Sean's Manufacturing Index
August 2026 numbers · pulled 2026-09-27 · updated monthly · revised in public
If you read one thing on this page
Manufacturing wages are climbing faster than wages in the rest of the economy, and the jobs number is hiding the reason.
Factory pay is up 0.68 percentage points more than private-sector pay generally over the past year. That is higher than it has been in 88% of the last 20 years. And it is happening while the total number of manufacturing jobs sits basically still.
Here is why that combination matters to you.
When a plant is hiring because business is booming, it hires people who don't have jobs. Wages move a little. When a plant is hiring because it needs someone who already knows how to run the machine, it hires them away from another plant. And the only way to do that is money.
That second kind of hiring does not add a single job to the national count. One plant gains, another loses, the number doesn't move. But every one of those moves resets what an experienced hand is worth.
A flat jobs number with rising wages is not a boring economy. It is a bidding war that nobody is reporting.
More openings, same number of people moving. Employers want more hands than they are getting. That is the pressure, sitting there, not yet resolved.
What the jobs number actually counts
Once a month the government publishes how many manufacturing jobs the country has. It is one number. It gets quoted everywhere. It is also the least interesting thing in the report.
To build it, every employer in the country gets sorted into a bucket. The buckets come from a codebook called NAICS — the North American Industry Classification System. Say it “nakes.” Your plant has a number in it whether you have ever heard of it or not.
Here is the part that trips people up. The code follows your employer, not your work. An electrical contractor is an electrical contractor whether his crew is wiring a school or wiring a billion-dollar chip plant. He never becomes a manufacturer because of who hired him.
Which means a single data center gets split across three different industries:
- The crews building it are construction.
- The people running it once it opens are information, filed next to phone companies and publishers.
- Only the plants making the guts — the chips, the transformers, the switchgear, the steel — are manufacturing.
So when somebody tells you manufacturing is coming back because there are twenty thousand people building plants, check which bucket those twenty thousand are in. They are almost never in the one being quoted.
And most of them don't stay. Meta's big Louisiana campus peaks over 5,000 construction workers and settles at about 1,000 permanent jobs. That is not a scandal. That is construction — you finish the job and you go to the next one.
Flat is not quiet
Over the past year manufacturing added +23,000 jobs. Call it flat. That is the headline.
Underneath it:
About 5 jobs moved for every 1 job the headline showed. That is the whole idea behind this page. The total barely budged. The shape of the thing changed underneath it.
These are the five industries that feed the data-center and automation buildout. Every one of them grew faster than manufacturing as a whole.
| Industry | Jobs | Change |
|---|---|---|
| Electrical equipment, appliance & componentscode 335 | 443,400 | +3.48% |
| Primary metalscode 331 | 369,900 | +1.82% |
| Fabricated metal productscode 332 | 1,460,800 | +1.74% |
| Machinerycode 333 | 1,097,300 | +1.29% |
| Computer & electronic productscode 334 | 998,500 | +0.35% |
| All five together | 4,369,900 | +1.49% |
| Everything else in manufacturing | -0.49% |
Electrical equipment is the one to watch. Transformers and switchgear — the gear that moves power around. It grew +3.48% while the sector it belongs to grew +0.18%. Everybody talks about chips. The actual choke point is getting electricity from the wire to the machine.
The part that doesn't add up
One of those five is barely moving, and it is the one you would expect to be booming.
| Navigational, measuring & control instruments3345 | 419,200 | +2.02% |
| Semiconductors & other electronic components3344 | 370,400 | -1.38% |
| Computers, peripherals & communications equipmentrest of 334 | 208,900 | +0.19% |
Instruments and controls are growing. Chips and components are shrinking. During a chip boom.
The easy answer is that a plant under construction counts as construction, so the new fabs haven't hit the payroll yet. That is true. It is also only half an answer, because it explains why there is no gain. It does not explain the loss.
Look at the long run instead and it gets clearer:
The offshoring already happened. Half the jobs in this industry left between 2000 and 2010. Since then it has sat between roughly 365,000 and 395,000 for sixteen straight years — right through the CHIPS Act. The recent dip is a step down inside a flat range, not the tail end of the collapse.
What is actually in this bucket, in case you have a hunch about which part moved:
- 334413 — chips themselves, plus transistors, diodes, LEDs and solar cells
- 334412 — bare circuit boards, before anything is put on them
- 334418 — loaded circuit boards, the contract-assembly work
- 334416 — capacitors, resistors, coils, small transformers
- 334417 — connectors, plugs and sockets
- 334419 — everything else: switches, relays, antennas, sensors
Here is the test, and the answer surprised me. A jobs count can't tell you whether work went overseas or whether fewer people are doing it. But output can. If the work had left the country, American production would be falling too. So put the two side by side.
| August | Output made here | People | Output per person |
|---|---|---|---|
| 1990 | 0.3 | 572,600 | 0 |
| 2000 | 7.4 | 691,400 | 4 |
| 2010 | 46.8 | 372,200 | 46 |
| 2017 | 99.4 | 360,200 | 100 |
| 2022 | 132.8 | 394,400 | 122 |
| 2024 | 148.8 | 394,800 | 137 |
| 2026 | 190.1 | 370,400 | 186 |
Output and output-per-person are index numbers set to 100 in 2017, so you read them as multiples of that year, not as units of anything.
Output went up while the people went away. Every decade, without exception. Between 2000 and 2010 — the decade everyone points to as the offshoring — American production of this stuff rose more than sixfold while half the workforce left. This past year, output is up 12.4% and employment is down 1.4%.
That rules something out. Work that moves overseas takes its output with it. You cannot ship the jobs to another country and post record domestic production in the same year. Whatever is happening here, it is mostly not the plant closing and reopening somewhere cheaper. It is the same work being done by fewer hands.
One honest caveat, and it matters. The government measures chip output in a way that counts a faster chip as more output, not just more chips. So the size of these numbers is inflated — nobody is making six hundred times the physical parts they made in 1990. But the direction survives that objection completely. Domestic output rising and employment falling cannot both be true of an industry that packed up and left.
What the numbers do say is worth saying plainly: fifteen years of promises about bringing chip work home, and the industry that would show it has no more people in it than it did in 2010.
Why you see it before they do
If you work in a plant, you knew the labor market changed before any of this was published. Here is why the numbers are late.
A job sent overseas looks exactly like a job that disappeared. The count sees a separation. It cannot see whether the work stopped or just moved.
People who get cut leave the industry. They don't go to another plant. They go drive a truck or sell insurance, and manufacturing never counts them again.
Your plant's turnover can spike while the national number sits still. Average enough plants together and anything real gets averaged away.
The turn happens one company at a time. The day your employer stops cutting and starts hiring outside again — that is the turn. For you. It takes a year or more of that happening in enough places before it shows up in a chart.
So the chart is a rear-view mirror. Useful, honest, and behind. Which is a convenient thing for me to say, so I put an expiration date on it below.
What I think happens next
Two predictions. Both dated, both able to be wrong, both scored here in public.
The bleeding in the rest of manufacturing — food, paper, chemicals, plastics — slows down. Specifically, the yearly decline gets better than −0.5%. It is -1.02% today and it has been getting worse, not better.
The expiration date: if it is still worse than −1.0% on March 31, 2027, I was wrong. I don't get to say “the data is late” twice.
Once the losses stop covering for it, the wage pressure shows up in the open. Factory pay pulls ahead of everyone else's by more than 0.76 points — that is the line I'd call notable — and ideally more than 1.24, which would be a level this gap almost never reaches. It is 0.68 today.
How fast matters as much as whether. A quick jump means the pressure was already built up and the numbers were just slow — which is the whole argument above. A long slow grind means something milder is going on and I oversold it.
And I don't get to win on a recession. This gap hit its all-time high in April 2009, because when layoffs start they take the lowest-paid first and the average goes up. That is not a raise. So if the five industries above stop growing, this prediction fails even if the wage line cooperates.
Where this comes from
Every number here is public and free. It comes from the Bureau of Labor Statistics — the monthly jobs report and the survey of job openings and quits — pulled through FRED, the St. Louis Fed's data library. Last pulled 2026-09-27. You can check any of it yourself.
The five industries are codes 331, 332, 333, 334 and 335 — metals, fabricated metal, machinery, electronics, electrical equipment. I picked them before I looked at the results and I am not changing the list to make the chart look better. For the record: dropping electronics would make the gap look bigger. I'm leaving it in.
What I'm not sure about. This is one month's reading, so it tells you where things stand, not where they are heading — that takes a few more months. Metals and machinery serve the whole economy, not just data centers, so crediting their growth to the buildout is a judgment call. The government revises these numbers. And the argument about why the data runs late is an argument, not a measurement — it explains why I expect a turn, it is not evidence one has started.
The biggest hole. A jobs count can't tell the difference between work that went overseas and work that fewer people now do. Putting production numbers next to employment would separate them. That is the next thing I add.
This page updates every month and gets revised in public. If I got something wrong, I would rather fix it here than quietly.