Big tech emissions under scrutiny

However, a report by The Guardian suggests that the emissions from their in-house data centres could be 7.62 times higher than the official figures they have provided.

The claims of reduced carbon emissions rely on the use of market-based accounting methods, where companies purchase Renewable Energy Certificates (RECs) to offset their carbon footprint. However, these methods may not accurately reflect the emissions generated by the actual energy consumption at their data centres. While the companies claim to be making strides towards carbon neutrality, the reality could be vastly different.

According to The Guardian, the real emissions from in-house data centres between 2020 and 2022 for companies like Google, Microsoft, Meta, and Apple are significantly higher than the figures reported. In the case of Meta, the company's market-based report for 2022 listed its data centre emissions as 273 metric tons of CO2 equivalent, but under location-based calculations, this number jumps to 3.8 million metric tons – a 19,000 times increase. Similarly, Microsoft reported 280,782 metric tons of CO2, which, under more accurate accounting, could be as high as 6.1 million metric tons.

AI driving emissions

The growing demands of AI are exacerbating the issue. AI-driven technologies, such as ChatGPT, are far more energy-intensive than traditional Cloud applications with Goldman Saches reporting that a single ChatGPT query requires nearly 10 times more electricity than a Google search, and data centre power demand is expected to rise by 160% by 2030​.

As tech companies race to expand their AI capabilities, the energy consumption of data centres is becoming unsustainable, contributing heavily to emissions. Commenting on this situation, Elio van Puyvelde, Chief Information Officer at Nscale, an AI cloud provider, comments: “The race for AI dominance is heating up, but at what cost? The increase in emissions, 7.62x higher than initially reported by these tech giants, are largely down to legacy data centres unable to cope with the power demands of AI. And while big tech invests heavily in renewables, the sheer scale of the AI boom threatens to overwhelm those efforts.”

Optimising energy consumption in AI

While big tech companies are increasingly turning to renewable energy sources, there is growing recognition that simply buying more green energy will not solve the problem. Van Puyvelde highlights the need for optimisation of AI hardware and software to manage energy consumption levels more effectively: “Big tech is in a bind, but there’s a way out. Ensuring data centres are built where there is stable supply of renewable power is one thing, but the industry must focus on maximising efficiency too. The reality is AI workloads are so often poorly optimised, wasting energy use and delivering poor returns. Accelerating AI hardware and software optimisation is one of the best routes to managing AI energy consumption levels. Companies must optimise and fine-tune AI models as well as invest in energy efficient AI accelerator hardware. The open source community is already hard at work finding ways to make AI models run more efficiently and with less impact on the environment. Open source innovation will drive more efficient and sustainable AI use.”

The path forward

According to the report, the emissions from data centres are expected to rise dramatically as AI applications grow in scope. While companies like Google and Microsoft have set ambitious goals to reduce their carbon footprint, such as Google’s 24/7 renewable energy initiative, the gap between their reported emissions and the actual energy consumed remains concerning. Without a clear commitment to transparency and better energy optimisation practices, big tech’s efforts to become greener could fall short in the face of increasing AI energy demands.