AI Isn't Cutting Jobs Yet — It's Quietly Cutting Paychecks, and That's the Story Nobody's Watching

AI Isn't Cutting Jobs Yet — It's Quietly Cutting Paychecks, and That's the Story Nobody's Watching

Let me tell you something that's been nagging at me all week. Every headline about AI and work is screaming about jobs. Layoffs here, restructuring there, 166,000 tech workers shown the door in one stretch earlier this year. And every time I read one, I keep thinking the same thing: you're all watching the wrong number.

The first real damage from this AI buildout isn't showing up in payrolls. It's showing up in paychecks. And this week, Apollo Global Management's chief economist put it on paper in a way that should make every founder, every engineer, and every worker with a chatbot open sit up straight.

The Setup — What Apollo's Data Actually Shows

Torsten Slok, Apollo's chief economist, and analyst Sania Edlich ran a difference-in-differences analysis across 321 occupations. They measured which jobs are most exposed to AI using something genuinely clever — actual usage data from Anthropic's Claude. Not a survey of what managers say they'll do someday. Not a consultancy's PowerPoint projection. Real chat traffic from a real AI model, mapped to real occupations.

Then they compared wage growth in the high-exposure jobs against the low-exposure ones, before and after 2023 — the first full year after ChatGPT went mainstream. Bloomberg put Slok on television Friday to walk through the findings, and the line he kept landing on was this: AI is hitting paychecks, not payrolls.

Employment levels in those exposed occupations? No statistically significant change. People kept their jobs. But their pay? That's a different story entirely.

The Numbers — 6.7 Points, 321 Occupations, $28 Billion a Year

Here's the number that matters: wages in high-AI-exposure occupations grew 6.7 percentage points more slowly after 2023 than wages in low-exposure work. Not 0.7. Six-point-seven.

Run that over a couple of years and it compounds into real money. Apollo puts the annual hit to workers at around $28 billion a year. And before you tell me that's rounding error in a $30 trillion economy — no. That's $28 billion that used to land in paychecks, now landing in margins. Every single year, for as long as the repricing runs.

The other part of the finding that should bother you: the squeeze is strongest among lower-income workers. The people who can least afford a pay freeze are the ones getting the deepest haircut. That's not an accident. That's where AI substitutes most cleanly — routine cognitive work, support roles, entry-level analysis. The jobs that used to be the first rung of the ladder, now with a rung missing.

The Two Readings — the Comforting Headline vs the Quiet Transfer

There are two ways to read this, and both are true at the same time — which is exactly why it's dangerous.

Reading one is the reassuring one. AI isn't mass-firing people. Employment is stable. The robots aren't taking the jobs. That's what every tech optimist will quote at you, and technically, they're not lying.

Reading two is the one I live in. Companies don't have to fire anyone to capture the AI dividend. They just stop giving raises. They rewrite the job description, add "AI proficiency" to the requirements, and let the market do the repricing. The productivity gain flows to the balance sheet, not the payroll. That's not a bug in the AI transition. That's the whole point of it.

I've run businesses long enough to know how this works. When a tool makes a worker twice as productive, the question is who captures the value. If labor has leverage, wages rise. If labor doesn't — and right now, labor doesn't — the value flows straight to capital. The Apollo data is just the first clean measurement of that transfer happening at scale.

The Part Nobody's Talking About — the Measurement Blind Spot

Here's the part that keeps me up. Every policy debate, every news cycle, every government program is built around employment numbers. Payrolls. Unemployment rate. Jobless claims. Those are the numbers politicians watch, because those are the numbers that move votes.

None of them measure pay. A worker who keeps their job but loses four years of raises is invisible to the entire statistical apparatus. The unemployment rate looks fine. The payrolls number looks fine. And underneath it, $28 billion a year quietly migrates from workers to shareholders without a single headline.

Worse — even the "payrolls are fine" story is already fraying. July's nonfarm payrolls came in negative — minus 23,000, against a forecast of plus 85,000. The labor market isn't as solid as the "no job losses" narrative suggests. And notice the exact wording of the Apollo finding: no statistically significant employment effect yet. That word is doing a lot of work.

The Secondary Bottleneck — When the Efficiency Dividend Goes to Capital

Now connect this to the bigger picture, because this is where it gets interesting for anyone who runs infrastructure. The AI buildout — the thing I keep writing about — is the most capital-intensive construction project in human history. The three biggest cloud companies on earth are spending 102% of their cloud revenue on capex. The bill for all of it has to be paid by somebody.

We've been watching the debt markets price that risk. We've been watching cloud prices get ready to move. But the wage line is the quietest valve in the whole system. Every dollar of wage compression is a dollar that doesn't have to come out of margins — and the buildout needs a lot of dollars. Wage repression isn't a side effect of the AI economy. It's a funding mechanism for it.

That's the secondary bottleneck nobody's talking about: not jobs, not skills, but the repricing of labor itself. And it's the most politically dangerous valve of all, because workers feel it personally, every month, on a check that's a little thinner than it should be.

What This Means for Founders and Independent Hosting Providers

First — hire the repriced talent. If you run a real business with real revenue, the AI-exposed labor market is handing you senior people at entry-level prices. The best time to poach is when the market is scared.

Second — don't assume the compression lasts. Wage gaps this wide get corrected. Either the market rebalances, or the government gets involved, or the workers organize — and any of those three makes labor suddenly more expensive. Sign multi-year commitments for talent while you can.

Third — build your business on value, not on cheap labor. The founders who win the next decade are the ones who use AI to do more with fewer people — not the ones who underpay the people they kept and hope nobody notices.

Fourth — watch your customers' customers. If your clients serve lower-income consumers, that $28 billion annual squeeze is coming out of their revenue. Budgets for hosting, software, and services get cut from the bottom first.

Fifth — get ready for the political bill. When the wage data keeps softening while the stock market keeps setting records, that gap becomes a political weapon. Data center moratoriums, AI taxes, labor mandates — the backlash has a million shapes, and every one of them eventually lands on infrastructure operators.

The Bottom Line

Stop watching payrolls. Start watching paychecks.

The AI buildout isn't coming for your job — yet. It's coming for your raise. And the quiet transfer happening right now, six-point-seven points at a time, is the story that actually explains where this economy is going: the value created by artificial intelligence is being collected by capital, and distributed to almost nobody else.

That's not a prediction. It's a paycheck.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: Apollo Global Management (Torsten Slok and Sania Edlich), Bloomberg Television, The Next Web, Crypto Briefing, U.S. Bureau of Labor Statistics.