September 16, 2026, (Inside AI) — Corporate America is adopting artificial intelligence at three times the pace it did personal computers four decades ago, yet the expected productivity windfall remains stubbornly absent from official economic data. About 44% of workplaces reported using AI in May, according to a tracker supported by Harvard Business School. Despite that rapid uptake, U.S. nonfarm labor productivity grew just 2.2% in the second quarter compared to a year earlier, the Bureau of Labor Statistics reported on September 3. Unit labor costs rose 1.2%, driven by a 2.6% increase in hourly compensation.
The gap between adoption and measurable output gains echoes the 1980s more than the internet-fueled 1990s, when productivity surged to 2.7% annually and quarterly rates sometimes topped 4%. During that earlier boom, massive capital expenditure on computers coincided with slowing inflation and rising consumption. Today, wars and supply-chain disruptions are eroding real incomes, leaving less room for a similar consumer-driven payoff.
Part of the problem may lie in how statisticians measure value. Epoch AI, a research institute, estimated last month that annual GDP measurements could understate the value U.S. AI giants capture from data center spending by as much as two percentage points by 2028. The issue arises when products designed in America are manufactured and sold overseas, such as Nvidia chips made in Taiwan and shipped to Europe. Those transactions often escape traditional output metrics.
Data has long struggled to reflect quality upgrades from technology. The baffling lack of a productivity windfall from early 1980s computer investments, famously noted by Nobel laureate Robert Solow, prompted "hedonic adjustments." These mechanically boosted 1990s output and reduced inflation as dollars went further in real terms. Most of that spending surge reflected computers getting 40% faster annually, not companies buying more hardware.
Perhaps services are equally improved by today's large language models, but the statistics cannot distinguish it. That would mean inflation is being overstated. Yet beyond the legal services consumer price index, which the U.S. Bureau of Labor Statistics stopped publishing because of unreliability, there are few suspicious jumps in price growth. Even in wholesale trade and finance, where AI adoption exceeds 60%, inflation tracks worldwide trends.
It is more likely the numbers are roughly correct, and the AI era will be nothing like its internet counterpart, when PCs already sat on every desk. To use large language models, businesses are retraining employees and hiring consultants, according to Federal Reserve surveys. Those costs probably explain why output and employment gains for AI users remain hard to spot.
The investment splurge is nevertheless a direct boon for developers, even as concerns about potential threats prompt calls to throttle chatbot advancements. Dario Amodei, CEO of AI developer Anthropic, on September 12 argued for a slowdown in the development of artificial intelligence, joining a growing number of executives and politicians raising concerns about potential threats from the technology.
Why The Payoff Lags The Hype
The disconnect between AI adoption and productivity statistics has several explanations. First, firms are still in the early stages of integrating the technology. A single contract clause that once took a lawyer hours to research can now be done by a specialized large language model in minutes. Such time savings repeat millions of times daily in offices and factories, but they do not appear in productivity data because the gains are diffuse and often absorbed by other tasks.
Second, the costs of adoption are substantial. Retraining employees and hiring consultants consume resources that offset some of the efficiency gains. Federal Reserve surveys show these expenses are widespread. As a result, the net effect on output and employment remains difficult to detect.
Third, measurement challenges persist. The Epoch AI estimate suggests that by 2028, annual GDP figures could miss up to two percentage points of value created by U.S. AI giants. This undercounting stems from the global nature of technology supply chains, where design happens in America but manufacturing and sales occur abroad.
Historical parallels offer some guidance. The 1980s saw a similar pattern: heavy investment in computers without an immediate productivity bump. It took until the late 1990s for the gains to materialize, aided by hedonic adjustments that better captured quality improvements. Today, those adjustments are already in place, yet the expected surge has not arrived.
Industry executives are now calling for a slowdown in developing the technology. Dario Amodei, CEO of Anthropic, on September 12 argued for a slowdown in the development of artificial intelligence, joining a growing number of executives and politicians raising concerns about potential threats from the technology. Such fears may be overblown, but either way the broad societal payoffs will take time to show up.
Read: AI Spending Slowdown Fears Rattle Investors After Industry Warnings
The investment splurge continues to benefit developers directly. Nvidia, for instance, has seen soaring demand for its chips, even as questions linger about when the broader economy will feel the effects. For now, the productivity revolution remains a promise rather than a measurable reality.
