside a portfolio more difficult to assess.

AI can improve efficiency, support scientific research, optimize logistics, help energy systems operate more intelligently, and improve access to information. It can also help companies reduce waste, forecast demand, manage buildings and identify climate risks more effectively.

The heavy use of water and electricity by data centers, a foundational part of AI infrastructure, raises a whole set of natural resource considerations.

Electricity and water use are already visible. The harder part for analysts and portfolio managers is understanding what they do to margins, capital spending and returns.

The AI boom does not need another sustainability label. It needs ESG analysis capable of distinguishing between growth that is merely resource-intensive and growth that remains economically attractive after those resources have been paid for.

Many sustainable portfolios have historically reduced exposure to fossil fuels, mining and other visibly resource-intensive sectors while holding relatively large positions in technology and communications. In a pre-AI world, that substitution could look straightforward; it is less straightforward now.

Google’s recent investment in Finland is a good example of why power procurement has moved beyond a sustainability disclosure. The company announced around $15 billion of AI infrastructure investment and signed a 22-year agreement with Fortum covering up to half of the output of one Finnish nuclear plant.

The agreement is environmentally relevant but it is also part of a much more conventional business problem: securing enough reliable power to support future computing capacity.

The following day, Finnish opposition politicians were already raising questions about whether such rapid data-center expansion could tighten national power supply and raise electricity prices. Whether those concerns ultimately prove justified is less important than the broader point: AI infrastructure is now large enough that local energy capacity can become part of the investment case.

For an analyst, the useful part is not simply that the electricity is nuclear or low carbon. It is the visibility that a long-term supply agreement may give the company over power availability and future costs. It also tells investors something about the scale of the infrastructure required to support AI growth.

A grid constraint can become a revenue-timing problem. A long permitting process can delay the return on invested capital. An expensive power market can affect operating margins. A more efficient cooling system can lower water use while also improving the economics of a site.

A data center in a water-abundant region does not face the same operating constraints as one built where industrial demand is already competing with households and agriculture. A company able to reduce freshwater use or adopt a cooling system better suited to local conditions may have more flexibility in choosing future sites.

The question then becomes: Does water availability affect cost, location, permitting or the pace at which capacity can be added?

If the answer is no, the datapoint may not be financially material. If it does, it deserves more attention than a generic ESG score.

If AI demand continues pushing electricity requirements higher, analysts should increasingly focus their analysis on the following:

• What additional investment will be needed simply to maintain those commitments?
• Will clean-power contracts become more expensive?
• Will grid infrastructure have to be financed or supported differently?
• Will new capacity have to move to locations where electricity and water are easier to secure?

A company that secures power earlier may be able to bring new capacity online faster than a competitor; a company with more efficient cooling may have a wider choice of sites; a company that locks in energy supply through long-term agreements may have better visibility over future operating costs.

Google’s environmental reporting shows how quickly the scale of energy procurement is changing. In 2025, the company contracted more than 12 GW of new clean energy, more than it had procured in the previous two years combined. It also said improvements in hardware, software, computing efficiency and clean-energy procurement avoided more than 58 million tons of CO2-equivalent emissions across operations and the supply chain.

Those figures show substantial investment in efficiency and clean energy. They also show how much physical infrastructure is now sitting behind the digital growth story.

If emissions rise because capacity is growing quickly ahead of strong demand, the financial conclusion may be very different from a situation in which emissions rise because infrastructure is becoming less efficient. If emissions targets require steadily more expensive power contracts or carbon removal, the cost should eventually appear somewhere in cash flow. If better hardware efficiency allows the company to provide more compute using less energy, the sustainability improvement may also improve unit economics.

This is where ESG analysis can add something to ordinary financial analysis rather than running alongside it.

The same distinction should apply across the AI ecosystem.

AI can create attractive opportunities across power generation, transmission, storage, electrical equipment, cooling, and semiconductors while also increasing environmental pressure. The investment question is whether companies can earn attractive returns while solving those constraints.

At the portfolio level, the exposure becomes harder to read.

A portfolio may still fit a sustainable mandate, while its environmental exposure is now coming from a different part of the value chain.

A portfolio manager should therefore look beyond sector labels and ask:

• How much of expected earnings growth across the portfolio depends on the same data-center expansion?
• How many holdings are exposed to the same power bottlenecks?
• Would a sharp increase in electricity costs hit several companies at once?
• Does the portfolio contain companies that benefit from solving those constraints as well as companies that create the demand?

These questions are more useful than deciding whether “AI” belongs inside an ESG allocation.

They also show why the relationship between ESG and valuation should not be reduced to adding a premium or discount to a stock.

Not every environmental metric will matter financially; the challenge is identifying the ones that actually change the investment case.

For investors, ESG information becomes relevant when it changes the economics of the business. AI makes that particularly visible because its environmental footprint is increasingly tied to growth, infrastructure and capital spending.