Promised vs. Produced Tracker · methodology
The Promised vs. Produced Tracker (“Tracker”) measures how many announced American factories produce. This page explains how we build it.
With AI, anyone can build a tracker now. So the question is which ones to believe. Here is the standard we set. One factory site, not a company pledge. Every sector of manufacturing. Failures kept in the count. The promised date next to the real one. Every project sourced and free to reuse. Agents find the announcements. A person checks every project. 52 corrections logged in public. We check our coverage against the trackers that came before. And we publish what we miss. Coverage: about 80%. Weakest on clean fuels, Gulf Coast chemicals and 2025–26.
We count factories that meet these tests:
Located in a US state.
Covers a single physical facility, not a multi-site announcement. We track the promise made about that facility.
If a site is announced again with bigger numbers, it stays one project, dated and sized from the first announcement. Later figures go in the notes and are never added up. If the company later changes what the factory will make, it is still the same promise. The change goes in the notes.
Their announcement is published on or after January 1, 2017. The same rules apply across administrations.
Their announcement promises either at least $1 billion in capital OR 2,000 new jobs (directly in the factory).
Regional, supported, induced and construction jobs don’t count.
Its work is manufacturing as the federal government defines it: NAICS sectors 31–33.
Refineries, chemical plants and shipyards count. Mines, power plants, data centers and warehouses don’t.
Produces in one of the following industrial sectors:
All others are categorized as “Other.”
Six things for every project.
For each project, we record the date of announcement, the promised date of first output, the promised capital in USD and the promised jobs, where available.
When the project was first announced.
When the announcement said the factory would start producing.
The scale of the project, in USD.
The scale of the project, in jobs.
When the factory first makes goods for sale. Test runs don’t count.
The first time the factory makes goods related to what was promised, for commercial purposes.
One of six: as announced, under construction, paused, producing, closed or cancelled.
We also assign each project one of six current statuses:
An illustrative project, drawn the way the Tracker draws each row
We build the Tracker in three stages.* Agents run Source and Screen. A person runs Verify, and only rows that pass it reach the Tracker.
Source searches for new projects and collects links to each one’s announcement and to reports on its current status.
Passes on
Screen reads those links and fills in each field, noting which link every value came from.
Passes on
At Verify, a person checks every value against its link and corrects what is wrong. Only then does the project join the Tracker. Every check is public.
Passes on
The verified rows, as published.
Keeping the agents controlled
Tools for the verifiers
Which articles each field may come from
* Like the ELT medallion pipelines used in data engineering.
¹ Currently Anthropic Claude. The pipeline’s code does not depend on a particular foundation model.
The Tracker’s data is public on GitHub under CC BY 4.0, along with every correction and every check. Anyone can reuse it with credit.
| Promised | On the Tracker | Future work |
|---|---|---|
| Capital | $1B | $100M |
| Jobs | 2,000 | 200 |
Next, we plan to lower the floor from $1B in promised capital to $100M, and from 2,000 promised jobs to 200. At the current floor, the Tracker has 165 projects.
A person checked every figure, and every check is public: who checked it, when, which page they read and what it said. See the check log.
Found something wrong? Correct us on GitHub. No GitHub account? Email us. We check every correction against a public source before it changes the Tracker.
We recount the Tracker every quarter and publish each recount on the Tracker page.
How other trackers compare, as checked in September 2026. Hover over or tap a column name to see what it means.
Sixteen other trackers cover parts of the same ground.
Those that record what was promised.
Those that follow what happened after the promise.
Those that do some of both, but within a single sector.
Scroll the table sideways.
| Tracker | What it countsWhat one entry in the tracker is, and where. | SinceThe earliest year it covers. | BarThe smallest project it includes. | Follows to productionWhether it keeps checking each project after the announcement, and whether it records when the factory starts producing. | Cancelled projectsWhether cancelled projects stay in its counts. Dropping them makes the totals look better than they are. | Open dataWhether anyone can download and reuse the data, and under what license. | Our recallThe share of that tracker’s qualifying projects that we had in October 2026. See the coverage check below. |
|---|---|---|---|---|---|---|---|
| Promised vs. Produced Tracker | U.S. factory sites, all sectors | 2017 | $1B or 2,000 jobs | Yes, promised and actual dates | Kept | Yes, CC BY 4.0 | |
| “Promised” trackersThose that record what was promised. | |||||||
| Good Jobs First* | U.S. subsidy awards | 1976 | None | No | Not available | View only | 91% |
| fDi Markets (Financial Times)* | Foreign investment projects, worldwide | 2003 | None | Partly, opened or not | Dropped | Paid | Not checked |
| Reshoring Initiative* | U.S. reshoring jobs, totals only | 2010 | Not available | No | Not available | Paid | Not checked |
| Avison Young | U.S. manufacturing announcements | 2025 | $100M | Partly, to groundbreaking | Kept | No | 73% |
| IndustrialSage | U.S. manufacturing announcements | 2025 | $50M | No | Not available | Yes, CC BY 4.0 | 72% |
| “Produced” trackersThose that follow what happened after the promise. | |||||||
| Engineered-Vision | U.S. manufacturing megaprojects | Not available | $1B | Yes, year range | Kept | No | 98% |
| Dezernat Zukunft | German investment projects | Not available | Not available | Yes | Not available | Not available | Not checked |
| E2 (Environmental Entrepreneurs) | U.S. clean-energy announcements | 2022 | Large-scale | Cancellations only | Kept separately | No | Not checked |
| American Enterprise Institute (AEI) | Chinese investment deals abroad | 2005 | $95M | Troubled deals only | Kept separately | On request | Not checked |
| American Factory Tracker | U.S. manufacturing projects | 2018 | $100M or 100 jobs | Yes, expected date | Dropped from totals | No | Not checked |
| Sector-specific trackersThose that do some of both, but within a single sector. | |||||||
| Rhodium Group & MIT CEEPR | U.S. clean-tech facilities | 2018 | None | Yes, status only | Kept | Yes, CC BY 4.0 | 79% |
| SEMI* | Chip fabs, worldwide | Not available | Not available | Yes, expected date | Not available | Paid | Not checked |
| Semiconductor Industry Association (SIA) | U.S. chip sites | Not available | Not available | No | Not available | No | 94% |
| Global Energy Monitor | Steel plants, worldwide | 2017 | 0.5 Mt a year | Yes, start year | Kept | Yes | 86% |
| We Work For Health | U.S. pharma announcements | 2025 | Not available | No | Not available | No | 79% |
| Kiel Institute for the World Economy | European military orders | 2020 | Not available | Expected delivery only | Not available | Yes, non-commercial | Not checked |
* No update in the past 12 months, or no update date we could find.
Not checked: we didn’t measure our recall against it, because the tracker is paid, has no project-level data, isn’t about U.S. factories, mostly repeats a list we checked, or needs permission first.
Not available: the tracker doesn’t say, or we couldn’t see it.
We compared the Tracker with eight of the trackers above to see what we miss.
We have about 82% of the qualifying projects on those lists.
For each list, we counted the projects that meet our bar and how many of them the Tracker already had.
| List | Meet our bar | We had | Recall |
|---|---|---|---|
| Engineered-Vision | 51 | 50 | 98% |
| SIA | 18 | 17 | 94% |
| Good Jobs First | 87 | 79 | 91% |
| Global Energy Monitor, iron and steel | 7 | 6 | 86% |
| Clean Investment Monitor | 91 | 72 | 79% |
| We Work For Health | 14 | 11 | 79% |
| Avison Young, 2025–26 only | 11 | 8 | 73% |
| IndustrialSage, 2025–26 only | 39 | 28 | 72% |
After removing duplicates, the lists turned up 30 projects we should have had. We estimate we hold about 82% of the qualifying projects on these lists. Most of the misses are clean-fuel and chemical plants, and many were announced in 2025 or 2026. The next recount adds them.
Projects that no list has can’t show up here, so our real coverage is probably a bit lower.
These are the limits we know of.
We cover the U.S. only and use only public sources. The AI steps can’t be rerun to give exactly the same result.
The data is CC BY 4.0. Use it for anything, commercial work included, and name the Tracker when you do.
Short form, for slides, posts and articles
Source: IndustriousAF, Promised vs. Produced Tracker (as of Q3 2026)Full citation
IndustriousAF. (2026). Promised vs. Produced Tracker (Version 1.8) [Data set]. https://doi.org/10.5281/zenodo.23116675The data behind every figure here is on GitHub.
The next recount
We recount every promise each quarter. The next one is due at the end of Q4 2026.
Our privacy policy explains what we do with your email.