Apollo’s ‘zero AI payoff outside tech’ claim meets a user-value challenge

On October 24, Apollo chief economist Torsten Slok said AI had delivered no financial payoff outside tech. His charts compare revenue per worker and margins at seven large companies with the S&P 493.

Economist Brad DeLong responded the same day by pointing to benefits reaching users and the creation of new businesses. His response shifts attention from existing companies’ accounts to value retained by users.

The excluded firms are Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla. The other 493 include technology businesses too. Comparing groups does not establish the return on every individual AI project.

Productivity has a different denominator from profit margins. The US Bureau of Labor Statistics’ September 3 revision put second-quarter nonfarm business productivity growth at a seasonally adjusted annual rate of 1.4%, or 2.2% year on year. It measures real output per hour, without isolating AI’s contribution.

Slok’s productivity chart also cites the San Francisco Fed’s total-factor-productivity data. Its October 8 update reports four-quarter growth through the second quarter of 1.42%, versus -0.07% after adjusting for utilization. The series accounts for changes in labor composition and capital inputs, then adjusts for labor effort and capital’s workweek. It is intended as an indicator of technological change, with quarterly-data limitations including incomplete corrections for competition and reallocation effects.

Ramp, whose data underpin Slok’s spending discussion, measures transactions rather than profits. Its current index covers more than 85,000 US businesses; 55.9% paid for an AI product or service in September. Adoption means a positive payment that month. Free tools and employees’ personal accounts can escape this spending-based measure.

In Ramp’s token-management sample, volume in the week ending October 11 rose 52.4% from the comparable week a month earlier; spending rose 23.0%. Ramp says this subset is not representative of its total AI spending. Usage and the bill are separate measures.

How users value AI

DeLong sees user benefits, new businesses and stronger platform incumbents as important channels. He nevertheless remains skeptical about current market valuations. He argues that these channels matter more than expecting a dramatic surge in measured productivity or established companies’ profits.

An April 13 Stanford Digital Economy Lab working paper linked by DeLong measures user benefits through choice experiments. Representative US adult samples were offered compensation for giving up AI chatbots for a month and asked to accept or reject it. The waves ran in July 2025 and March 2026.

For 2026, estimated mean monthly compensation was $124.50 and the median $11.40. Accounting for the user population, the researchers estimated annual consumer surplus at $172 billion. This values benefits to users, separately from corporate revenue or accounting profit. Usage frequency was the strongest predictor of valuation.

Cost governance is another distinct test. A July 29 report by Mavvrik and Benchmarkit surveyed 396 enterprises in April and May. It found 62% reported changing a business decision after unexpected AI costs, and 25% delayed or cancelled an initiative. Only 11% forecast spending within 10%. These are respondents’ experiences across industries, rather than the financial series for Slok’s listed-company groups. Mavvrik sells AI cost-management services.

BLS puts productivity growth since the fourth quarter of 2019 at an annualized 2.1%, matching its long-term rate since 1947. The agency schedules its preliminary third-quarter productivity and costs release for November 5 at 8:30 a.m. Eastern Time.