Using a standard accounting decomposition of nonfinancial corporate value added, the Fed authors find that the majority of the post-2020 labor share decline traces to a widening markup wedge, with capital share roughly flat and the residual small. They explicitly rule out capital deepening, compositional shifts toward capital-intensive sectors, and a productivity surge outrunning wages as the dominant explanations.
The editorial argues that two-thirds of the cumulative decline in labor share since the 1970s arrived after 2020 — a five-year move that consumed roughly half a century of slow erosion. Because the traditional explanations are slow-moving variables, they cannot account for a cliff-shaped step-down concentrated in such a short window.
If AI and automation were driving wage stagnation in the 2020s, capital share would be rising as firms substitute machines for workers. Instead capital share is flat and markups are widening, suggesting the labor share story is about pricing power and corporate profits rather than technological displacement of workers.
By submitting the Liberty Street Economics post and driving it to 295 points with 243 comments, the submitter signals that the developer/tech community views the Fed's markup-centric framing as a meaningful counterweight to the prevailing AI-displacement narrative shaping tech labor debates.
The New York Fed's Liberty Street Economics blog just published a decomposition of the US labor share of income showing it has fallen to its lowest level since the post-war series began. The piece — written by the Fed's research staff and posted under the title *The Post-COVID Decline in the Labor Share* — pulls apart the nonfinancial corporate business sector's accounts and concludes that the drop concentrated in the post-2020 period is mechanically attributable to expanding profit markups, not to capital deepening, not to a compositional shift toward capital-intensive sectors, and not to a productivity surge that out-ran wages.
The headline number: labor compensation as a share of nonfinancial corporate gross value added is now below every reading recorded between 1947 and 2019, and roughly two-thirds of the cumulative decline since the 1970s arrived after 2020. That's a five-year move that ate roughly half a century of slow erosion. The post-COVID step-down is large enough that earlier explanations — China shock, automation, declining union density, skill-biased technical change — can't carry it on their own, because those are slow-moving variables and this is a cliff.
The Fed authors use a standard accounting identity: labor share = (wage bill) / (value added), and value added decomposes into labor cost, capital cost, and pure profit (the markup wedge). Hold productivity and capital fixed, and the only way labor share falls sharply is if the markup wedge widens. Their estimate is that markup expansion accounts for the majority of the post-2020 decline, with capital share roughly flat and the residual small.
This is the Fed's own research arm pushing back, gently, on two stories that dominate tech discourse. The first is the AI-displacement thesis — the idea that the 2020s wage stagnation is the leading edge of a software-eats-labor wave. If that were the main driver, you'd see capital share rising (the robots get paid) and labor share falling. You don't. Capital's slice is approximately where it was in 2019. What rose is the residual — the wedge between price and marginal cost — which in plain English means firms charged more without paying inputs more.
The second story this complicates is the supply-shock narrative: that 2021–2023 inflation was a transient pass-through of energy, shipping, and chip shortages. Markups don't snap back the way input costs do. Once a firm discovers it can hold a 2022 price into a 2026 cost environment, the gap becomes operating leverage, and the labor share absorbs the difference. The Fed paper is careful not to make a causal claim about *why* markups widened — concentration, demand composition, post-pandemic pricing power, or genuine product differentiation are all candidates — but it does say the residual is real and persistent.
The macro reading-room is going to argue about this for months. Jason Furman has been making a version of the markup case since 2023; Larry Summers leans on supply-side stories; Olivier Blanchard splits the difference and emphasizes labor-market mismatch. This NY Fed post lands closer to Furman than to either of the others, which is notable given the Fed's institutional preference for supply-and-mismatch framings during the active inflation episode. Posting a markup-centric decomposition under the Liberty Street masthead is a signal that the in-house view has shifted.
For the AI-capex debate specifically, this matters because the Mag 7 vs. S&P 493 divergence Torsten Sløk flagged this week — and the broader question of whether $300B in datacenter spend is funded by genuine productivity gains or by margin expansion at the top of the corporate distribution — runs through this exact decomposition. If the answer is *margin expansion*, then the AI capex cycle is being financed by the same wedge that's compressing labor's slice, not by efficiency dividends that will eventually flow back to wages.
Three practical implications, in descending order of how directly they hit a developer's life.
Comp negotiations have a macro tailwind, not a headwind. If markups, not automation, are eating labor share, then the standard tech-worker fear — 'AI is about to commodify my skill set and crash my wage' — is at least premature. The mechanism currently compressing aggregate wages is pricing power, which is sector-specific and concentrated. Software engineering is not, on the firm-level data, a sector where the median company has pricing power over its labor inputs; FAANG-tier compensation is set by talent competition, not by a national wage anchor. Read the Fed paper as evidence that the squeeze is happening upstream of you, in product markets, not in your specific labor market.
The AI productivity story needs receipts. Every roadmap deck currently promises that LLM tooling will lift output per engineer 20–40%. If that's true, and if it's broad-based, capital share should be visibly rising in the data within 2–3 years as firms convert software into output. So far, capital share is flat. Either the productivity gains aren't showing up in the national accounts yet (plausible — Solow's paradox redux), or they're being captured as markup rather than reinvested. Either way, anyone building internal tooling ROI cases should be honest that the macro evidence for the productivity dividend is, as of mid-2026, missing.
Regulatory risk on concentration is now a first-class infra concern. The FTC and DOJ have been quietly reorienting around markup and concentration metrics for two years. A Fed paper saying *markups drove the post-COVID labor share collapse* gives that effort top cover. Expect more aggressive merger review in cloud, ad-tech, and vertical SaaS — the categories where pricing power is most defensible — and price your acquisition assumptions accordingly. If you're at a sub-scale SaaS company assuming a strategic exit, the policy environment for the buyer just got worse.
The useful question for the next 18 months isn't whether the labor share recovers — slow-moving aggregates rarely snap back — but whether the markup wedge holds when the AI capex cycle either delivers measurable productivity or doesn't. If capital share starts climbing in the 2027 vintages of the same data series, the automation story belatedly arrives; if it doesn't and markups stay wide, this is just a concentration story wearing an AI costume. The Fed has now staked a position. Watch the next four quarters of nonfinancial corporate accounts for whether they're forced to revise it.
FTA's conclusion:"Is this decline a distinct change from the recent behavior of the labor share in the U.S.? Along the two key dimensions we investigate, our answer is no. <later> ... and they provide little evidence that it will evolve differently from past episodes."This concl
The submitted title is a bit sensationalist given the article’s conclusion:> Is this decline a distinct change from the recent behavior of the labor share in the U.S.? Along the two key dimensions we investigate, our answer is no. First, the labor share’s trajectory post-COVID broadly follows the
Raise the minimum wage to ~$25-30/hrBefore anyone says inflation, there will be more consumer spending and thus more cash flow in the middle and lower class
it feels like every share of income is at its lowest except for the ultra wealthy.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
People are misreading the conclusion - the Covid related drop is normal and matches previous episodes, but the massive overall drop since 2000 is not.The situation on the ground is unchanged - the amount of labor being generated per person has not really changed, but the overall pie has grown massiv