Executives Overestimate AI's Energy Footprint by 16x, Bain Finds — and Still Can't Agree on Its ROI

Executives estimate AI will consume 11.4% of global energy within three years; Bain’s own Intersect model puts the real figure at 0.7% — a 16x overestimate that consumers manage to beat, guessing 19%, a 27x miss. That’s the headline finding in Bain’s CEO Sustainability Report 2026, authored by Wissam Yassine, Harry Morrison, Martha Moreau, and Krystle Jiang, and it lands alongside a harder number: the share of Americans more worried than excited about AI grew from 37% in 2021 to 50% in 2025, with nearly two-thirds saying they’ve taken concrete action over AI concerns.

The report’s more consequential split is between companies, not perceptions. Among the top-20% “Shapers,” 90% still see major sustainability opportunity in AI, with confidence down only 4 points from 2025; among “Laggards,” confidence collapsed from 57% to 41%. Internally, the gap tracks a familiar failure mode: 75% of business leaders cite unclear ROI as their top barrier to sustainable AI investment, while their own sustainability teams are prioritizing data quality and regulatory compliance instead — two departments solving different problems under the same budget line. Bain frames three practical entry points — Make Better (operational efficiency), Sell Better (sustainable product design), and Protect Better (climate risk quantification) — as the fix, arguing environmental impact ranks only tenth among companies’ stated barriers to sustainable AI adoption, well behind the coordination problem itself.

The ROI-measurement gap isn’t unique to sustainability spend. Gartner’s September survey of 161 chief audit executives found 93% of audit functions already using AI, yet 60% lack a formal strategy and 54% haven’t begun measuring return at all — the same pattern Bain describes: near-universal adoption outrunning the governance and measurement needed to prove it’s worth the spend, function after function.