Rathbones: AI’s power problem – the climate risk and opportunity investors can’t ignore

As AI investment accelerates, the electricity required to run and cool next-generation data centres is emerging as a key sustainability and infrastructure issue. It is one that is moving from a technical detail to a board-level and investor-engagement priority. This is alongside growing scrutiny of water availability and local infrastructure capacity.

The rapid build-out of AI-focused data centres is driving a step-change in demand. It is increasing pressure on power grids, that were not designed for sustained, city-scale industrial loads, and accelerating investment in networks, storage and efficiency.

In some locations, individual AI campuses are already drawing power comparable to a mid-sized city. David Harrison, Head of Sustainability at Rathbones Asset Management, says this raises questions for sustainability-minded investors around emissions, water availability and allocation, and grid resilience. But while also expanding the opportunity set across power and electrification, subject to physical, regulatory and social constraints.

He continues: “One example is the Stargate AI data-centre campus in Abilene, Texas, described as a purpose-built AI ‘supercluster’. The project—led by Oracle and OpenAI, with financing support from SoftBank, is estimated to cost more than $15bn to build. The cost excludes Nvidia GPUs for computation, which entail considerable upstream water and energy use due to advanced semiconductor manufacturing. 

“The scale is striking. When fully built out, Stargate is expected to require around 1.2 gigawatts of power—comparable to the electricity demand of a mid-sized US city. For context, the UK’s largest power station, Drax, generates around 4GW. And Stargate is far from unique. Estimates suggest global AI data-centre spending will exceed $750bn in 2026 alone, according to Gartner research and Bloomberg figures.

“Data centres already consume more than 4% of total US electricity demand, while water use is rising (as liquid cooling becomes standard for AI loads) and could exceed 12% by the end of the decade, according to McKinsey estimates. After years of flat demand, AI is changing the trajectory for power markets,” Harrison adds.

From challenge to opportunity

Harrison says sustainability investors face a dual imperative: manage the near-term risks tied to higher power demand, while identifying the beneficiaries of faster investment in grids and electrification.

He explains, “Rising electricity demand has clear implications for emissions, water stress at local and regional level and grid resilience. But it is also accelerating investment across the energy system—from transmission and distribution to efficiency, storage and cooling technologies.

“Beyond large incumbents, specialist ‘picks and shovels’ firms are also gaining relevance. Quanta Services plays a key role in building and upgrading US power-grid infrastructure, while Switzerland’s Belimo supplies components used in data-centre cooling systems—an area seeing rising demand as AI workloads intensify. Water infrastructure providers, and treatment and recycling technologies, are also becoming increasingly important as sites scale and local constraints tighten.

“Capturing this opportunity requires exposure across the value chain. Improving efficiency—in generation, transmission, cooling and energy management—will be as important as adding new capacity,” Harrison says.

A moving target

The AI infrastructure build-out remains early-stage, and the technology and power solutions are evolving quickly. New opportunities are likely to emerge in energy storage, microgrids and advanced power-management tools to support the reliability and sustainability of AI workloads.  However, they are deeply linked to lengthy planning processes, grid reinforcement and public capital deployment.

Harrison says that rising electricity demand is already influencing pricing in parts of the US, increasing the likelihood of political and regulatory intervention, including decisions over how power is allocated. He adds, “The impact will vary by market, depending on energy mix and access to scalable low-carbon supply, particularly where water stress or grid congestion is already acute.

“What is clear to us is that underinvestment in energy infrastructure is no longer an option.  Underinvestment poses material economic and political risk, especially in AI-heavy regions. A balanced, efficient and technologically robust power system is becoming a prerequisite for growth in an AI-driven world—and a critical consideration for long-term investors.”

Why engagement matters

Engagement across the AI datacentre and power ecosystem is becoming a priority, says Harrison, “Early dialogue helps us understand how businesses are approaching hard constraints around power, water and grid access, and how those constraints may shape strategy and capital allocation.

“Discussions with large US technology firms have provided insight into siting decisions, the efficiency gains being delivered through technology, and how companies are working with local authorities in more energy-constrained regions. That transparency helps establish benchmarks for best practice and informs our wider engagement approach.

“We are also collaborating closely with asset owners on this issue. The power demands of AI will not be resolved overnight, but sustained engagement can help drive continuous improvement—and differentiate between companies that are part of the solution and those that risk worsening the problem.”