New study shows AI emissions path — and how to bend the curve
CLIMATEFEATURED
Suspendisse faucibus mauris massa, ut scelerisque est ultricies nec. Aliquam quis dictum tortor. Nunc tincidunt gravida ornare. Sed pulvinar faucibus dolor ac efficitur.
Why it matters: Accenture's study offers fresh modeling — and metrics — as policymakers, companies and other stakeholders try to understand AI's path.
The big picture: Its "base case" scenario sees carbon emissions linked to AI rising to 3.4% of the global total in 2030, with data centers powering the tech using as much electricity as Canada.
"Cooling those data centers may consume 3.02 billion cubic meters of fresh water, more than the total annual freshwater withdrawals of countries like Norway or Sweden," the research finds.
Threat level: Beyond ecological harms, Accenture's report warns of companies scaling their AI investments facing rising costs, regulatory risks, public backlash and more.
The intrigue: The consulting giant urges companies to employ a new frame that looks holistically at AI costs and benefits, as opposed to just measures like efficiency.
It calls it the "Sustainability-Adjusted Intelligence Quotient."
It's designed to measure "how efficiently AI systems convert money, electricity, water and carbon into actual performance."
State of play: The analysis offers companies several categories of advice for making AI less costly and resource-intensive.
One of them is "smart silicon" — that is, ways to integrate hardware and software more efficiently, such as emerging systems that cut down on the "costly movement of data between memory and processors."
What we're watching: Other advice explores topics like data center location and ways to sell idle computing capacity.
And sometimes the only winning move is not to play.
The study notes that organizations sometimes default to using large language models when more "task-specific" ones would do the trick.
Source: Axios