Apollo Global Management concluded that corporate America outside seven technology giants has yet to see a financial return from artificial intelligence. Its analysis, published on Saturday, separates the Magnificent Seven from the "S&P 493" and finds flat revenue per worker and profit margins for the rest. For the seven leaders, revenue per employee rose from $550,000 in 2022 to almost $800,000 this year. The finding matters because it shifts the test of AI value from model progress to profit growth outside Big Tech.
What Apollo measured in the S&P 493
Analyst Torsten Slok based the verdict on three indicators: revenue per worker, profit margins and productivity growth. The Magnificent Seven in the study are Alphabet, Amazon, Apple, Meta Platforms, Microsoft, Nvidia and Tesla. Their curves moved sharply upward after AI adoption accelerated, with margins rising since the third quarter of 2023. By contrast, the S&P 493 figure for revenue per worker has lingered near $550,000 since 2019, with no lift from AI spending.
Slok described mainstream companies as starting to treat token spending as a cost center that pressures margins rather than a source of growth. The distinction rests on how intelligence is used inside a firm. AI handles calculation, research and generation well, but business results depend on judgment, where context changes the answer. That gap explains why buying access to models does not automatically convert into higher sales or lower unit costs.
The result runs against predictions of an coming "age of abundance" with less work and faster output for all. Slok noted that some venture capital-backed startups keep raising spending on AI tokens, while established firms pull back. Mainstream corporate America, in his wording, does not need the latest model and is unwilling to pay for it. As long as agentic AI and token use stay confined to a narrow segment, the macroeconomic effect stays limited.
What this means for companies adopting AI
For operating businesses, the message is to link every AI budget line to revenue or margin, not to experimentation alone. Large firms can test agents in support, sales operations and document processing, then keep only cases where output per employee rises. Small firms gain more from narrow tools with clear pricing than from broad platform commitments, since they lack spare capacity to absorb token costs. Vendor selection then turns on measured throughput and error rates in a specific workflow.
The limits named by Slok deserve direct checks before scaling: understanding of business culture, accountability for decisions, ethical trade-offs, fairness and trust. These factors shape reputational risk and, through it, sales and profit. The analysis does not prove that AI can never raise productivity outside technology; it shows no evidence so far in aggregate data. Buyers should ask suppliers how judgment is reviewed, who signs off on agent actions, and where costs stop growing when usage rises.
The marker to watch is the S&P 493 itself: revenue per worker and profit margins in quarterly reports through 2026. If cheaper models start lifting those lines, the payoff phase has begun. If the lines stay flat while token bills grow, discipline on AI budgets will tighten further.
