“Uncertainty” around the surging power consumption of AI-focused data centres is confounding Google’s efforts to cut carbon emissions and “challenging” its net-zero strategy, the company has admitted, as Bill Gates began construction of a novel nuclear power plant that may help.

Despite a 2021 commitment to reach net zero emissions across its operations and “value chain” by 2030, Google’s newly released 2024 Environmental Report paints a much different story – with emissions increasing by 13 per cent from 2022 to 2023 and 48 per cent compared to its 2019 baseline.

This, despite the company undertaking a range of technological refinements including the newly released Trillium chip – whose sixth generation, Google claims, is 67 per cent more energy efficient than the previous chip – as well as contracts to purchase 4GW of clean energy generation capacity in Texas, Belgium, and Australia, and the establishment of a water risk framework designed to improve cooling in its data centres.

Despite these initiatives, Google admitted that its investment in countless new servers to run its AI services has complicated a net-zero picture that could not, back in 2021, foresee the explosion of AI adoption since OpenAI released its ChatGPT generative AI (genAI) tool in November 2022.

Now, as Google throws money and computing power at building out its own genAI capabilities, the company believes reaching net-zero emissions by 2030 is an “extremely ambitious goal” because the future environmental impact of AI “is complex and difficult to predict.”

“As we further integrate AI into our products, reducing emissions may be challenging,” the report notes, “due to increasing energy demands from the greater intensity of AI compute, and the emissions associated with the expected increases in our technical infrastructure investment.”

While AI powered applications for organising information, improving prediction and optimising systems like traffic lights are expected to help reduce global carbon emissions by 5 to 10 per cent by 2030, the power consumption of those applications will continue to increase as companies like OpenAI, Google, Microsoft, and Meta build massive, power hungry data centres to support those tools.

Goldman Sachs Research, for one, recently noted that a ChatGPT query consumes nearly 10 times as much power as a Google search and estimated that AI will be a key force in growing data centre power demand by 160 per cent by 2030 – driving what it called a “sea change in how the US, Europe, and the world at large will consume power – and how much that will cost.”

The increasingly AI-focused tech industry is well aware of the challenges, with Google noting in its report that “we know that scaling AI and using it to accelerate climate action is just as crucial as addressing the environmental impact associated with it” – and that resolving this paradox “will require us to navigate significant uncertainty…. solutions for some key global challenges don’t currently exist.”

Taking the nuclear option?

Developing those solutions has become increasingly crucial to the data centre industry as it explores ways to sustainably power data centres using solar, wind, water – and, if a new project founded and backed by Microsoft founder Bill Gates gets its way, small nuclear reactors.

Nuclear innovator TerraPower, which Gates established in 2008, “has proven that we can do nuclear better,” Gates wrote as the company last month broke ground on the construction of a 345 megawatt nuclear facility in Kemmerer, Wyoming, whose sodium-powered Natrium reactor is expected to come online by 2030.

As well as generating power, Natrium uses a molten salt energy storage system that will allow it to also function as a large-scale battery – enabling it to fill in gaps in coverage when the output of large-scale renewable systems ebbs, as at nighttime.

Gates isn’t the only tech luminary betting on nuclear: OpenAI founder Sam Altman chairs TerraPower rival Oklo, whose Aurora nuclear reactor is similarly aiming to offset the explosion in AI computing power with an economically viable small nuclear solution.

Yet while these and other small modular reactors (SMRs) such reactors could conceivably be collocated with large AI data centres in the future, one recent analysis noted that SMRs produce significantly more nuclear waste than conventional reactors – creating new issues even as they help companies like Google get their net zero efforts back on track.