The AI Productivity Gap: Why Markets Face a ‘Painful Repricing’ as ROI Lags

The global markets are approaching a critical juncture where the hype surrounding artificial intelligence must finally translate into balance-sheet reality. Torsten Slok, the chief economist for Apollo Global Management, warns that a “painful repricing” is imminent if the current mismatch between aggressive earnings expectations and the actual time required for firms to generate a return on investment (ROI) is not resolved.

While the tech sector has successfully integrated AI to bolster margins, the broader economy remains stuck in a holding pattern. Data from Bloomberg and Macrobond highlights a stark divergence: profit margins for the so-called Magnificent Seven surged from roughly 15% to 25% between early 2023 and 2024, yet the remaining S&P 493 have seen margins stagnate at approximately 10%. This gap suggests that structural productivity gains are not yet a universal reality.

The delay in broader adoption stems from significant regulatory hurdles, data protection requirements, and the sheer complexity of workflow integration. According to Slok, the market has priced in returns much sooner than the “ROI runway” outside the tech sector allows. This observation aligns with historical trends monitored by the U.S. Bureau of Labor Statistics, which show that transformative technologies often require years of organizational restructuring before they manifest in national productivity data.

Evidence of this friction is appearing in the industrial heartland. Ford recently recruited 350 “gray beard” engineers—industry veterans and former employees—to provide human oversight for AI tools that proved ineffective on their own. Despite deploying AI vision systems across 33 plants, the automaker found that the technology lacked the necessary nuance without the experience of senior staff. Charles Poon, Ford’s vice president of vehicle hardware engineering, noted that the tools are only as effective as the information used to train them, admitting the company had previously undervalued the institutional knowledge of its most experienced personnel.

This reliance on human expertise challenges the narrative of immediate, cost-saving automation. In many cases, human labor remains more cost-effective than the high-priced automation tools currently being marketed. Bryan Catanzaro, Nvidia’s vice president of applied deep learning, has acknowledged that the cost of AI still significantly exceeds that of human workers. This cost-benefit imbalance has led to a phenomenon Slok describes as tokenmaxxing, where companies incentivize AI use through internal leaderboards simply to justify the investment, often driving up costs without achieving strategic goals.

A study from the MIT found that only 5% of companies realized a meaningful return from generative AI pilot projects last year. Peter Cappelli, a professor at the Wharton School, argues that leadership often focuses on what is possible rather than what is practical. In one case study involving the digital services firm Ricoh, an attempt to automate insurance claims resulted in costs three times higher than manual processing due to consultant and AI fees. While the division eventually saw a three-fold increase in productivity, the transition was neither cheap nor rapid.

The pressure to demonstrate AI competency has also birthed “AI shame,” a corporate environment where managers feel compelled to mandate technology use without articulating clear business objectives. A Boston Consulting Group Global AI at Work report found that while some employees saved up to eight hours a week using AI, half of them received no guidance on how to repurpose that time for strategic work. Slok suggests that the current shift toward token optimization is an early warning that the road to AI implementation will be significantly bumpier than investors currently anticipate.

If firms do not see ROI quickly, Slok predicts a sharp deceleration in spending. This potential retreat poses a direct threat to the valuations of companies currently leading the AI boom, as the market begins to realize that the Jevons paradox—where technological efficiency leads to increased demand and job creation—may take far longer to manifest than the current trading cycle suggests.