AI Is Suppressing Wages, and the 10-Year at 4.85% Has Already Priced It

BLS data through Q2 2026 marks the cleanest break yet between AI-exposed and AI-light workers. The Employment Cost Index shows inflation-adjusted wages and salaries fell 0.4% year over year through June, while labor's share of nonfarm business output dropped to 52.8% — the lowest reading since the series began in Q1 1947. Productivity rose 1.4% annualized in Q2, but unit labor costs rose 1.2% alongside it, meaning the productivity gain flowed straight to gross operating surplus rather than to workers.

The 6.7-point Apollo gap is not a hiring signal — it is a bargaining signal. Workers stay employed; their wage curve flattens. That mechanism is the through-line tying the bond market's hawkish lean to the [Federal Reserve]'s rate path. Goldman Sachs Research, in an Aug 19, 2026 analysis, estimated AI substitution has been a net drag of roughly 16,000 U.S. jobs per month over the past year (25,000 substituted, 9,000 augmented), with the pain concentrated in Gen Z and entry-level white-collar roles.

MIT's Daron Acemoglu told CNBC he expects wages — not employment — to be where AI's effects first become politically visible, given labor-market flexibility and a thin social safety net. [Anthropic] CEO Dario Amodei has warned that 50% of entry-level white-collar roles could disappear within 1–5 years. The Challenger, Gray & Christmas July report counts 149,023 announced tech-sector cuts in 2026, up 67% year over year, with AI named as the #1 cited reason for the fifth straight month.

The 10-year Treasury's grind from 4.18% in early August to 4.95% on Sep 9 and an intraday 4.99% on Sep 11 looks, at first glance, like a growth scare. It is not. The 2-year has done most of the work — it jumped 24 bps in a single week per the Sep 11 CME FedWatch read — which is what happens when the market moves the front end, not the term premium. The 30-year mortgage at 6.76% is the cleanest consumer-level evidence that 4.85% on the 10-year is already in the real economy.

What changed the curve was Fed Chair Kevin Warsh's Jackson Hole speech on Aug 28, 2026, in which he pledged to return inflation to 2% and explicitly said rates could rise further. Bank of America Global Research now expects three 25 bp hikes in 2026, starting in September. The Reuters poll of 93 economists, run Sep 4–9, still shows 70% expecting a hold at the Sep 15–16 meeting, but that share is down from 90% a month earlier — and the 85.5% implied probability on CME FedWatch reflects how quickly the front end has caught up to the speech.

Markets keep treating 4.85% as a 'growth premium' on the 10-year. The more honest read is that it is a labor-share premium. With labor's share at a 79-year low and real wages negative, the Fed faces a textbook stagflationary bind: services inflation stays sticky because unit labor costs keep rising even as paychecks stagnate, while goods inflation pressures from the AI capex buildout (chip demand, power, data-center construction) push the other way. Apple CEO Tim Cook warned on Sep 12 that memory-chip-driven price hikes are unavoidable; Tesla CEO Elon Musk agreed. That is a third-derivative input into the same curve.

For bond investors, the actionable signal is that 4.85% is a regime, not a stop. The 10-year cannot rally on a soft CPI print alone; it needs either a real-wage re-acceleration (which AI is currently preventing) or a clean unemployment break higher (which the Aug 162k payroll print pushes further out). Long duration is therefore the wrong expression of stagflation fear. The right expressions are short-duration credit, the 2s/10s steepener if the front end moves first, and the equity book where the wage-suppression thesis translates directly to margin expansion — large-cap tech, IT services that have already taken their AI-restructuring hits, and banks that are re-platforming headcount around AI rather than just cutting it.