The “AI ROI gap” is the distance between how much money is being spent on AI and how much measurable return it has produced so far. Two kinds of evidence define the gap: the mismatch between infrastructure spending and AI revenue, and the still-modest rate at which businesses have actually put AI to work.
On the revenue side, Sequoia’s David Cahn laid out the arithmetic in his June 2024 essay “AI’s $600B Question.” Scaling up from NVIDIA’s data-center sales, he estimated that AI products would need to generate on the order of 600 billion dollars in annual revenue to justify the infrastructure being built - a figure far above what the industry was actually earning. The gap between the implied revenue and the real revenue is one face of the ROI problem.
On the adoption side, official data shows uptake that is real but uneven. The US Census Bureau’s Business Trends and Outlook Survey found that, across late 2025 and early 2026, the share of US businesses using AI in a business function hovered roughly between 17 and 20 percent. Adoption was heavily skewed by size and sector: large firms and knowledge-intensive industries like Information and Finance ran well ahead (the Information sector near 40 percent), while small firms and sectors like retail lagged. Broad, economy-wide AI use - the kind that would generate the revenue to close the gap - was still building rather than achieved.
Why a business reader should care: the ROI gap is the empirical core of the bubble debate. The optimistic case (an adoption J-curve, with returns arriving after a lag) and the pessimistic case (overbuilt capacity chasing demand that never fully arrives) make the same observation today; what differs is whether they expect the gap to close. Tracking real adoption and real revenue, not announcements, is how a business reader can judge which way it is going.