Buyers name these record families on their own pages, and the first limit on each is ownership.
AI data licensing means a company grants an AI developer rights to use its data for training.
Programs are listed alphabetically, never by fee, and every buyer figure is a dated claim, checked 2026-11-09 unless marked, with the Mercor listings re-read on 2026-11-17. Sources are buyers' own pages, eCFR text and the Federal Register. Run records are one sector of AI training data licensing.
What a Fab, Device Maker or Materials Lab Holds That AI Buyers Ask For
Chip and materials companies hold six record families that AI buyers name, and the one asked for most precisely is the run or build record that ties a recipe to its measured result, failed builds included.
Mercor's semiconductor listing, read 2026-11-17, asks for records the company owns, with recipe or process settings linked to measured performance, 1,000+ runs, builds, wafers or experiments, and data that has not been published. The table compares the six families, as of 2026-11-17.
| Record family | Typical contents | Who names it (stated by) | First limit | Source |
|---|---|---|---|---|
| Run and build records | recipe or process settings linked to measured performance; every build kept, failed ones included, with dates; the same recipe built again on different days | Mercor Semiconductor & Materials listing | ownership; unpublished only | Mercor listings feed, semiconductor |
| Process, yield and equipment data | yield, metrology and equipment logs; SPC and yield systems, MES, OSIsoft PI | Mercor | customer lots; classification under the EAR | Mercor listings feed, semiconductor |
| Lab notebooks | lab notebooks with notes on why builds failed; electronic lab notebooks, LIMS | Mercor (semiconductor and R&D listings) | employee and contractor terms | Mercor listings feed, semiconductor and company-rd-research-records |
| Chip and device designs with test results | design and test files where available; chip or device designs with their verification and test results; Cadence, Synopsys | Mercor | licensed third-party IP; customer designs | Mercor listings feed, semiconductor |
| Measurements under several conditions | measurements under more than one condition; instrument outputs, sensor or time-series data and simulation outputs linked to sample IDs, units and calibration | Mercor (semiconductor and R&D listings) | classification under the EAR | Mercor listings feed |
| Research episodes and decisions | hypotheses, versioned protocols with equipment settings, failed and null experiments, stop/go decisions | Mercor R&D records listing | published or co-owned work | Mercor listings feed, company-rd-research-records |
Mercor's R&D listing, read 2026-11-17, adds equipment settings and instrument outputs linked to sample IDs. Equipment time series are a data type of their own, covered under process histories as sensor, IoT and telemetry data.
Buyer Statements That Name Semiconductor and Materials Records
Four buyer statements on record reach semiconductor or materials records, and Mercor runs the one listing written for fabs, foundries, device makers and materials labs. The table compares them alphabetically, as of 2026-11-17, with each value the stating party's own claim, checked 2026-11-09 unless marked.
| Program or terms | What it names for chip and materials companies | Threshold stated | Range stated (claim) | Source |
|---|---|---|---|---|
| Handshake AI Data Partnerships | "Hardware & engineering" and "Manufacturing" among 11 industries | no minimum (Data Partnerships page) vs 20+ full-time employees and 3+ years (license and referral pages) | program-wide, not sector-specific; not used here | joinhandshake.com/ai/data-partnerships |
| Mercor R&D records listing (title "Chem / Mat Science R&D records: hypotheses, experiments, results and failed work") | dated research records; priority domains include materials and batteries, industrial engineering and process R&D | not stated | USD 10,000 to 300,000 | Mercor listings feed, company-rd-research-records |
| Mercor Semiconductor & Materials Company listing | owned run or build records with settings linked to measured performance; 1,000+ runs, builds, wafers or experiments; unpublished | 10 to 2,000 employees; anywhere | USD 100,000 to 750,000; "Depends on run count and how well runs link to results" (re-read 2026-11-17) | Mercor listings feed, semiconductor |
| micro1 Data License Master Terms | s.5.7 excludes export-controlled information unless the Order Form allows it | not a sector rule | none in the terms; License Price per Order Form | micro1.ai/legal/data-license-master-terms |
Mercor's semiconductor listing names fabs and foundries, device makers (solar, photonics, quantum, power electronics), fabless chip designers, process and test equipment makers, and materials and battery R&D labs as its industries, and it states that labs and startups that have wound down qualify (read 2026-11-17).
Each range is a claim, not a paid amount, and ranges are never averaged, ranked or converted to a price per run. The deal tracker lists no semiconductor or materials transaction as of 2026-11-17. The terms of every company data partnership program sit on the program page.
Customer, Foundry and Licensed-IP Records a Chip or Materials Company Cannot License
Mercor lists data owned by a customer or foundry partner and licensed third-party IP as not a fit, and its scoping question asks whether the company owns the data and whether any of it has been published (read 2026-11-17). Four situations need a contract check before any offer:
- Wafers run for a fabless customer: layouts and customer-specific recipes belong to the customer relationship, and who owns the run data under a foundry agreement is a question for the contract; this page states no general rule.
- A fabless designer's records from a foundry: process design kit material and foundry data sit under the foundry's own terms, and this page states no general rule on them.
- Licensed third-party IP inside designs: Mercor excludes it, so each design is read for embedded blocks before it is counted.
- Co-development or government-funded projects: the project contract is read first, and information is controlled unclassified information only where it is created or held for an agency (32 CFR 2002.4(h)).
Secrecy duties travel with these records. Under 18 U.S.C. 1839(5), misappropriation includes disclosure by a person who acquired a secret under a duty to maintain secrecy, and Directive (EU) 2016/943 Art. 4(4) makes acquisition, use or disclosure unlawful where the person knew or ought to have known the secret came from someone using or disclosing it unlawfully.
The ownership check comes first, and customer and foundry contracts are the first layer of a rights audit.
Recipe and Yield Know-How: What a License Does to Its Secrecy
A company's own recipes and yield history are often its trade secrets, and a trade secret exists under 18 U.S.C. 1839(3) only while the owner takes reasonable measures to keep it secret and the information has independent economic value from not being generally known.
License terms decide what happens next. micro1's master terms (last updated 2026-11-12, re-read 2026-11-17) state a worldwide, nonexclusive, irrevocable, perpetual license, sublicensable through multiple tiers (s.1.2), and no duty to delete or retrain Licensee Materials (s.4.4).
Labelbox Prism states that "Models already trained on the data cannot be unlearned" and that its license is exclusive by default (checked 2026-11-09). The secrecy question is settled in the terms before any data moves.
The conditions in four jurisdictions and the effect of a license on trade-secret status are set out in the comparison of trade secrets and confidential information in AI data licensing.
Where Run Records Sit Under the EAR: CCL Categories, ECCNs, EAR99 and Published Data
Under the Export Administration Regulations, process and design records are "technology", and the Commerce Control List sorts technology into group E of ten categories, among them category 1 (Materials, Chemicals, "Microorganisms," and Toxins), category 2 (Materials Processing) and category 3 (Electronics), the categories whose names match chip and materials work. This follows 15 CFR 738.2(a) and (b), read on eCFR 2026-11-17.
Each entry carries an Export Control Classification Number whose first digit is the category and whose following letter is the group, as in 3A001 (15 CFR 738.2(d)(1)).
Items subject to the EAR that are not on the list are designated EAR99 (15 CFR 734.3(c)). No ECCN for any type of run record is stated on this page, because classification is the company's own task, made file by file.
Published data sits outside the EAR. Under 15 CFR 734.7(a), unclassified technology is published when it has been made available to the public without restrictions on further dissemination, including (a)(5) the submission of a computer-readable dataset with the intention that it be made public if accepted for publication, and 734.3(b)(3) also excludes fundamental research and published patents.
Mercor's requirement of "Data that hasn't been published" asks for exactly the records that exclusion does not cover.
A deemed export is the release of technology to a foreign person in the United States, treated as an export to that person's most recent country of citizenship or permanent residency (15 CFR 734.13(a)(2) and (b)). Deemed exports, ITAR and encryption questions are answered under export controls and technical data licensed to AI buyers.
The 2025 BIS Rules on Advanced Chips and Model Weights
Two Bureau of Industry and Security rules of January 2025 concern semiconductors and AI, and their published abstracts address chips, model weights and foundry due diligence, not a seller's run records.
The interim final rule "Framework for Artificial Intelligence Diffusion" (Federal Register document 2025-00636, published 2025-01-15) revises the EAR's controls on advanced computing integrated circuits and adds a control on AI model weights for certain advanced closed-weight dual-use models.
A second interim final rule (2025-00711, published 2025-01-16) adds due diligence procedures to assist foundries and OSAT companies on advanced computing integrated circuits. The current status of either rule is not covered here.
Benefits, Risks, Downsides and Alternatives for a Semiconductor or Materials Company
Licensing can earn income from builds a company has already run, failed ones included, but customer and foundry terms, recipe secrecy and export classification carry risks a fee does not offset, and not licensing is a valid outcome.
Benefits
- Failed builds and repeat runs are named as wanted: Mercor's listing asks for every build kept, failed ones included, and nice-to-have repeats of the same recipe on different days (read 2026-11-17).
- A range is stated for run records: Mercor states USD 100,000 to 750,000, a claim and not a paid amount (checked 2026-11-09).
- Wound-down labs qualify: Mercor states that labs and startups that have wound down qualify.
Risks
- Breaching a customer, foundry or IP license, since Mercor excludes customer- or foundry-owned data and licensed third-party IP, and 18 U.S.C. 1839(5) reaches disclosure under a secrecy duty.
- Loss of secrecy under an irrevocable grant, as in micro1's s.1.2 and s.4.4 terms or Labelbox Prism's no-unlearning statement.
- A release of controlled technology to a foreign person at the buyer, which is a deemed export under 15 CFR 734.13.
- Security incidents at buyers: the California Attorney General report on Mercor lists unauthorized access through compromised LiteLLM versions, with breach dates 2026-03-24 and 2026-03-30, and the lead case is In re Mercor.io Corporation Data Breach Litigation, N.D. Cal. No. 3:26-cv-02821-WHO.
Downsides
- Classification file by file before scoping, because no record type has a stated ECCN here.
- The 1,000+ run floor and the dated history Mercor asks for exclude smaller or newer archives.
- Trained models cannot unlearn what they learned: Labelbox Prism states so (checked 2026-11-09).
Alternatives
- License only records the company owns and has cleared as published or uncontrolled.
- License a de-scoped set that leaves out customer lots.
- Restrict access to named persons by contract.
- Wait for the company's own export classification before offering anything.
- Not licensing at all, which is a valid outcome; see alternatives to selling data, including not selling at all.
Where Fab and Materials Licensing Meets Neighboring Industries, Data Types and the Seller Path
Semiconductor and materials companies share process records with manufacturers and run-record questions with biotech labs, and the industry index sets out how each sector's limit differs.
The index of data licensing by industry also covers manufacturers licensing industrial data, biotech and life sciences labs licensing data and engineering and architecture firms licensing data, and a page on hardware companies licensing CAD and test data follows.
Published papers are scientific data and papers as AI training data, and board and mechanical designs are CAD and engineering files as AI training data. Trade-secret claims in the US run under the Defend Trade Secrets Act and licensed training data, and in the EU under the EU Trade Secrets Directive and licensed datasets.
Where run records carry personal data such as operator names, the DOJ rule at 28 CFR Part 202 (effective 2025-04-08) is a separate check for buyers linked to countries of concern, set out under the DOJ bulk sensitive data rule.
- Check data sale readiness.
- Count runs and systems in the data inventory for an AI data sale.
- Sign NDAs before sharing data with an AI buyer.
- Describe the runs in the data manifest.
- Set named access and encryption under security during an AI data sale.
- Negotiate the clauses of AI data licensing agreements.
This page is information, not legal advice or an export classification.
Buyer, Program and Law Pages Named on This Page
Each program, term set and case above has its own page with dates and sources.
- Handshake AI
- Labelbox Prism
- Mercor
- Mercor data breach class actions
- micro1 Data License Master Terms
- Program terms compared
Frequently Asked Questions
Can a semiconductor company license its process data to AI companies?
Yes, for run records it owns and has not published. Mercor's listing asks for 1,000+ runs, builds, wafers or experiments with settings linked to results, and excludes customer- or foundry-owned data (checked 2026-11-09, re-read 2026-11-17).
How much do AI companies pay for semiconductor run records?
The deal tracker lists no semiconductor price from a filing or docket as of 2026-11-17, and the only figure on record is a claim. Mercor states USD 100,000 to 750,000, "Depends on run count and how well runs link to results" (checked 2026-11-09). The wider picture is under what company data is worth to AI companies.
Can a wound-down hardware or materials startup sell its lab records?
Yes, through whoever holds authority over its assets. Mercor states that labs and startups that have wound down qualify, and Labelbox Prism states that distressed and wound-down companies are eligible. The routes are under selling data from a shut-down company.
Does licensing run records to an AI buyer need an export license?
It depends on the records' classification and on who at the buyer sees them. Published data is outside the EAR, while releasing controlled technology to a foreign person in the US is a deemed export.