AI adoption stalls in large enterprises as doubts grow over returns

AI adoption among large enterprises has dipped slightly, even as overall use of the technology continues to rise across the corporate landscape.

New figures from the US Census Bureau show that uptake among companies with more than 250 employees fell from a peak of around 14 per cent to 12 per cent in recent summer surveys.

The decline, though modest, points to hesitancy among larger businesses that have already committed billions to AI infrastructure, including data centres and rollout support. Analysts suggest the slowdown reflects frustration with unclear returns on investment.

Across all companies surveyed, however, adoption remains on an upward trajectory. The Bureau’s biweekly survey of 1.2 million US businesses found that 9.7 per cent of respondents had used AI in the past two weeks, up from 8.8 per cent previously. Forward-looking sentiment is also strengthening, with 13.7 per cent of companies expecting to adopt AI for producing goods or services within six months. Yet nearly two-thirds still report no plans to use AI at all, underscoring that mainstream adoption is still emerging.

Industry leaders caution that the figures do not mean companies are walking away from AI altogether. Sheila Flavell CBE, chief operating officer of FDM Group, said the dip demonstrates the limits of deploying AI without adequate training and strategy. “AI can only deliver value when people know how to use it effectively,” she explained. “When organisations implement practical, hands-on training, it builds confidence and helps employees understand how AI can support their roles.”

Others argue that security and governance remain pressing concerns. Andy Ward, senior vice-president international at Absolute Security, noted that over a third of chief information security officers have already banned tools such as DeepSeek, citing privacy and control risks. “AI can transform detection and response, but if it’s deployed without robust resilience strategies, real-time visibility and clear governance, it risks adding more vulnerabilities than it solves,” he said.

Some experts argue the Census Bureau’s methodology may understate true AI adoption because it counts only use in producing goods and services, excluding applications in marketing, administration and customer service. Analysts at UBS point out that, despite the recent dip, AI adoption overall is progressing more quickly than US e-commerce did in its first two decades.

But scepticism is growing around whether the scale of corporate investment is justified. Torsten Sløk, chief economist at Apollo Academy, has warned that the slowdown highlights caution among big corporates, while Arpit Gupta, associate professor at NYU Stern, suggested “trillions in AI capex should probably be reconsidered.”

A recent MIT report adds to the uncertainty, finding that 95 per cent of companies struggle to generate financial returns from AI. Together, these signals have fuelled speculation that an AI bubble could be inflating — with large enterprises pausing to reassess before committing to further rollouts.