Apollo's chief economist sees no AI payoff outside big tech yet. What that means for your AI budget

What Slok said

The Daily Spark note compares the "Mag 7" (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla, as TechCrunch lists them) with the rest of the index. Revenue per worker and profit margins are rising for the seven, Slok writes, but flat for the other 493. A third chart uses the San Francisco Fed's total factor productivity series, a measure of output growth not explained by more labour or capital, and Slok says it still shows no AI boost.

His reading of company behaviour goes further: "mainstream corporate America is pulling back and starting to view token spending not as a source of growth but as a cost center that weighs on margins." In charts built on data from the fintech company Ramp, he adds that venture-backed companies are spending more and more on tokens, while mainstream firms do not need the latest model and are unwilling to pay for it.

On whether that will change, he is sceptical: turning cheaper AI into revenue and margin growth "may be harder than it sounds". "AI may be a great calculator, but in business, intelligence is mostly judgment." He argues that AI does not understand accountability, trust, culture or fairness, which shape reputational risk and, in the end, revenue and margins. TechCrunch, which reported the note on 25 October, adds that recent studies have also found AI has yet to deliver significant productivity gains.

What the note does and does not show

This is a short market commentary with five charts, not a study of individual firms. It covers large US-listed companies in aggregate. It says nothing directly about small businesses, the UK or the EU, and an index average can hide companies that gained from AI alongside others that spent and saw nothing.

Flat margins also have more than one possible cause, from interest rates to wage costs, and the note does not separate AI's effect from those. Those limits are Evoogen's reading of the note, not points Slok makes. What his charts do show is that, in aggregate, any gains for the S&P 493 are not yet visible in revenue per worker or margins.

A worked example: when AI spend pays for itself

Here is a hypothetical with stated assumptions, to show how quickly the maths can go either way. A five-person office pays £150 a month for AI subscriptions. If the tools save each person 30 minutes a week, that is about 10.8 hours a month (5 people × 0.5 hours × 4.33 weeks). At a staff cost of £15 an hour, that time is worth about £162.

On paper the tools just about pay for themselves. In practice they only do if the saved half-hours turn into billable work, faster replies or fewer missed enquiries. If the time disappears into the day, the £150 is the kind of cost line Slok describes. The same subscription aimed at one measured task, such as answering missed calls or drafting quote follow-ups, can be checked against a before-and-after number.

What this means for a smaller business

Evoogen's reading: the note supports picking AI projects by the result they should move, not by the model they use. Before paying for a tool, write down the number it should change: hours spent on a task, response time, enquiries answered out of hours. Measure it for two weeks before and after, and set a date to cancel if the number has not moved.

It is also a reminder that the newest model is rarely the point. Slok observes that mainstream firms do not need the latest model and will not pay for it. Our view is that for routine office work the gain is more likely to come from fitting AI into a specific workflow with clear handover to a person than from paying more per token.