IT Sustainability Think Tank: How IT directors can spot false green claims from Big Tech suppliers | Computer Weekly

"One of the most glaring red flags when considering the sustainability of software and IT services is when suppliers tout the benefits of their artificial intelligence (AI)-powered offerings whilst remaining conspicuously tight-lipped about the environmental impacts of the resource-hungry datacentres powering them. Even when a supplier can cite corporate-level commitments (say, for purchasing renewable power or offsetting emissions), if they're unable or unwilling to provide granular-level transparency about impacts at the individual workload level, then this should raise concerns."

"Global average figures that smooth out peaks and troughs, or rely heavily on offsetting to disguise true consumption, can mask uncomfortable truths about how green youruse of their services (in your region, at the time you're using them) actually is. The environmental impact of AI becomes particularly pertinent when considering its application in sustainability use cases. Discussions at COP29 highlighted the 'sustainable AI paradox' - the fact that the very AI systems being deployed to solve climate challenges are themselves energy and water-intensive - and so it's imperative that any sustainability solution deploying AI demonstrates clear net environmental benefits."

The sustainability technology market spans ESG reporting, nature capital, sustainable manufacturing, smart energy grids, green IT, and the circular economy. IT directors must distinguish genuine environmental credentials from averages, aggregates, and spin to prevent greenwashing. Suppliers promoting AI-powered offerings should disclose datacentre energy and water impacts at a granular workload level rather than relying on corporate commitments or offsets. Global averages and heavy offsetting can hide real consumption patterns by region and time. Any sustainability solution using AI must demonstrate clear net environmental benefits that account for training, inference, operational impacts, and verification.

Read at ComputerWeekly.com