Census Monthly Wholesale Trade & Manufacturing API (MWTS/M3) Data

It is the federal government's answer to two questions every semiconductor planner asks - how much moved through distribution this month, and how much did domestic factories ship - answered in one queryable time-series service.

The Monthly Wholesale Trade Survey (MWTS) covers merchant wholesalers except manufacturers' sales branches and offices: monthly sales, end-of-month inventories and the derived inventories-to-sales ratio, by kind of business across NAICS sector 42. The electronics-relevant lines are already broken out: professional and commercial equipment (4234), whose June 2026 preliminary print sits at $64,158 million; computer and computer peripheral equipment and software (42343) at $16,234 million; and household appliances and electrical and electronic goods (4236) at $14,211 million, inside a durables total of $397,510 million against $794,101 million for all merchant wholesalers.

Manufacturers' Shipments, Inventories, and Orders (M3) covers the factory side: value of shipments, new orders, unfilled orders and total inventories split into materials, work in process and finished goods, tabulated across dozens of industry categories with Computer and Electronic Products (NAICS 334) as the electronics aggregate. One caveat matters before any model touches it: the standalone semiconductor series was published only in historical shipments files through 2010, because new orders and unfilled orders estimates exclude semiconductor data - so current-month electronics reads ride the 334 aggregate, not a semiconductor-only line. Get a sample of this dataset cut to the categories and months you name.

Every observation is one cell: a survey program, an industry category, an item type, a month, an adjustment flag and a value. The block below uses the source's own published anchors:

# MWTS cell - one category x measure x month x adjustment flag
program_code     : MWTS
category_code    : 42          total merchant wholesalers, except MSBOs
data_type_code   : SM          sales of merchant wholesalers
time_slot_name   : Jun 2024    (time_slot_id 2024-06)
seasonally_adj   : yes
cell_value       : 794101      dollars in millions

# the electronics-relevant wholesale lines sit one code deeper
category_code    : 4234        professional and commercial equipment -> cell_value 64158
category_code    : 42343       computers and peripherals and software -> cell_value 16234
category_code    : 4236        household appliances and electrical/electronic goods -> cell_value 14211

# M3 cell - manufacturing side of the same channel
category_code    : 34S         Computer and Electronic Products (NAICS 334)
data_type_code   : VS          value of shipments
seasonally_adj   : yes         cell_value 34696 (June 2026)
seasonally_adj   : no          cell_value 19172 (January 1992, same measure)

Read the first and last cells together and the panel earns its keep. Total wholesaler sales ran $142,980 million in January 1992 against $794,101 million in June 2026 - up more than fivefold with the schema unchanged, which is why a backtest written against the 1990s cells runs unchanged against last month's. The seasonally_adj flag is the guardrail: every measure publishes twice, adjusted and not, and mixing the flags inside one trend line manufactures turning points that never happened. Dollars arrive in millions, so unit economics need a price index supplied on your side.

Eleven fields carry every observation, in four bands.

An identification band: program_code names the survey behind the cell, category_code carries the NAICS-based industry list entry - 42 through 4249 on the wholesale side, 334's groups on the manufacturing side - and geo_level_code pins the single national geography.

A measure band: data_type_code selects the item type, sales on the wholesale side, shipments and orders and inventories on the factory side, with ratios and percent-change variants alongside; cell_value holds the estimate itself.

A calendar band: time_slot_id, time_slot_date and time_slot_name fix the reference month three ways, and the time predicate accepts single periods or from/to ranges.

A quality band: seasonally_adj separates the adjusted twin from the raw one, and error_data marks sampling-error records so they never masquerade as estimates.

Three chips summarize the footprint:

• Geography: United States national totals only - a single geography level by design. Sub-national wholesale or manufacturing reads need a companion dataset, not a parameter.
• Time frame: monthly from January 1992 to the latest reference month on both surveys, roughly four hundred observations per series, with each month's first estimate flagged preliminary and superseded in later vintages. Annual revisions reach back years, so treat history as versioned rather than frozen.
• Granularity: one observation per industry category x item type x month x adjustment flag. Wholesale detail descends from the NAICS 42 aggregate through subsector totals (423 durable, 424 nondurable) to selected five-digit lines such as 42343; M3 publishes at group level with 334 as the electronics aggregate.

Within Datadory's catalog of 1,744 datasets averaging a 7.81 quality score, this record holds 7/10, carried by verified field definitions, a thirty-four-year monthly panel and the cleanest join key in the slice - a NAICS code that connects wholesale cells to every other industry-coded dataset we pool. The best semiconductor materials and equipment datasets ranking places it among federal statistical feeds.

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

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

Pick the channel your pipeline already speaks: query-shaped lookups for live dashboards keyed on category-measure-month, flat files sized for overnight warehouse loads, or a direct landing in Snowflake, BigQuery or Redshift. Cadence changes are a settings conversation, not a re-integration project. > > Every delivery ships with the field dictionary above unchanged and validation cells drawn from real months, with adjusted and unadjusted twins kept as separate columns - so the seasonal-flag check that should precede every trend chart is a select statement, not an audit.

Ranked by how directly one monthly cell settles their day job:

1. Data scientists & ML engineers. A fixed-schema monthly panel reaching back to January 1992 spans multiple inventory cycles - training features without a parsing project or schema drift.
2. Investors & quant researchers. The inventories-to-sales ratio is the classic restocking gauge: when ratios rise while sales flatten, distribution is stocking ahead of demand, and that turn prints here before it reaches anyone's guidance call.
3. Market researchers & consultants. Category-level dollar values put figures under American electronics distribution sizing, citable with federal methodology behind them.
4. Competitive intelligence teams. A distributor up 4% in a market up 9% is losing share - category baselines make that visible instead of anecdotal.
5. Policy analysts & journalists. Chip incentives, tariffs and reshoring arguments land on numbers with a named statistical program behind them.

For contrast inside the same slice: the Census M3 Manufacturers' Shipments, Inventories & Orders page covers the factory side alone, and the Monthly Wholesale Trade Survey sales and inventories page covers the wholesale side alone - this dataset joins both through one service.

Data scientists and ML engineers get a thirty-year panel with stable codes, ready for cycle features without a labeling pass. Investors and quants get a restocking signal that leads distributor prints, revision flags intact for honest backtests. Market researchers and consultants get channel sizing with federal provenance behind every figure. Competitive intelligence teams get the baseline their internal numbers should be judged against. Across all of them the constant is the join key: one NAICS category code connecting these cells to every other industry-coded dataset in the pool. Persona-by-persona detail lives on the data science, investor and competitive intel industry pages.

Inside this six-dataset slice, each source answers a different question and this one owns the channel. The SIA Semiconductor Market Data & Factbook owns global billings - monthly WSTS shipments across 200+ categories back to 1976 plus forecasts. The OEC Integrated Circuits Trade Profile owns geography, with bilateral HS 8542 flows across roughly 240 economies. UCI SECOM owns the fab floor at 1,567 wafer tests x 591 sensor channels. The Wikipedia Market Tables own the competitive lineage back to 1975. This feed owns what happens between factory and end customer inside the United States - the one leg none of the others carries, and the only one that answers where inventory actually sits.

Four things, stated plainly.

First, the geography is one row deep: national totals only. If the question needs states, metros or ports, this dataset answers a different question than the one you are asking.

Second, values are dollars in millions, not units. Pairing against physical volumes requires a price deflator on your side; nothing here counts chips or boxes.

Third, history is versioned. Each month's first estimate carries a preliminary flag and gets superseded in later vintages, and periodic comprehensive revisions restate long runs - keep the vintage you trained on or re-pull deliberately.

Fourth, the semiconductor-specific manufacturing series ends where the aggregate begins: current-month electronics reads ride the Computer and Electronic Products aggregate because new orders and unfilled orders exclude semiconductor detail. Name the categories, measures and months you care about when requesting the sample and it arrives shaped to that scope, folded codes confirmed against live cells.