Apollo study questions whether AI’s economic benefits have reached beyond big tech companies

AI consulting firm Apollo believes there is still little clear evidence that AI’s economic gains have spread beyond the largest technology companies. In a report released on Saturday, 25, economist Torsten Slok compares the evolution of revenue per employee and margins of the so-called Magnificent Seven with those of the rest of the S&P 500 companies.

The charts reveal a gap between the two groups. This is an interpretation of corporate indicators: the comparison does not directly measure the return on each AI investment nor does it demonstrate that the technology is solely responsible for the difference.

In addition, the universe analyzed has limitations. The group of the other 493 companies includes some technology firms, so it does not represent all economic activity outside the sector.

Why revenue per employee doesn’t settle the debate

Revenue per employee is not equivalent to production per hour worked. The former divides revenue by the number of people; the latter relates production to hours. Changes in these indicators can have multiple causes, which requires caution when attributing them to a specific technology.

Data from the Bureau of Labor Statistics show that productivity in the non-agricultural business sector of the United States grew by 1.4% in the second quarter of 2026, at an annualized rate. Compared to the previous year, the increase was 2.2%. The statistic does not indicate how much of this progress is attributable to AI.

The Federal Reserve Bank of San Francisco also records an acceleration in production per hour since 2023, but leaves open the question of AI’s contribution. In a September analysis, the institution compared an annualized average of 2.5% between the first quarter of 2023 and the second quarter of 2026 with 1.5% between 2005 and 2022.

The difficulty of separating investment from results is already present in the debate about how companies measure the return on AI.

Lower spending can also reflect efficiency

The Ramp AI index monitors payments from more than 85,000 corporate customers in the United States. In this survey, the adoption of a tool is considered when there is a monthly payment; free uses or personal accounts may be excluded from the measurement.

In the October letter, economist Ara Kharazian linked the slowdown in spending to prices and efficiency. In a subset of customers using the token management service, the volume processed grew more than the expense between the weeks ending September 13 and October 11. The sample does not represent the entire market.

Stable margins do not rule out profitable projects. To evaluate the return, it is necessary to relate project costs and results, rather than concluding that all AI use has failed based on a corporate average.

Frequently Asked Questions

• Does the analysis prove that AI does not provide a return?

  No. The comparison made by Apollo does not isolate the individual result of each project nor does it establish a cause-and-effect relationship.

• Does less spending mean less usage?

  Not necessarily. Lower prices and greater efficiency can reduce spending even with a higher volume of processing, as Ramp notes.

• Does the official productivity data measure only AI?

  No. The BLS indicator measures the relationship between production and hours worked in the sector analyzed.