Investors grapple with AI’s carbon impact and mounting investment risks

Global investment in the AI sector has surged in 2026, with the sector set to attract some $1tn in inflows this year, more than half of which will be deployed in the US, according to Goldman Sachs. By extension, AI's footprint within institutional portfolios has also increased, and this extends beyond listed equities, as Jenny Buck, CIO of the LGPS pool London CIV, says.

"As you start to peel back the onion, how much exposure have you got through your real estate or your infrastructure with data centres, how much of your private credit is lending to these guys, and how much in your private equity portfolio?" 

Private Markets and Portfolio Construction | 5 November 2026 | Central Court, London

But despite the pervasiveness of the sector, very few long-term investors have to date set out their investment beliefs on AI, as a research note by Silicon Boundaries, a joint initiative of the Oxford Sustainable Finance Group at the University of Oxford and Theia Finance Lab, warns.

Of the world's 200 largest pension funds and sovereign wealth funds, only 99 disclosed their investment beliefs and only one, Temasek, explicitly mentions AI, the research by Jakob Thomä and Ben Caldecott points out. While some of the largest investors such as NBIM, Legal & General Investment Management, Royal London Asset Management, Railpen and CalPERS have set out more detailed views in their voting and engagement policies, overall, there appears to be silence on some of the defining issues for the sector, such as whether there is a need for policy intervention or how to best navigate AI's environmental and carbon footprint.

"We believe that investors should clarify what they believe about AI and write it down, whether or not they publish it, both to keep their own policies consistent and to explain their position to their clients, limited partners or beneficiaries. The need will grow as AI attracts more scrutiny and drives wider changes across society," Caldecott and Thomä stress.

AI's energy consumption

Ethical concerns aside, AI exposure raises a series of immediate concerns for investors seeking to navigate the mounting impact of climate change, the most immediate one being its role in driving up global energy demand.

As of last year, global data centre electricity consumption stood at 485 TWh, approximately the equivalent of Germany's annual electricity generation, and it is now on track to double to about 950 TWh by 2030, the IEA predicts in its latest energy and AI outlook, Key Questions on Energy and AI.

This, of course, would not be a problem if AI was powered by renewables, however, this is not the case for the time being. In the US, the heart of the data centre boom, only a third of data centres are currently powered by renewables PPAs, IEA data shows, while European data centres derive about 40% of their power from clean energy sources.

Meanwhile, some 47GW of onsite natural gas power projects for data centres in the US were either under active development or proposed as of 2026, the IEA's project pipeline shows, although the agency expects only 15 to 27 GW of onsite gas to be operating by 2030, mostly in the US.

Tracking AI's carbon footprint

As the sector grows, institutional investors struggle to navigate its growing carbon footprint. While corporates, including hyperscalers, tend to report on their carbon footprint in absolute terms, investors tend to assess their carbon footprint using measures such as weighted average carbon intensity, which takes each issuer's scope 1 and 2 emissions per million dollars of revenue and multiplies it by that holding's share of the portfolio. For the time being, the sheer scale of revenue of hyperscalers means that AI's carbon footprint at portfolio level has not been unusually high, says Sanjay Joshi, consultant at Hymans Robertson and a member of the IFoA working party on responsible AI.

At the same time, actuaries recognise that the net amount of emissions from AI is significant and growing, warranting closer attention on how to account for it, Joshi says.

This is further complicated by the fact that AI will also have indirect effects that could result in improvements to tackling climate problems, he adds.

Joshi warns that the rollout of data centres could easily become prone to the Jevons effect: "the reduction in emissions intensity of energy could also make it cheaper, which would drive more activity, which could mean that the absolute amount of emissions increases," he says.

AI hyperscalers themselves are acutely aware of these challenges. Ahead of its planned IPO, Anthropic cites the uncertain regulatory and political environment surrounding its operations as a risk factor for its business and its data centre buildout in its S1 filing. 

"Our business is subject to an evolving and uncertain regulatory and political environment, which could increase our costs, restrict our operations, or adversely affect demand for our models, platform, and applications," the firm cautions.

Investment risks

These concerns are closely linked to the wider investment risks the sector is facing. If the sector is to grow as anticipated, more money will have to be deployed in the buildout of data centres, which is increasingly confronted with material constraints to its growth while a handful of tech giants, chip makers and cloud providers mutually invest in, lend to and buy from each other, cautions Gaurav Saroliya, global macro portfolio manager at Allianz GI. "I think there is going to be a very meaningful correction, expectations are lofty. If inflation picks up materially and the Fed is forced to respond more aggressively, that could become a trigger," he says.

As the sector continues to grow, many investors struggle to keep pace. For long term asset owners, the challenge will be to track their exposure to a sector that extends across asset classes and may well be too big to fail.