Businesses would rather scale back their sustainability commitments than miss out on realising the benefits of generative artificial intelligence, according to a new report from Capgemini. Of those implementing the technology across most or all functions, 47% have “had to relook” at their original environmental goals.
In July, Google came under fire after its annual environmental report revealed that its emissions had increased by 48% in four years thanks to the expansion of its data centres to support AI developments. It also stated that its goal to reach net-zero emissions across all its operations and value chain by 2030 is now “extremely ambitious” and “will require (Google) to navigate significant uncertainty.”
For “Developing Sustainable GenAI,” the Capgemini Research Institute surveyed executives from 2,000 large organisations worldwide that were already working with GenAI. Almost half (47%) said their organisation’s greenhouse gas emissions had increased in the last year by an average of 6%, and a similar proportion (48%) linked a rise to their AI usage.
Generative AI demands a substantial amount of energy and water
GenAI has an aggressive environmental impact. The graphics processing units central to the technology’s operation require rare Earth metals that must be mined, releasing greenhouse gases. The hardware behind it also requires frequent upgrades, with studies suggesting this could create up to five million tonnes of e-waste by 2030.
It is estimated that data centres will be responsible for up to 4% of global power demand by 2030, driven, at least in part, by AI. Training OpenAI’s GPT-4, with 1.76 trillion parameters, consumed an amount of energy equivalent to the annual power usage of five thousand U.S. households. This doesn’t even include the electricity required for inference, where the AI generates outputs based on new data.
A substantial amount of water is also required to cool the servers. Running an inference of between 10 and 50 queries on a large language model uses about 500 ml of water.
SEE: Sending One Email With ChatGPT is the Equivalent of Consuming One Bottle of Water
The E.U. has the lofty goal of reducing the region’s 2030 greenhouse gas emissions to at least 11.7% lower than projected at the start of the decade. However, demand for bit barns in Europe is predicted to triple in that time, increasing their share of the region’s total energy demand by 3% and pushing that goal out of reach.
Businesses may not know, or even care, about the emissions attached to their AI usage
Many businesses use AI now, with 80% having increased their investment in it since 2023, according to Capgemini. Nearly a quarter are now integrating generative AI into some or most of their locations or functions, up from 6% in 2023.
SEE: 31% of Organizations Using Generative AI Ask It To Write Code
However, the new report highlights that awareness of AI’s electricity and water demands is patchy. Only 38% of executives surveyed claim to be aware of the environmental impact of the GenAI they use, and 12% say their company measures its footprint.
Of those surveyed that are aware of the impact, 51% say that AI use is one of the primary reasons for their organisation’s rise in emissions. They also expect it to increase the proportion of their emissions that come from internal operations within the next two years by 2.2%.
The lack of businesses monitoring the environmental impact of their GenAI usage is not due to a lack of effort. Almost three-quarters (74%) of those surveyed said doing so is challenging due to limited transparency from hyperscalers and model providers.
A report from the Uptime Institute found that fewer than half of data center owners and operators track metrics like renewable energy consumption and water usage. The emissions of data centres owned by Google, Microsoft, Meta, and Apple are likely to be about 662% higher than officially reported, according to The Guardian. This is largely due to renewable energy certificates and carbon offset schemes, which allow companies to claim they use renewable energy when they don’t.
SEE: Power Shortages Stall Data Centre Growth in UK, Europe
On the other hand, executives may not be concerned about the impact of AI usage on their company’s emissions. Only a fifth of respondents to the Capgemini survey ranked the environmental footprint among the top five factors when selecting or building GenAI models.
Cost competitiveness was ranked among the top five considerations by 53% of executives. However, this is fundamentally connected to energy use, according to Samuel Young, AI practice manager at research firm Energy Systems Catapult.
He said: “When implementing at scale, organisations quickly become sensitive to inference costs. They therefore have an incentive to adopt less energy-intensive models, which can reduce carbon impact.”
Read the full article here
