Enterprise AI Spending Shows Zero Payoff Outside Big Tech

The growing divide between the Magnificent Seven and the S&P 493

Only the hyperscalers selling computing hardware, cloud instances, and foundation models are recording concrete financial gains from the boom.

Torsten Slok, chief economist at private equity giant Apollo Global Management, highlighted a stark divergence in public financial data. While revenue per employee continues to rise sharply across the tech giants—Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla—the remaining 493 companies in the S&P 500 show flat productivity and static operating margins.

Silicon Valley software vendors continue pitching automated workflows, such as managed agents designed for enterprise pipelines, but regular businesses have yet to see those integrations translate into actual income or overhead reductions.

Without concrete returns on balance sheets, mainstream corporate buyers are declining to renew expensive experimental contracts, viewing consumption-based token billing as an unnecessary tax.

Why generative tools hit a wall in regular business

The fundamental issue stalling wider adoption is that typical corporate functions rely heavily on nuanced organizational context, accountability, and liability. Language models can summarize documents or generate marketing drafts, but they cannot assume responsibility when mission-critical operations fail.

As Slok pointed out, business operations depend on ethical trade-offs, trust, and unwritten cultural rules. When an automation pipeline errs, the cost of auditing and remediating the fallout quickly wipes out whatever labor hours were saved.

This dynamic has turned pilot projects into dead ends. Venture-funded startups may keep increasing their API consumption, but traditional industrial firms, healthcare systems, and consumer brands are drawing a hard line on IT overhead.

What it means for enterprise IT budgets

For CIOs and procurement leaders in the US and Europe, the pressure has shifted from fear of missing out to showing immediate return on investment. Expect fewer open-ended pilots and strict requirements that software vendors demonstrate measurable labor savings before securing enterprise-wide contracts.

As infrastructure providers look to monetize workflows further—such as pricing enterprise agent software on per-chip licenses—non-tech buyers will likely stick to legacy software stacks until foundation model costs fall by another order of magnitude.

The broader economic transformation promised by artificial intelligence cannot run indefinitely on Big Tech selling infrastructure to other tech firms. Until ordinary corporations in retail, manufacturing, and services extract real balance-sheet value from automated workflows, macroeconomic growth will stay pinned.