The numbers lie.
Capital is flowing into silicon at a record pace. Labor is stalling in the crosswinds of automation. The gap is the danger. Morgan Stanley Chief Economist Seth Carpenter recently signaled a warning that the market is currently ignoring. The central question is no longer if machine intelligence works. The question is whether the global economy can rewire its fundamental structures before the labor market fractures. We are witnessing a massive capital reallocation that has yet to show up in the bottom line of national accounts. The promise of an AI productivity boom remains a ghost in the machine.
The structural lag of total factor productivity
Technology does not translate to growth instantly. History proves this. The electrification of factories in the early twentieth century took decades to move the needle on output. We are seeing a repeat of the Solow Paradox. You can see the computer age everywhere but in the productivity statistics. Per recent data from the Bloomberg Terminal, corporate expenditure on AI infrastructure has surged 42 percent year over year. Yet, real GDP per hour worked remains stubbornly tethered to its pre-2023 trend line. This suggests that firms are spending on the capability of efficiency rather than the realization of it.
The friction is human. Integrating large language models into legacy workflows requires more than a software license. It requires a complete overhaul of management hierarchies. Most firms are simply layering expensive tech on top of broken processes. This creates a drag. It increases operational expenditure without a commensurate reduction in headcount or an increase in throughput. The result is a margin squeeze that the equity markets have yet to price in fully.
Visualizing the divergence of investment and output
The following data represents the growing chasm between capital expenditure in AI and the actual productivity gains recorded across the G7 economies as of May 2026.
AI Investment vs Productivity Growth Index (2023-2026)
The labor market shock is localized
Aggregate unemployment figures are deceptive. The headline rate remains low, but the churn beneath the surface is violent. We are seeing a hollowing out of mid-level cognitive roles. Analysts, paralegals, and junior coders are being displaced by automated agents. This is the labor market shock Seth Carpenter referred to in his latest Thoughts on the Market. The economy is not losing jobs in the traditional sense. It is losing career paths.
The transition is not seamless. Displaced workers cannot move from data entry to AI ethics overnight. There is a skills mismatch that is widening the wealth gap. High-skill engineers are seeing wage inflation, while the median white-collar worker faces stagnation. This creates a two-tier economy. One tier builds the automation. The other tier is optimized by it. The political ramifications of this divergence will likely dominate the back half of this decade.
Sector adoption and displacement metrics
The impact of this technological shift is not uniform. Some sectors are absorbing the shock better than others. The table below breaks down the efficiency gains versus the displacement rates observed in the first quarter of this year.
| Industry Sector | Efficiency Gain (%) | Displacement Rate (%) | Net Wage Growth (%) |
|---|---|---|---|
| Financial Services | 18.4 | 12.1 | +4.2 |
| Manufacturing | 6.2 | 4.5 | +1.8 |
| Healthcare Admin | 22.1 | 15.8 | -0.5 |
| Software Development | 35.7 | 8.9 | +9.4 |
| Customer Support | 41.2 | 28.3 | -12.2 |
Customer support is the canary in the coal mine. The 41 percent efficiency gain is staggering. It is also a death knell for entry-level service roles. This sector shows the most extreme version of the productivity boom. It is a boom for the corporation, but a shock for the employee. The Reuters April Payroll Report highlighted that while the service sector added jobs, the quality and pay of those jobs have shifted toward manual labor that cannot yet be automated. This is a regression, not a progression.
The cost of compute vs the value of output
Energy is the hidden bottleneck. The productivity boom is being throttled by the power grid. Training and running these models requires a level of electricity consumption that is driving up utility costs for the entire industrial sector. If the cost of the energy required to run the AI exceeds the value of the labor it replaces, the net productivity gain is zero. We are approaching that threshold in several European markets. The efficiency of the algorithm is being canceled out by the inefficiency of the infrastructure.
Investors are currently valuing companies based on their AI potential. They should be valuing them based on their AI utility. A company with a massive GPU cluster but no change in its operating margin is just a company with an expensive hobby. We are entering the era of accountability. The hype cycle has peaked. The execution cycle has begun. The next data point to watch is the June 15 release of the revised productivity figures for the second quarter. If we do not see a meaningful uptick in output per hour, the market’s faith in the AI miracle will likely break.
