AI datacentres: the climate risk moving onto Big Tech’s balance sheet

Shareholders are starting to ask tougher questions about artificial intelligence (AI) at technology company AGMs, spanning governance, human rights and risk oversight. 

Yet one of the biggest AI issues may be hiding in plain sight: the infrastructure needed to power it. As data centres become a larger part of Big Tech’s capital spending, featuring prominently on balance sheets, climate risk is becoming a direct financial issue, not just a sustainability disclosure.

For a long time, climate risk felt like a moot point for technology companies compared to heavy industry. A software firm or online platform did not own many smokestacks, mines or refineries. Its climate exposure was often indirect: emissions from suppliers, electricity purchased from the grid, the manufacturing of devices and chips, or the way customers used its products. Climate risk sat one step removed from the company’s own operations.

But now, AI is changing that, and its environmental impact can’t be ignored. The race to build and train AI models is pushing companies such as Microsoft, Alphabet, Amazon and Meta into a much more infrastructure-heavy business. AI needs data centres, and data centres need land, power, water, cooling systems, backup generation, grid connections and long-term local permits. These are not abstract emissions on a spreadsheet, but rather physical assets exposed to physical climate risks.

Three ways AI is changing climate risk

1. Energy: constant demand, rising pressure

The first and most immediate consideration is energy and AI’s carbon footprint.

AI data centres consume huge amounts of electricity. A traditional office can reduce lighting or heating in a pinch. A data centre, however, cannot simply “switch off” without affecting customers, cloud services or AI workloads. It needs reliable power all day, every day.

That creates several risks. In regions where the electricity grid is already strained, new data centres may face delays getting connected. They may have to pay more for power, fund their own energy infrastructure, or sign long-term contracts at higher prices. In some cases, public opposition may rise if residents find data centres are competing with households for grid capacity.

Northern Virginia is one such example. It has become one of the world’s largest data centre hubs because of its fibre connections, skilled workforce and proximity to major customers. But that concentration also creates pressure on the local power grid. And this is not just limited to the US. In Europe, Ireland’s rapid data centre growth has raised concerns about electricity demand, grid stability and the country’s ability to meet climate targets. 

2. Water: a growing social constraint

The second consideration is water.

Data centres generate heat. To keep servers running efficiently, companies need cooling. Some facilities use air cooling, but many rely partly on water-based cooling systems, particularly where energy efficiency is a priority. This creates a tension that is easy to understand: the hotter the weather gets, the more cooling is needed; but hot weather often comes with drought, water scarcity and pressure on local supplies.

That matters in places such as Arizona, Texas, Spain, India and parts of Australia, where water stress is already a political and economic issue. Imagine a fast-growing desert city facing drought restrictions while a new AI data centre seeks approval to use millions of litres of water for cooling. Even if the facility is legally permitted, the reputational and regulatory risk is obvious.

This is also where climate risk becomes social risk. A data centre may bring jobs and tax revenue, but local communities may ask whether those benefits justify the strain on water resources. Investors should pay attention to that because it could affect planning approvals, operating costs and long-term asset value.

3. Physical damage: when real costs come to bear

The third consideration is physical damage.

Data centres may feel invisible to most people, but they are buildings filled with expensive equipment. They can be affected by floods, storms, wildfires, extreme heat and supply-chain disruption. A flood can do real damage to backup power systems. Wildfire smoke can affect air filtration and worker access. Extreme heat can reduce cooling efficiency and raise electricity demand at precisely the moment the local grid is under the most pressure. 

A coastal data centre may need higher flood defences. A facility in a wildfire-prone area may require stronger backup systems and air filtration. A site in a hot region may need more expensive cooling technology. These are real costs, not theoretical ESG talking points.  But arguably, tech companies are not used to having physical confinements on their ability to build, grow and scale ideas.

What this means for investors 

AI changes the investment debate because historically, a technology company’s climate problem was often framed as reputational: are its suppliers decarbonising, is it buying enough renewable power, and are its emissions disclosures credible? Those questions still matter. But AI datacentres add another layer. Climate risk is moving from the footnotes of sustainability reports into the tangible asset base. It affects where companies build, how much they spend, how quickly they can grow and how resilient those assets – in some ways, their new foundations – are over time.

For investors, this is exactly the kind of issue that ESG-risk assessment is designed to capture. At Charles Stanley, we assess and monitor material climate-related factors across investments, including assessing ESG-risk, carbon footprint, and carbon emissions help assess the potential financial impact on investments. We also consider forward-looking climate-related risks and opportunities through tools such as scenario analysis and Paris Agreement alignment. 

But AI is not only a warning story. It’s also an opportunity story. If AI infrastructure is becoming one of the defining investment themes of the decade, then the supporting systems matter too. Grid upgrades, transmission equipment, nuclear power, geothermal energy, battery storage, advanced cooling, water recycling, energy management software and climate-risk analytics could all become more valuable. Investment in these areas, when balanced with a detailed assessment of the risks, may present opportunities for investors. 

So, the key point is simple: AI may be digital, but its constraints are increasingly physical. Investors who only look at model performance or cloud revenue may miss the harder question underneath… can these companies secure enough clean power, water and resilient infrastructure to make the AI boom profitable and durable?