📰 Key Takeaways

In a new research note, Apollo Global Management chief economist Torsten Slok says AI hasn’t delivered any profit growth to large companies outside the tech sector. Apollo is the private equity giant that once owned TechCrunch. Slok’s analysis focuses on what he calls the “S&P 493,” which is the S&P 500 minus seven companies: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla. He found that profit margins and revenue per employee for these companies both look flat, with no evidence that AI has improved productivity.

Slok put it bluntly: despite massive spending and heavy market hype, AI has “delivered zero financial payoff” to US companies outside the tech sector. As for whether this is just a lag while non-tech companies catch up to the tech giants, Slok is skeptical. He thinks it’s hard for US companies to turn increasingly cheap intelligence into higher revenue growth and better margins.

He went on to explain that AI may be a great computing tool, but in business, intelligence is largely about judgment. He noted that AI can’t understand accountability, trust, company culture, unwritten rules, ethical trade-offs, or fairness, and those are the things that drive reputational risk and ultimately shape revenue growth and profit margins. The original article also notes that other recent studies have likewise found no significant productivity gains from AI.

💬 JudyAI Lab Take

Apollo chief economist Torsten Slok’s research finds that in the “S&P 493” (the S&P 500 after removing the seven big tech giants), neither profit margins nor revenue per employee have improved, so AI hasn’t yet delivered profit growth to large companies outside the tech sector. That conclusion is a sharp contrast to the excitement in the market, and it’s worth looking at calmly.

For AI builders, this story is a reminder about design thinking: cheaper compute and intelligence don’t automatically mean better business outcomes. Slok argues that intelligence in the business world is largely judgment, and things like accountability, trust, company culture, unwritten rules, ethical trade-offs, and fairness, all of which affect reputational risk, are things AI can’t understand right now. In other words, where many adoption projects get stuck may not be whether the model is strong enough, but whether the product actually designs those judgment steps into the workflow. The original article also notes that other recent studies haven’t found significant productivity gains either, so this isn’t just one person’s view. As for whether it’s only a lag, Slok himself stays skeptical, so we don’t need to rush to make excuses for AI.

Next time you evaluate an AI feature, write down which part of the work it replaces as “computation” and which part is still “judgment” that a human has to be accountable for. Then validate it with measurable metrics like profit margin or output per person, instead of just looking at usage.

📅 Source Info

🔗 Further Reading