AI infrastructure overspend hits $23.4bn, study says

AI and machine-learning infrastructure accounted for $23.4 billion in annual technology infrastructure overspend across eight economies, according to Centre for Economics and Business Research work commissioned by Wasabi Technologies.

The study estimated total technology infrastructure overspend at $63.4 billion a year, attributing it to provider charges that businesses described as opaque, unpredictable or difficult to forecast. It said AI and machine-learning infrastructure made up 37% of that total, more than any other technology category.

The research was conducted among 690 senior business and IT decision-makers in the U.K., the U.S., France, Germany, Italy, Japan, Australia and New Zealand. Wasabi said the margin of error was plus or minus 3.7 percentage points at the total sample level.

Among businesses reporting AI and machine-learning infrastructure overspend, 38% cited higher-than-expected token, inference or usage-based AI charges as a contributing factor, while 49% cited unexpected AI platform and model licensing fees and 45% cited integration, deployment and model operations expenses.

The study said 84% of organisations made at least one cost-driven trade-off over the past year. CEBR modelled that recovering the overspend across the surveyed markets could support $21.8 billion of new productive investment, $7.6 billion of workforce training investment and 111,450 additional jobs, contributing to a long-run GDP uplift of $20.4 billion.