Nearly two-thirds of S&P 500 companies mentioned AI during their Q3 quarterly analyst calls.
Around half indicated they’re working to leverage AI to improve productivity and efficiency.
But “only a handful,” to quote Goldman’s David Kostin, were able to “discretely quantify” the impact of AI on profitability.
That’s the sort of disparity which has some observers nervous: Pretty much everyone’s talking about AI, and almost everyone’s spending money on it one way or another, but almost nobody can say, when pressed, how (or even if) the technology’s contributing to the bottom line.
The figure on the left, below, shows you what Jensen (and Sam) hath wrought, so to speak. Before 2023, just one in 10 S&P 500 companies bothered mentioning AI on earnings calls. In Q3 2025, that figure was more than six in 10, and I gotta tell you: Some of those discussions are forced, which is to say management teams increasingly view AI adoption as compulsory.
The figure on the right’s marginally useful, even as it merely affirms the intuitive: The most oft-cited use cases for AI in the context of improving productivity are coding and customer support.
In his latest, Goldman’s Kostin drew a distinction between companies which “quantified the AI-related productivity uplift for specific use cases” and those which were unequivocal and direct about the benefits which accrued to the bottom line. Long story short, the former were many, the latter were few where “few” means two out of 500.
“About 8% of S&P 500 firms cited the AI-driven improvement in alternative metrics including the amount of code being written by AI, the share of customer support interactions handled by AI agents and the number of documents processed by AI,” he wrote, before cautioning that the relatively large increase in profitability observed for many of those companies “may reflect a general focus on efficiency or unrelated coincident business dynamics rather than reflecting the specific impact of AI adoption” given that those firms had lower margins versus the median company prior to the last few quarters.
“Most” of those firms, he went on, “have not explicitly quantified the profit uplift from AI adoption,” nor have virtually any other firms for that matter. The two exceptions in Q3 calls were ServiceNow and CH Robinson, both of which specifically cited AI in tallying better operating leverage and lifting their profitability guidance.
Notably, AI-related stock gains remain concentrated almost entirely in infrastructure names. The figure below’s remarkable in that regard.
Since Nvidia’s Q2 2023 earnings release (the “report heard ’round the world”), Goldman’s AI infrastructure basket has bested the equal-weighted S&P by an absurd margin, while an index designed to include stocks seen as benefiting from AI-enabled revenues has managed comparatively little in the way of outperformance.
Tellingly, Goldman’s AI productivity basket has underperformed over the period. That poor showing reflects “uncertainty surrounding the distribution, magnitude and timing of AI-related productivity gains,” Kostin remarked, flatly.
Finally, someone has figured out that for AI users (actual customers of the product itself, rather than its processing hardware sales) there are few if any sustainable business models showing actual profit and revenue growth for users. (And the job users do in implementation creates increasing numbers of errors.) Productivity improvement is not actually a revenue related metric. Rarely, if ever, are sales increased by rising productivity. AI use may enhance the prospect of cost reductions in operations but most if not all of those productivity gains will harm labor, consumption and economic growth. So where’s the growth in revenue and real profit? Little if any yet. Google, for one, has figured out that by owning virtually all the AI powered advertising market it can increase its revenue and profits, at the risk of being broken up as a monopolist.
“The most oft-cited use cases for AI in the context of improving productivity are coding and customer support.”
He also might have accurately phrased it “The most oft-cited use cases cited by Wall Street analysts and industry figures are coding and customer support.” More often than not, those successes are extrapolated to enterprises across the broader economy as opposed to a limited subset of industries.
The most cited use cases are more evidence that the powers that be at most companies don’t realize how much time their minions waste putting together useless presentations or content so they can justify their own jobs to board members. AI has been great for me to pull together their BS presentations in much less time since it’s mostly just an exercise in making things sound good and AI is well suited to that.
OMG – You must hope and pray that “the powers that be” DO NOT see this post. AT best it will be a convenient excuse to burden you with more work. At worst, it will be the judication for firing you and replacing you with a much-lower cost young employee since, “after all, AI is doing all the work we’ve been giving him!”
This is America where every person and every product much pay its own way. That includes AI.
AI talk is quasi obligatory now, as was ESG talk and diversity talk previously. My sense is that companies are no longer getting price reaction boost from AI talk on the earnings call, so perhaps we’ll see less of it going forward. Or not. Many large public companies have margin incentives to RIF, and dressing it up as AI efficiency sounds good.