US Census Annual Survey of Manufactures (NAICS 333120)

It is the federal government's own account of American construction-machinery factories, kept between the big five-yearly counts. The Annual Survey of Manufactures is the intercensal survey of US manufacturing establishments with one or more paid employees, and NAICS 333120 is its cell for construction machinery manufacturing - one of roughly 470 manufacturing industries the program tabulates. For that cell you get number of establishments, employment, production-worker hours and wages, annual payroll, cost of materials, value added, sales/value of shipments, inventories at three stages (finished goods, work in process, materials and supplies) and capital expenditures split between total and machinery-and-equipment specifically.

Two structural facts matter before anything else. First, the tables arrive in benchmark cycles - 2013-2016, then 2018-2021, with the 2022 benchmark release published April 2025 carrying the 2018-2021 statistics, plus earlier collections - so a long series is assembled by joining cycles rather than reading one endless ribbon. Second, the survey lives on inside the Annual Integrated Economic Survey, whose first main release (2023 reference year) landed on 2026-02-26; how much six-digit industry detail carries forward is exactly the thing to confirm when you scope a sample.

Why anyone selling into or studying heavy equipment cares: shipments size the domestic factory output, inventories tell you whether dealers are stocking or destocking, and capex tells you whether manufacturers themselves are betting on capacity. Get a sample of this dataset and the first rows arrive scoped to the industry cut you name.

Flat, annual, one row per industry-year - shaped to load straight into a warehouse table. Illustrative rows in the delivered column order, magnitudes following the publisher's own documented examples for the NAICS 333120 cell:

year: 2021   naics2017: 333120   indlevel: 6   indgroup: 333
estab: 900   emp: 52,000   payann: 3,800,000     unit: thousands of dollars
rcptot: 37,000,000   valadd: 15,000,000   cstmtot: 20,000,000

year: 2021   naics2017: 333120
empq1pw: 30,000   hours: 85,000   payanpw: 2,600,000

cextot: 1,200,000   cexmch: 900,000
invtotb: 5,500,000   invtote: 6,000,000   invfinb: 2,000,000

The block already computes into a picture. Value added of $15 billion against $37 billion shipped puts the industry's value-added share near 40% - the rest is bought-in materials, which the cost line confirms at $20 billion. Of $1.2 billion in capital expenditure, $0.9 billion went to machinery and equipment specifically: three-quarters of the sector's investment buys machines to make machines. Inventories closed half a billion higher than they opened, with $2 billion sitting in finished goods - the series that shows whether the dealer channel is filling or draining. Request a sample and these columns extend across the years and geographies you name.

Sixteen documented fields define the core row: calendar and classification keys, the headline money measures, production-worker detail, capital spending and inventory brackets. Definitions are verified against the publisher's own variable metadata rather than inferred from documentation prose.

Every estimate ships with quality companions too - relative standard error percentages and imputed-percent ranges - so noisy cells can be down-weighted instead of trusted blindly, and separate cubes carry product-class shipments and state-level cuts. Those ride along as additional fields on request: name the ones your model needs and they ship in the sample.

Geography - United States national totals by NAICS industry and product class, with US and state-level cuts in the area tables. State detail is where regional dealer networks, plant footprints and wage gradients become visible.

Temporal - benchmark cycles organize the archive: 2013-2016 and 2018-2021, with the 2022 benchmark release (published April 2025) carrying the 2018-2021 statistics, and earlier collections reaching further back; the successor survey's first main release covers the 2023 reference year. Treat the archive as dated blocks to be joined, not one continuous ribbon.

Granularity - annual establishment-level aggregates by NAICS industry from sector down to six-digit detail, by state, and by product class for shipments. One row per industry-year keeps joins trivial; the price is that within-year dynamics need faster indicators layered on top.

More on the industry this serves lives on our construction machinery heavy transportation equipment data hub.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the channel your team already works in: live queries against a specific industry-year, flat files sized for overnight warehouse loads, or a direct pipe into Snowflake, BigQuery or Redshift. Cadence is yours to set - and to change when your models change.

Every delivery ships with the full field dictionary, sample rows for validation, and a schema that holds steady between cycles.

• Market sizing with provenance - anchor US construction-equipment TAM decks in officially counted shipments and payroll rather than a vendor's reconstruction; see our market researchers use cases page.
• Equipment-cycle reads - pair value of shipments against inventory brackets to time the OEM cycle; the shipments-inventories-capex trio is the classic accelerator story for investors use cases.
• Capacity-bet detection - rising machinery-and-equipment capex flags manufacturers expanding capacity before it shows up in output.
• Dealer stocking signals - finished-goods inventories beside total inventories reveal whether the channel fills ahead of softening order books.
• Forecast feature panels - a schema-stable annual panel with verified definitions drops straight into a feature store; see our data scientists use cases.
• Citable official figures - journalists and academics cite federal numbers with named provenance instead of trade-association press releases.

Market researchers and consultants and data scientists rate this their top relevance: one group needs official denominators for market-size deliverables, the other a schema-stable annual panel with documented variables. Journalists, academics and students cite it because provenance survives scrutiny. Investors and quant researchers read shipments, inventories and capex as annual anchors for the heavy-equipment cycle. Developers building data products get flat industry-year rows shaped for warehouse loads, delivered daily, weekly, or hourly. Sales and competitive-intelligence teams use the payroll and shipment geography as the denominator layer under territory plans and rivals' growth claims.

Four things worth having upfront.

First, the archive arrives in benchmark blocks, not one continuous run - 2013-2016, then 2018-2021, then the successor survey's 2023 reference year. Joining cycles is routine, but definitions occasionally shift between them; the crosswalk ships with your sample.

Second, everything here is a surveyed estimate, not a census count, and each figure carries companions measuring exactly how surveyed: relative standard errors and imputed-percent ranges. Down-weight high-error cells rather than charting them bare.

Third, the dollar columns are denominated in thousands of dollars. It sounds trivial until someone charts payroll against revenue in mixed units and off by three orders of magnitude - convert once, centrally, or not at all.

Fourth, how much six-digit NAICS 333120 detail the successor survey carries forward is not stated on its program pages. If your model depends on that exact cut, say so in the sample request and we confirm the window before anything recurring starts.