The AI Boom Arrived and Workers’ Share Hit a Record Low

Nonfarm output per hour rose 1.4 percent in the second quarter of 2026, the Bureau of Labor Statistics said on September 3. The labor share fell to 52.8 percent, the lowest reading since the series began in the first quarter of 1947.

Real hourly compensation dropped 3.3 percent in the quarter. Unit profits at nonfinancial corporations jumped at a 43.0 percent annual rate, the fastest since the second quarter of 2021. That split is the live test of an AI productivity boom, and it landed the same day Dean Baker, senior economist at the Center for Economic and Policy Research, argued that sharing the gains is not complicated.

The Labor Share Just Set a Record Low

BLS measures labor productivity as real output divided by hours. In the second quarter, output rose 1.7 percent at an annual rate and hours rose 0.3 percent, which is how the bureau got 1.4 percent. From a year earlier, output per hour was up 2.2 percent, a softer pace than the 3.0 percent annual gain in 2024.

Hourly compensation rose 2.6 percent in the quarter, so unit labor costs rose 1.2 percent. After consumer prices, that pay gain vanished. Real hourly compensation was down 0.1 percent over the last four quarters. The labor share is the slice of output that shows up as pay, and 52.8 percent is the floor of a series that starts in 1947.

THE Q2 2026 SPLIT

• Output per hour: Nonfarm labor productivity rose 1.4 percent at an annual rate, with output up 1.7 percent and hours up 0.3 percent.
• Workers’ slice: The labor share of output was 52.8 percent, the lowest level in a series that begins in 1947.
• Real pay: Real hourly compensation fell 3.3 percent in the quarter and 0.1 percent over four quarters.
• Profits: Unit profits at nonfinancial corporations rose 43.0 percent in the quarter and 17.8 percent over four quarters.

Manufacturing was the loud side of the report. Factory output rose 5.4 percent, the largest gain since the second quarter of 2021, and manufacturing productivity rose 2.4 percent. Durable-goods output jumped 8.9 percent. Those are production numbers, not a census of chatbots, and they still did not lift the labor share.

Through the current business cycle, from the fourth quarter of 2019 through the second quarter of 2026, nonfarm productivity has grown at a 2.1 percent annual rate. That matches the long-term rate since the first quarter of 1947 and beats the 1.5 percent rate of the 2007-2019 cycle. A faster cycle average has not restored the old split between pay and output.

2024 Already Delivered a 3 Percent Year

Baker’s essay treats 3.0 percent average annual productivity growth as the postwar bar. From 1947 to 1973, he writes, productivity grew at roughly that pace and wages and living standards rose for most people. He adds that it would be impressive if AI produced 3.0 percent growth over the next decade.

The bureau’s annual table already printed a 3.0 percent labor-productivity gain in 2024, with total factor productivity up 1.5 percent. In 2025, labor productivity slowed to 2.2 percent and total factor productivity slowed to 0.8 percent. The second-quarter 2026 print then slowed again, to 1.4 percent at an annual rate. The golden-age rate showed up for a year, then faded, which is a weak opening act for a decade-long AI miracle.

PRODUCTIVITY PRINTS AGAINST THE POSTWAR BAR

The Economic Policy Institute’s tracker, updated July 30, 2026, is the cleaner long view of who got the gains. From 1948 through 1979, productivity grew 2.5 percent a year and typical-worker pay grew 2.1 percent. From 1979 through 2026, productivity grew 1.4 percent a year and typical pay grew 0.6 percent. By the first quarter of 2026, the institute’s index (1948=100) stood at 418.8 for productivity and 254.9 for pay. In that accounting, pay has trailed productivity since 1979, including through the 2024 3.0 percent year.

Firm-level studies still find real time savings. A Quarterly Journal of Economics paper on 5,172 customer-support agents found a generative assistant raised issues resolved per hour by 15 percent on average, with the largest gains among less experienced workers. Nicholas Bloom, the Stanford economist, told a San Francisco Fed audience in March 2026 that Daron Acemoglu’s 2024 paper implied about 0.5 percent extra productivity over ten years, and that the Congressional Budget Office’s assumed AI add is about 0.1 percent. Those are small numbers next to a 3.0 percent year, and they are smaller still next to a 43.0 percent jump in unit profits.

Baker Says the Sharing Tools Are Sitting Unused

Baker’s September 3 essay is a history lesson aimed at people who talk about universal basic income as if the United States had never seen fast productivity growth. He is skeptical that AI will produce a huge uptick in output per hour, a skepticism he shares with the Congressional Budget Office’s baseline, and he says the sharing problem is old even if the models are new.

It’s entertaining to see the country’s great minds struggling with how to deal with the fallout from an AI productivity boom that may never occur, but the reality is that we have known for many decades how to ensure that the gains from productivity growth are widely shared.

Dean Baker, senior economist, Center for Economic and Policy Research

He names three tools that once pulled pay along with output: unions, a minimum wage that rose with productivity, and a shorter workweek. He also wants an end to what he calls upward redistribution through longer patents and copyrights, private-equity-friendly bankruptcy rules, and special protections for giant platforms.

THE THREE TOOLS BAKER WOULD REBUILD

• Unions: About a third of private-sector workers were in unions in the 1950s and 1960s, he writes; first-contract arbitration, as in Canada, would force a deal when firms stall.
• The wage floor: If the minimum wage had kept rising with productivity after 1968, he puts it close to $28 an hour, above the median wage in his telling.
• Hours: The 40-hour week dates to the Fair Labor Standards Act of 1938; he would cut it to 36 or 32 hours and mandate vacation, spreading the same work across more people.

The irony is not that those tools are mysterious. It is that 2024 already matched his 3.0 percent bar, 2025 still ran at 2.2 percent, and none of the three tools snapped back. Private-sector union membership stayed at 5.9 percent. The federal minimum wage stayed at $7.25. The standard week stayed at 40 hours. A boom year did not conjure the 1950s.

Why Private Union Density Is 5.9 Percent

The bureau’s union release for 2025, published February 18, 2026, is the current map of bargaining power. The overall membership rate was 10.0 percent, or 14.7 million workers, little changed from 2024. In 1983, the first year of comparable data, the rate was 20.1 percent and membership was 17.7 million. Public-sector membership was 32.9 percent, more than five times the private-sector rate of 5.9 percent.

Private-sector membership was 7.4 million; public-sector membership was 7.3 million. Workers represented by a union, including nonmembers covered by a contract, totaled 16.5 million, or 11.2 percent. Among full-time wage and salary workers, union members had median usual weekly earnings of $1,404, against $1,174 for nonunion workers. The bureau warns that those medians do not hold occupation, industry, or location constant.

The industries closest to AI software sit at the bottom of that table. Union membership was 0.8 percent in finance and 1.3 percent in professional and technical services. Utilities, a much older capital-intensive sector, were at 17.8 percent. If generative tools raise output in offices first, they are raising it in shops that almost never bargain.

Hayley Brown, writing for CEPR on January 26, 2026, noted that Canada’s union membership density in 2025 was nearly three times the U.S. rate and higher than Hawaii’s 24.7 percent and New York’s 21.3 percent, the two densest states. Baker’s point is that the U.S. drop is a legal design, not a cultural one. He flags the Taft-Hartley Act of 1947, which let states bar union-security clauses, and an National Labor Relations Board he says imposes trivial penalties when firms fire organizers.

HOW THE SHARING RULES WERE REWRITTEN

1. 1938: The Fair Labor Standards Act sets a federal minimum wage and the 40-hour week with overtime pay.
2. 1947: Taft-Hartley restricts union-security agreements and opens the door to state right-to-work laws.
3. 1947-1973: Nonfarm productivity grows at about 3.0 percent a year in Baker’s accounting, with unions covering roughly a third of private-sector workers.
4. 1968: The last period Baker cites in which the federal minimum wage moved in step with productivity; he says unemployment then was under 4.0 percent.
5. 2009: The federal minimum wage is set at $7.25 an hour and stays there.
6. 2024: BLS records a 3.0 percent annual labor-productivity gain.
7. 2025: Private-sector union membership holds at 5.9 percent.
8. September 3, 2026: BLS prints 1.4 percent quarterly productivity growth and a 52.8 percent labor share.

October 2025 union data were never collected because of the federal shutdown, so the 2025 rates are 11-month averages. CEPR says that truncation has historically moved the national rate by about 0.1 percentage point. It does not change the private-sector picture: 5.9 percent is not a third of the workforce.

A $7.25 Floor Against a $28 Productivity Path

The federal minimum wage was $7.25 an hour in August 2026, unchanged since 2009. EPI, in a May 21, 2026 paper, called that a 17-year freeze and said the floor had lost about 30 percent of its purchasing power. The same paper proposed tying the federal wage to two-thirds of the national median, about $17.70 at the time, and projected about $20 in 2030.

Baker’s figure is higher because he indexes to productivity, not to prices or to the median. If the minimum wage had kept rising with productivity after the late 1960s, he puts it close to $28 an hour. In his account, more than half the workforce would get a direct raise and another 20 to 30 percent would see a spillover. That is a different instrument from a $15 state floor. It is also an instrument Congress has not touched for 17 years, including through the 3.0 percent productivity year in 2024.

Nineteen states raised their minimum wages on January 1, 2026, and EPI estimated that 8.3 million workers would benefit. For the first time, more workers lived in states with a $15 or higher floor (66.4 million) than in states stuck at $7.25 (60.2 million). The federal number still sets the floor in the latter group. A national productivity link was not on that January list.

The 40-Hour Week Has Not Moved Since 1938

Baker’s simplest arithmetic is the one AI panels keep stepping around. He writes that 150 million jobs at 40 hours a week, 50 weeks a year, equal 300 million jobs at 20 hours. The identity is just hours: 6,000 million either way. If output per hour really soared, the same output could be produced with fewer hours per person, or with more people each working less.

The Fair Labor Standards Act of 1938, not 1937 as Baker’s essay has it, made 40 hours the overtime threshold. Almost 90 years later, the threshold is still 40. In a May 6, 2026 Beat the Press post, he suggested dropping the overtime trigger to 34 or 36 hours and raising the overtime premium from 50 percent to 75 percent, along with mandated vacation of the sort every other rich country already requires. That would spread work. It would also require Congress to reopen a statute that has not been used that way since the New Deal.

Shorter hours only share a boom that exists. The second-quarter print, 1.4 percent, is not a boom. The 2024 print, 3.0 percent, was closer, and hours policy did not move then either. The unused hours lever is the same story as the unused union lever: the legal settings stayed where they were when output per hour ticked up.

Wall Street Needs 4 to 5 Percent Growth

Baker’s AI Bubble Monitor, posted August 31, 2026, is the valuation half of the same argument. He writes that current tech prices only make sense if productivity growth jumps from the 1.5 to 2.0 percent range of recent years to 4.0 to 5.0 percent a year, unless profits take an even larger bite out of wages. On September 3 he posted that the new BLS figures were “going in the WRONG direction for your AI taking all the jobs story,” and that fans should be seeing 3 to 4 percent or higher.

The bureau’s own cycle average, 2.1 percent since late 2019, is the long-run mean, not a 4.0 percent breakout. A 52.8 percent labor share is the opposite of wages absorbing a miracle. If anything, the second quarter looks like a profits boom riding a modest productivity gain, which is the pattern EPI has been charting since 1979.

On September 6, Baker applied the same arithmetic to retirement. If AI is about to wipe out jobs, pushing people to work later is a strange response. He put it without hedging.

I know arithmetic and logic have no place in Washington policy debates, but if we are concerned that AI is going to kill all the jobs, it is batshit crazy to be pushing people to delay retirement. https://t.co/b3UqfgSnVp

— Dean Baker (@DeanBaker13) September 6, 2026

“I know arithmetic and logic have no place in Washington policy debates, but if we are concerned that AI is going to kill all the jobs, it is batshit crazy to be pushing people to delay retirement,” he wrote. The next Productivity and Costs report is scheduled for November 5, 2026. The labor share has already printed its answer for this quarter.