Viewing AI Through a Sustainability Lens

Let’s look at demand first. Already around 10% of American companies have meaningfully integrated AI into their production processes, according to official data.⁴ Our research included 30 interviews with AI project leads at a whole range of firms. What began in finance and tech has broadened to healthcare, logistics and heavy industry. Mayo Clinic, for instance, expects AI to detect cancers several months earlier.⁵ This could in turn carve out billions in long‑term savings.

We believe that Microsoft sits at a privileged position on this demand curve. The company’s Azure product is a distribution channel for enterprise AI, from Copilot to fine‑tuning tools. Accenture is a clear example of a consultancy leaning into AI adoption. Their pitch is simple: when every business starts using AI you need someone to coordinate it all.

Of course, future demand for AI services is not guaranteed. Some recent evidence, including from a Massachusetts Institute of Technology (MIT) study, suggests poor returns from corporate experimentation in AI products.⁶ We do not interpret this result entirely negatively: experimentation, by its very nature, often has high rates of failure. AI adoption speeds will vary depending on the size of the problem being addressed, and whether the solution is materially better. Nonetheless, we need to remain vigilant, keeping a close eye on the growth in AI revenues.

We believe many companies in which we invest also stand to gain from the expansion of AI supply. At the silicon layer, Taiwan Semiconductor Manufacturing Company (TSMC) remains pivotal. No other company in the world is so skilled at manufacturing cutting-edge chips. Schneider is a huge supplier of data-centre equipment and is a longstanding partner of choice for hyperscaler firms from the early design phase of a campus. If you built a data centre without any cooling, it simply would not run for more than a few minutes. Trane’s portfolio spans facility‑level chillers and rack‑level liquid cooling. At the software layer, WEKA (a Growth Equity strategy investment) is helping decouple compute growth from energy use. Its platform helps cut hardware needs, aiding customers to avoid substantial carbon emissions.

Just like with AI demand, supply also faces numerous bottlenecks. It takes a long time to establish grid connections and to install transformers for data centres. A connection to a high‑voltage line can take five to ten years. Skilled labour shortages in mechanical and electrical engineering are non‑trivial. In recent years Amsterdam, Singapore and Dublin have paused new data centre capacity, further slowing construction. The practical implication is that companies can break ground on a huge data centre and still wait years to turn it on.

The fundamental risk, therefore, is a ‘maturity mismatch’ between the demand and supply of AI. Companies are in a frenzy of AI experimentation. AI-service revenues are rapidly growing. And yet meeting future demand for AI requires years of long, careful planning. There is no guarantee that companies or investors will always be willing to fund these huge, long-term projects. Software teams will therefore need to deliver sticky AI uses that justify larger clusters, so that boards keep funding projects that, in turn, make the next level of models feasible.

We are carefully monitoring the climate effects of the AI build-out. Some commentators see reason for optimism. Credible studies suggest AI could reduce global emissions by tightening industrial controls, routing logistics more intelligently and improving building operations. A recent study found that AI applications across power, food and transportation could reduce global emissions by 6% to 10% annually by 2035.⁷ Bear in mind, however, that these very same optimisation tools can make it cheaper and faster to extract fossil fuels. The impact of AI use on the climate therefore remains an open question.

AI training and inference have their own climate impacts. Data centres are becoming more efficient over time, meaning that they use less energy to achieve a given operation. Historically, hardware efficiency has compounded at roughly 40% per year, and software can deliver even larger improvements.⁸ Google has used AI to cut data centre cooling energy requirements substantially.

The unprecedented scale of the AI buildout nonetheless presents challenges. ‘Extreme ultraviolet’ tools allow for the manufacture of much more powerful chips, but they consume dramatically more power per wafer than ‘deep ultraviolet’ tools, the workhorse of chip manufacturing for decades. This is one reason why TSMC’s Scope 2 emissions (those caused indirectly, via energy consumed) have proved stubborn. Meanwhile, many data centres sit on grids that remain powered by fossil-fuel energy. On current trajectories, global data centre carbon emissions could rise from 0.5% of global emissions today to more than 1%.⁹

The solution is simple but hard to achieve: improve data centre efficiency and decarbonise the grid at the same time. The major cloud providers are leading on this front. Power-use effectiveness, a measure of how efficiently a data centre uses energy, now sits near 1.1-1.2 for the leaders, meaning that for every unit of energy used, only 20% goes on overhead.¹⁰ This is impressive. Alphabet, Amazon and Microsoft claim 100% renewable energy on an annual basis. We are hopeful that renewable energy, especially wind and solar, can supply the growing power needs of data centres.

We also look for progress on water use. A large data centre facility can consume on the order of 2.5 billion litres per year (which would supply 80,000 people for a year).¹¹ It is important to use this water sustainably. Amazon, Microsoft, Alphabet and Meta have all set ‘water positive’ targets for 2030, meaning they will replenish more clean water than they use. Amazon Web Services discloses a water‑usage‑effectiveness metric of roughly 0.15 L/kWh (meaning that for every kilowatt-hour of electricity consumed by AWS’s data centres, about 0.15 litres of water is used, mainly for cooling). Microsoft discloses about 0.30 L/kWh today. We think companies will find that, if they fail to use water efficiently, local politicians and campaigners will hold them to account. 

Companies embarking on this technological revolution while ignoring planetary boundaries risk losing our support – as well as their social licence to operate. Our expectations are clear. We strongly advocate for aligning compute usage with clean hours, such as times of surplus renewable energy; powering it on authentically zero-emissions energy (meaning additional, real, clean generation that is not double counted); and supporting AI applications that meaningfully accelerate the mitigation of, and adaptation to, climate change. We outlined some of these expectations in a recent letter to the large cloud companies, and will continue to engage with these companies in the future. The goal is to ensure the supply needed to meet rising demand is also the supply that gets the world to net zero.

While our focus here is on the environmental and economic implications of AI, we recognise – consistent with Generation’s broader firmwide research – that the technology’s societal effects, from labour markets to privacy, will be equally consequential in determining whether AI advances human and planetary well-being. We are working through the social risks and implications of AI, from reinforcing disinformation and inequality, to threats to wellbeing and relationships, to security issues. Expect to hear more from Generation on this topic soon.

With AI, the world faces a tremendous opportunity but also a tremendous challenge. In an ideal world, an AI revolution will provide the necessary funds to develop ever more sophisticated AI models, a virtuous circle leading to higher productivity for all. In an ideal world this expansion would not only facilitate lower-carbon business models and products beneficial for society but also would itself be powered by low-carbon energy sources, and with limited social risks. There are no guarantees. But we look forward to investing behind companies that are seeking to make this vision a reality.