Big tech AI related emissions could be 700% higher than they claim

Big tech AI related emissions could be 700% higher than they claim
The boom in AI use will double power demand by 2030 and their emissions are probably almost seven-times higher than they claim. / bne IntelliNews
By bne IntelliNews September 18, 2024

US big tech companies claim they are carbon neutral, but an investigation by The Guardian found emissions related to the rapid growth of data centres needed to drive AI has seen their emissions rise more than seven-times more than they claim, the paper reported on September 15.

According to a Guardian analysis, from 2020 to 2022, the real emissions from the company-owned data centres of tech giants Google, Microsoft, Meta and Apple were likely 662% higher than the companies officially reported. The image conscious tech giants appear to have been lying about the environmental cost of developing the burgeoning AI business as emissions are 7.62 times more than companies disclosed.

Amazon, the largest emitter among the top five tech companies, was excluded from the calculation as it relies on cloud services rather than the much more energy-intensive AI operations that require ten-times more calculations to function than the simple processing of online orders. However, its emissions far exceed those of Apple, the second-largest emitter, which produced less than half of Amazon's total in 2022.

AI’s growing energy demands

At a conference in Houston this week, leading industry figures warned that AI is set to dramatically increase the demand for power in the coming years by 70 GW to a total of 870 GW in the US alone.

"Natural gas will help power the rapid growth of artificial intelligence with its insatiable demand for reliable electricity," Chevron CEO Mike Wirth said September 17 at a conference in Houston during a keynote address opening the annual Gastech conference. "[This] means AI's advance will depend not only on the design labs of Silicon Valley, but also on the gas fields of the Permian Basin," he said.

Demand from large industrial sources, including data centres, could mean that total US Lower 48 peak electricity demand needs could rise to 870 GW by 2030, up from about 800 GW in 2023, according to gas market analysts with S&P Global Commodity Insights.

"Growth in peak demand will result in a stronger call on gas-fired generators," the analysts said in a long-term North American market outlook updated in September.

The energy consumption of data centres has already been significant. In 2022, they accounted for 1% to 1.5% of global electricity consumption, according to the International Energy Agency (IEA). However, the rise of AI is pushing these figures higher. AI systems like ChatGPT require up to 10 times more electricity than typical cloud-based applications. Goldman Sachs projects that data centre power demand could grow by 160% by 2030, with Morgan Stanley forecasting global data centre emissions to accumulate to 2.5bn tonnes of CO₂ equivalent by the same year, The Guardian reports.

Despite this surge in power demand and associated emissions, big tech firms continue to claim they are environmentally friendly. Amazon, for instance, asserted in July that it had achieved its carbon neutrality goal seven years ahead of schedule, implementing a 3% reduction in gross emissions. However, the company later dropped the claim. Critics claim that Amazon is actually expanding its fossil fuel use, through the use of new data centres and increased volume of deliveries using trucks burning fossil fuels.

Accused of using "creative accounting" when it comes to reporting their carbon footprint, the companies make use of renewable energy certificates (RECs). The tech firms have been actively seeking out green sources of power in order to reduce their emissions, but critics claim these purchases are little more than window dressing.

Companies purchase RECs to match a portion of their electricity consumption with renewable energy generation, even if the energy is produced far from the company’s facilities. RECs enable companies to report lower "market-based" emissions, which differ from "location-based" emissions – the actual emissions generated where the data is processed, The Guardian reports.

If the combined location-based emissions of Google, Microsoft, Meta, Apple and Amazon were considered as a country, their emissions in 2022 would rank them as the 33rd highest-emitting nation, just behind the Philippines.

Despite the accuracy of location-based accounting, the Greenhouse Gas (GHG) Protocol permits the use of RECs, sparking a debate within the tech sector. Amazon and Meta, through the Emissions First Partnership, advocate for the continued use of RECs regardless of their geographical origins. Meanwhile, Google and Microsoft are pushing for more stringent standards, calling for energy consumption to match renewable production on a 24/7 basis.

The disparities between official and actual emissions numbers are stark. Meta’s official scope 2 emissions for 2022 were reported as 273 tonnes of CO₂ equivalent, but when calculated using a location-based approach, that figure skyrocketed to over 3.8mn tonnes – a 19,000-fold increase. Similarly, Microsoft’s official emissions for 2022 were reported at 280,782 tonnes, while location-based emissions totalled 6.1mn tonnes.

Even more emissions are tied to third-party data centres, which big tech companies heavily rely on. These facilities contribute to scope 3 emissions, the most uncertain category of emissions reporting.

The demand for data centres is set to rise sharply and will at least double its power consumption in the coming years, as AI applications continue to expand. Google and Microsoft have already cited AI workloads as a factor driving their recent upticks in emissions.

Industry experts worry that existing power grids may not be able to meet the growing energy demand of data centres, which is expected to double by 2030. Industry professionals have warned that the US may face power generation deficits in the next two years as a result of exploding AI power demand.

Tech

Dismiss