Enterprise AI gains stall outside the tech sector

Apollo chief economist Torsten Slok reports that revenue per worker and profit margins have climbed for the largest tech companies, but remain flat across the rest of the S&P 500. Total factor productivity data from the San Francisco Fed shows no lift either. Outside of software companies buying API access from each other, corporate finance teams increasingly treat model token spending as an operational cost that weighs on quarterly margins.

The split makes sense when you examine how work happens in production. Frontier labs can sell tokens to developers building prototypes, but a manufacturer or a retailer has to justify that expenditure with faster cycle times or reduced headcount. Generative models drop into existing enterprise workflows as conversational assistants, not autonomous operators. Because a model cannot take accountability for business errors, human reviewers must check every output. That keeps the labor cost in place while adding cloud inference bills on top of it.

This changes what enterprise engineering teams need to prioritize. Upgrading to larger frontier checkpoints will not fix the financial return when the friction is integration and verification. Companies outside tech will likely get more value from smaller, domain-specific models hosted on their own infrastructure, where inference costs are predictable and boundaries are tight. Until agentic systems can reliably finish multi-step tasks without human babysitting, CFOs will continue to see generative tools as discretionary expenses.