On August 3, the Institute for Supply Management reported that its Manufacturing PMI hit 55.6% in July 2026 — the highest reading since May 2022, a seventh consecutive month of expansion, and a 1.6-point beat against a 54.0 consensus. Four days later, the Bureau of Labor Statistics reported that the U.S. economy lost 23,000 jobs in July against an expected gain of 80,000, and revised May and June down by a combined 103,000. Manufacturing's contribution to that wreckage was a gain — of 5,000 jobs.
The instinct is to pick a winner. Don't. Both numbers are measuring what they claim to measure. They diverge because one counts firms and the other counts people, and because the thing American factories bought in July was not workers. It was hours.
The scoreboard
Set the two releases side by side, and the shape of the problem is immediate.
ISM Manufacturing PMI — July 2026 (rel. Aug 3)BLS Employment Situation — July 2026 (rel. Aug 7)Headline PMI 55.6, up 2.3 pts from June's 53.3Total nonfarm payrolls -23,000 (consensus +80,000)Production 58.5 vs 52.2 (+6.3, highest since Nov 2021)Manufacturing payrolls +5,000, to 12,611,000New Orders 56.7 vs 56.0; Backlog 55.0 vs 50.5Private +30,000; government -53,000Employment Index 52.8 vs 49.7 — first expansion in 33 monthsMay revised -66,000; June revised -37,000 (103,000 combined)New Export Orders 53.0 vs 48.5; Prices 71.1 vs 73.0Unemployment rate 4.1%15 of 16 industries expanding; Chemical Products the lone declinerManufacturing still down 14,000 year over yearThe ISM release is available in full here; the BLS summary is here. One note on vintage before going further: with 103,000 jobs erased from two prior months in a single release, any payroll figure cited in the next several weeks should carry its data vintage. June manufacturing, for what it's worth, was revised up — from +3,000 to +11,000.
Why a diffusion index cannot tell you about headcount
The ISM Employment Index turning positive after 33 months of contraction is a genuinely notable event, and it was widely reported as though 52.8 meant factories were adding workers. It does not mean that, and it cannot.
ISM asks purchasing and supply executives a directional question: is your company's employment higher, lower, or the same than the prior month? The index is the share reporting higher plus half the share reporting the same. A 40-person job shop that adds two people registers identically to a 4,000-person plant that adds two people. Size is invisible. The index measures breadth of change across firms, not magnitude of change in bodies.
July supplied a clean proof case. ISM listed Food, Beverage & Tobacco Products among its 15 expanding industries. In the same month, BLS food manufacturing shed 6,200 jobs — the largest single-subsector loss in the report. Both are true: more food-manufacturing firms reported activity higher than lower, while the firms doing the cutting were larger, or cut deeper, than the ones adding. Conversely, ISM's single contracting industry, Chemical Products, lined up neatly with BLS chemicals at -2,000. Sometimes breadth and magnitude agree. In July, mostly they did not.
ISM itself put a caveat in the release that got less attention than the 33-month headline: 60% of panelists reported hiring, while 40% said managing headcount remains the norm. That is a survey describing a sector split between expansion and containment — which is exactly what a +5,000 payroll print looks like from the inside.
The actual answer: factories bought hours, not heads
Here is the number that reconciles the week, and it comes from the same BLS release everyone read as bad news.
The manufacturing all-employee average workweek was 41.7 hours in July 2026, against 41.1 hours in July 2025. Overtime rose to 4.1 hours from 3.7. Over the same twelve months, manufacturing headcount fell 14,000, or 0.11%, from 12,625,000 to 12,611,000.
Multiply out. Aggregate weekly hours worked in manufacturing rose roughly 1.35% year over year on a payroll base that shrank. That is the production surge ISM's panelists are describing — it is real, it is measurable in hard data, and it required no net hiring at all.
Now run it the other way, which is the version that should matter to anyone budgeting for 2027 labor. Hold the workweek at last year's 41.1 hours and ask how many workers it would take to deliver July 2026's labor input. The answer is about 12,795,000 — roughly 184,000 more workers than are actually on the payroll.
That is the boom. It was absorbed by the people already in the building, working an extra half-hour a week each, with overtime up 24 minutes. A 55.6 PMI and a +5,000 payroll number are not contradictory readings of the same economy; they are the numerator and denominator of the same operating decision.
Where the 5,000 actually came from
The headline gain is a durable-goods story with a nondurable-goods drag underneath it, and the two nearly cancel.
- Durable goods: +18,000 (7,869,000 vs 7,851,000)
- Nondurable goods: -13,000 (4,742,000 vs 4,755,000)
Within durables, the subsector detail maps almost perfectly onto what ISM panelists said they were building: transportation equipment +11,900, computer & electronic products +2,900, machinery +2,600, fabricated metal products +2,500, nonmetallic mineral products +800, primary metals +600. On the other side: food manufacturing -6,200, furniture and related products -2,300, chemicals -2,000.
ISM's panelist comments name the demand source directly. A Computer & Electronic Products respondent cited "growth in the semiconductor, AI, advanced packaging markets." A Machinery respondent said "products going into data centers are at full procurement and manufacturing ramp-up." This is a capital-goods cycle concentrated in the most capital-intensive, least labor-intensive corners of the sector. Advanced packaging lines and data-center equipment consume enormous capex per unit of output and comparatively few production workers. The demand is unambiguously there. The employment multiplier attached to it is small.
Real hiring, or slower bleeding?
One plausible reading of the Employment Index turn is that nothing is being added — attrition simply slowed, so headcount stopped falling. JOLTS lets us test that, and it doesn't hold.
June 2026 manufacturing data (the latest available) shows:
- Job openings 481,000, vs 391,000 a year earlier — +23%
- Hires 329,000, vs 288,000 — +14%
- Quits 192,000, vs 174,000 — +10%
- Layoffs and discharges 94,000, vs 104,000 — -10%
Gross hiring is genuinely accelerating. But quits are accelerating alongside it, and the layoff decline is modest. What that combination produces is elevated replacement churn, not a step-change in net labor demand: 329,000 people hired to backfill 192,000 who left plus 94,000 who were let go, with a few thousand left over. A 52.8 diffusion reading and a +5,000 payroll print are what that looks like from two different instruments simultaneously.
Practically, that means the sector's recruiting cost per net add is brutal right now. Firms are running full req pipelines to stand still.
The capex question the boom framing gets wrong
The standard forecast move after a four-year-high PMI is to project a capex wave two to three quarters out, usually branded as reshoring. Check the base it would be building from.
Total construction spending on manufacturing structures was $172.674 billion SAAR in June 2026 — down 31.0% from the September 2024 peak of $250.233 billion, and down 21.4% year over year from $219.564 billion. The factory-building wave peaked two years before this PMI print and is still receding.
That does not make the reshoring thesis wrong, but it does relocate it. The structures cycle — the CHIPS- and IRA-era shell construction — is behind us. What remains is equipment spending: the tools, lines, and packaging capacity going into buildings that already exist. Any forecast citing a capex acceleration off this PMI should be forced to say which measure it means. Structures are contracting. Equipment, per the ISM order and backlog data, is not.
Fact-checking the same-week framing
The White House released a statement on August 3, hours after the ISM print, titled "Under President Trump, U.S. Factories Expand at Fastest Clip in More Than Four Years." Some of it checks out against the hard data. Some does not.
Holds up. The 4.2% wage claim is accurate on the production and nonsupervisory series: $30.35 per hour in July 2026 vs $29.14 a year earlier. (The all-employee series is softer at $36.87 vs $35.51, or +3.8% — worth noting, but the cited figure is a real series, correctly stated.) "Fastest clip in more than four years" is a fair characterization of 55.6 against May 2022's 55.9.
Does not hold up. "Record production" is true of ISM's diffusion reading and false of measured output. The Federal Reserve's industrial production index for manufacturing stood at 98.70 in June 2026 (2017=100) — up about 1.8% since December 2025, and still 7.4% below its December 2007 all-time high of 106.61. American factories are producing meaningfully less than they did nineteen years ago.
Needs a category correction. The 83,000 "factory construction jobs added since inauguration" are construction-sector employment — people building plants, not people working in them. That is a legitimate number in its own right, but it is not manufacturing payrolls, and its spending base is the one down 31% from peak.
Needs a vintage stamp. The "more than 18,000 manufacturing jobs created in 2026" figure predates the July revisions. On the current data vintage, manufacturing is up 31,000 since December 2025 (12,580,000 to 12,611,000) — a better number, and still down 14,000 against July 2025.
Does not reconcile internally. ISM publishes its own translation: a 55.6 PMI "corresponds to a 2.8-percent increase in real GDP on an annualized basis." The same release cites a Q3 GDP tracking estimate of 6.2% annualized. If the ISM print is the evidence, the ISM model is the mapping — and it delivers less than half that figure.
TD Economics, reading the same release on August 4, split the difference credibly: manufacturing is "no longer merely stabilizing, but gaining momentum," while "hiring plans remain sensitive to demand visibility and cost pressures." With ISM Prices still at 71.1 on steel, aluminum, tariffs, and petroleum inputs, that cost sensitivity is not hypothetical.
What this means for workforce planning
Strip the politics out and the July data describes a specific, actionable condition: a sector running 41.7-hour weeks with 481,000 unfilled openings and net-flat headcount. That is not a demand problem. It is a matching problem, and it is being financed with overtime.
Overtime is the cheapest capacity in the short run and the most expensive in the medium run. At 4.1 hours of weekly overtime and climbing, the sector is carrying premium-rate labor, accumulating fatigue-related quality and safety exposure, and — per the JOLTS quits data — feeding its own attrition. The 10% year-over-year rise in quits and the 41.7-hour workweek are not unrelated facts.
The openings are concentrated where the hires are: durables, and specifically the semiconductor, advanced-packaging, and data-center equipment complex that added 20,000 jobs across transportation equipment, computers/electronics, machinery, and fabricated metal in a single month. Those are the roles employers describe as specialized and technical — which is to say, the ones where a 90-day req cycle is optimistic and where poaching from a competitor is the default fill strategy. Meanwhile nondurables, food manufacturing above all, is shedding the kind of headcount that does not transfer laterally into a packaging fab.
Three things to watch
- The August ISM Employment Index. Does 52.8 hold, or was the 33-month turn a one-month artifact? One print is not a trend, and this index has faked out before.
- The September JOLTS quits rate for manufacturing. If quits keep rising alongside hires, the churn diagnosis is confirmed and net adds stay near zero regardless of how strong the order book looks.
- The workweek. This is the leading indicator that matters. Firms extend hours before they hire and cut hours before they fire. If the 41.7-hour week keeps climbing without headcount following, the demand is real and the hiring constraint is structural. If the workweek rolls over while payrolls stay flat, the July PMI was the peak.
Until one of those breaks, the honest summary of American manufacturing in mid-2026 is this: order books are the best they've been in four years, output is still 7.4% below where it was in 2007, the buildings boom ended two years ago, and the people making the difference are the ones already on the floor, working longer.
Related reading
-
[The Navy Just Printed 1,000+ Parts Underway: Inside RIMPAC 26's USS Essex, the Pentagon's Largest Distributed-Manufacturing Live Fire](/article/rimpac-2026-uss-essex-distributed-manufacturing-3d-printing)
-
[A Korean-Owned Yard Wants to Go From 2 Ships a Year to 20 — and It Just Won Another $1.5B in Federal Ship Orders](/article/hanwha-philly-shipyard-throughput-bet-15b-nsmv-marad)
Sources
- Manufacturing PMI® at 55.6%; July 2026 ISM® Manufacturing PMI® Report — Institute for Supply Management, August 3, 2026
- Employment Situation Summary, July 2026 — U.S. Bureau of Labor Statistics, August 7, 2026
- BLS CES series CES3000000001 — All employees, manufacturing (with companion series for durables, nondurables, weekly hours, overtime, earnings, and subsector detail)
- BLS JOLTS — Manufacturing job openings, hires, quits, and layoffs
- FRED — Total Construction Spending: Manufacturing (TLMFGCONS), U.S. Census Bureau
- FRED — Industrial Production: Manufacturing (IPMAN), Federal Reserve Board
- Monthly Construction Spending, June 2026 — U.S. Census Bureau
- Manufacturing jobs continue upwards, adding 5,000 jobs in July — Manufacturing Dive
- July 2026 jobs report: US economy unexpectedly shed jobs — Fox Business
- Under President Trump, U.S. Factories Expand at Fastest Clip in More Than Four Years — The White House, August 3, 2026
- U.S. ISM Manufacturing Index (July 2026) — TD Economics, August 4, 2026