The AI Payoff Split: Flat Returns Outside Tech, a Market Wobble, and Open Models Gaining Ground

The flat 493

Apollo Global Management's read on the AI economy is blunt: revenue per worker is rising for the Mag 7 but flat for the S&P 493, and profit margins follow the same split, rising for the Mag 7 and flat for the rest. Apollo's argument is that there are still no signs of AI boosting productivity growth, and there is no evidence so far that revenue growth or margins are improving outside the tech sector. m:…1236

The spending behavior matches the split. Mainstream corporate America is pulling back on AI spending, Apollo says, starting to view token spending not as a source of growth but as a cost center that weighs on margins, and mainstream companies do not need the latest model and are unwilling to pay for it. VC-backed companies, by contrast, are spending more and more on tokens, which keeps the headline AI spend figures strong while the buyer base narrows. m:…1236

Apollo's explanation reaches beyond price. AI does not understand accountability, trust, culture, unwritten rules, ethical trade-offs, or fairness, Apollo argues, and those are the factors that shape reputational risk and ultimately drive revenue growth and profit margins. As long as agentic AI and token spending are confined to a small part of the economy, the macroeconomic effects of AI will be very limited, and the AI story will be decided in the S&P 493, not the Mag 7. m:…1236

Markets flinch

Equity markets gave the debate a live backdrop on Thursday. The S&P 500 lost 1.2 percent and the Nasdaq composite fell 2.6 percent, according to Emirates NBD research, while both the 2-year and 10-year US Treasury yields fell by just under 17 basis points, leaving the 10-year yield below 5 percent for the first time in roughly a month. The same session brought a different kind of policy move: the Central Bank of the Republic of Turkiye cut its policy rate by 100 basis points to 36 percent, its first reduction since January, as annual inflation eased to 29.7 percent in September. m:…4048

Other Thursday data points cut against a simple risk-off story. US jobless claims fell by 6,000 to 194,000 in the week ending October 17, below expectations and reaching a three-month low, and Brent crude rose 4 percent to settle at 113.17 dollars per barrel on Thursday as Wednesday's Houthi strikes on Saudi Arabia heightened concerns about energy supplies. The session is best read as a repricing in progress, not a verdict: equities down, yields down, and policy easing elsewhere all at once. m:…4048

The counter-evidence from accounting

The counterpoint to the flat-493 argument comes from a narrower, professional test. On DualEntry's Accounting AI Benchmark, Moonshot AI's Kimi K3.1 Pro debuted at number 4 with 86.1 percent accuracy across accounting tasks, 1.9 percentage points higher than its predecessor Kimi K3 at 84.2 percent. Its strongest category was reconciliation at 92 percent, followed by reporting at 85 percent, while it scored 74 percent on hard tasks against 90 percent on easy tasks. m:…1464

The leaderboard around it is what makes the result interesting for buyers. Google's Gemini 4 Argon ranks first at 91.1 percent and Zhipu AI's GLM-5.3 Flash ranks second at 88.1 percent, while Kimi K3.1 Pro at 86.1 percent sits ahead of Claude Opus 5.5 at 82.2 percent, GPT-6 Astra at 81.2 percent, and Claude Sonnet 5.5 at 80.2 percent. The benchmark tests 101 synthetic accounting cases across eight areas, with each model evaluated three times using the same test company, tools, and instructions, so the comparison is controlled even though the cases are synthetic. m:…1464

Put together, the two stories describe different questions. Apollo asks whether AI is lifting the broad economy, and answers that it is not yet visible in the S&P 493. DualEntry asks whether a specific model can do a specific professional job, and answers that open models are now competitive on that job. Both can be true at once, and the gap between them is where the next investment debate will sit: capability in a benchmark is not the same as margin in a business. m:…1236 m:…1464