DataPacts

Sensor, IoT and Telemetry Data for AI Training

Checked

At a glance

Sensor, IoT and telemetry data are the time-stamped readings that machines, devices and vehicles write as they run, and AI developers acquire them mainly for industrial AI and robotics.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

Mercor states that its manufacturing listing asks for machine alarm logs and process data with several years of history, and Labelbox Prism lists robotics telemetry among the data it wants (both checked November 9, 2026).

These records are usually licensed rather than sold outright, and two limits decide most cases: which party controls the machine data, and whether the set carries location. This page sets out the acquirers, the stated terms and those limits for machine records in AI training data licensing. Not licensing is a valid outcome.

The DataPacts Editorial Team explains how we source, date and check every figure.

What Counts as Sensor, IoT and Telemetry Data in an AI Data Deal

In a licensing deal, sensor data means records a company already holds from its own machines, products or fleet: process and alarm logs, readings from connected products, vehicle and device telemetry, and capture recorded with sensor rigs.

  • Industrial process and alarm records: the logs a plant system writes, which Mercor names in its manufacturing listing.
  • Run records with settings and measured results: the build and experiment histories that Mercor names in its semiconductor listing.
  • Connected-product and device telemetry: readings that a product sends back to its maker, governed in the EU by the Data Act rule below.
  • Sensor capture recorded to order: depth and motion data recorded by vendors such as Human Archive and Mecka with their own hardware.

Robot logs and teleoperation sessions belong to a neighboring type. They are covered under robotics data for AI training: teleoperation, manipulation, so this page covers machines, devices and fleets without a robot.

Which Acquirers Name Sensor Records, and in What Role

Seven companies name sensor, telemetry or machine records on their own pages, and they play six different roles, from a company data program that pays holders to a factory service that the plant pays.

The table compares each company's role for sensor data as of November 16, 2026, with programs checked November 9, 2026, in alphabetical order.

Company Role for sensor data What it names, acquires or handles Source
AIChamp paid factory-study service; does not acquire plant supplies sensor readings, production data, drawings and videos and pays the service; terms: AIChamp "does not sell, license or share Your Data" outside the project aichamp.com/terms.html
Human Archive (Human Archive, Inc.) data vendor with contributor app depth and motion data with first-person video, licensed to AI and robotics labs humanarchive.ai/privacy.html
Labelbox Prism (Labelbox, Inc.) licensee, company data program robotics telemetry among wanted data prism.labelbox.com
Luel (Luel Inc.) dataset vendor and marketplace for third-party listings Sensor filter shows 4; 30 datasets listed, category filters sum to 33 luel.ai/marketplace
Mecka (Carbon Based Technology Corporation d/b/a Mecka AI) capture vendor, sells datasets human demonstrations recorded with its own multi-sensor hardware mecka.ai
Mercor company data program machine alarm logs and process data; semiconductor run records Mercor listing feed
The General Data Company (General Data Labs, Inc.) RL-environment builder training data built from industrial companies' operating records; private models for them thegeneraldata.com

Two groups of acquirers are on record: robotics (Prism, Mecka and Human Archive) and industrial AI (Mercor's listings and The General Data Company's work from operating records). Luel states 4 datasets under its Sensor filter, and its own counts differ, 30 datasets against filters that sum to 33, so both numbers stand.

Capture vendors record their own data, so a company holding existing records is their customer, not their supplier. Luel lists third-party datasets and handles them rather than buying them, and AIChamp acquires nothing from the plant.

What Industrial Programs Ask For: Alarm Logs, Process Histories and Run Records

Mercor's two industrial listings are the most specific statements on record: the manufacturing listing asks for machine alarm logs and process data with several years of history, and the semiconductor listing asks for 1,000 or more runs with process settings linked to measured results.

The table sets out each listing's requirements as of November 16, 2026, every value Mercor's own statement checked November 9, 2026.

Listing Records asked for History or volume Tools named Not a fit Source
Manufacturing & Industrial Company production or maintenance records the company owns; at least one of CAM programs with setup sheets, work orders with inspection reports, or machine alarm logs and process data several years; company size 20 to 2,000 employees Mastercam, Fusion CAM, Esprit, Maximo, Fiix, UpKeep, MES, OSIsoft PI data restricted by customer contracts; marketing material only api.data.mercor.com listings /manufacturing
Semiconductor & Materials Company run or build records the company owns, with recipe or process settings linked to measured performance; unpublished 1,000+ runs, builds, wafers or experiments; every build kept, failed ones included, with dates MES, SPC / yield systems, LIMS, electronic lab notebooks, Cadence, Synopsys, OSIsoft PI data owned by a customer or foundry partner; published papers and datasheets only; licensed third-party IP api.data.mercor.com listings /semiconductor

Both listings name OSIsoft PI, a plant historian, together with MES and maintenance systems such as Maximo, so historians and plant systems are the expected source of the records.

Data restricted by customer contracts and data owned by a customer or foundry fall outside both listings. The manufacturer's side of these deals is set out under manufacturers licensing industrial data to AI.

Stated Ranges, Exclusivity and Recall for Sensor Data Sets

No acquirer publishes a price per sensor, device, hour of telemetry or gigabyte; the figures on record are whole-deal ranges that each party states for a wider data category.

The table compares the stated terms as of November 16, 2026, each value the party's own statement checked November 9, 2026, in alphabetical order.

Program Pay or range (stated) Exclusivity and scope Withdrawal, deletion or recall Source
AIChamp factory studies none to the plant; the plant pays USD 4,870 (Second Opinion); the Whole-Plant Study, earlier published at USD 14,610, is no longer listed (checked 2026-11-17) limited license to deliver the project only deletion within 30 days after project end, except legally required records aichamp.com/terms.html
Human Archive contributor app payouts for approved submissions via Stripe and Circle; per-hour rate not published licensed to labs under agreements prohibiting re-identification kept for the license duration; not recalled on account deletion humanarchive.ai/privacy.html
Labelbox Prism targeted datasets "typically start around $50K"; large deals "can reach $1M+"; upfront fee plus revenue share exclusive by default; seller keeps ownership and continued use kill clause: deletion within 30 days on stated triggers; trained models "cannot be unlearned" prism.labelbox.com/faq
Luel marketplace revenue share not published license terms not stated not stated luel.ai
Mercor, manufacturing listing USD 100,000 to 1,500,000, "Depends on record types and history" owned records only not stated in the listing Mercor listing feed
Mercor, semiconductor listing USD 100,000 to 750,000 owned, unpublished run records not stated in the listing Mercor listing feed

Mercor states USD 100,000 to 1,500,000 for the manufacturing listing and USD 100,000 to 750,000 for the semiconductor listing, and Labelbox Prism states that targeted datasets typically start around USD 50,000 and that large-scale enterprise data partnerships can reach USD 1,000,000 or more (all checked November 9, 2026). The ranges are not averaged or converted to a unit price.

On recall, Prism states that its license is exclusive by default, that a kill clause forces deletion within 30 days on stated triggers, and that models already trained on the data cannot be unlearned.

Human Archive states that approved submissions are kept for the license duration and not recalled when an account is deleted. The same terms across data types are laid out under payouts and program terms compared, dated.

Who Controls Machine Data: Device Makers, Operators and Their Customers

Machine data is often held by one company, generated at another's site and governed by a contract between them, so the first question is which party may license it.

In the EU, Regulation (EU) 2023/2854, Art. 4(13) and (14), allows a data holder to use non-personal data from a connected product only under its contract with the user, and bars it from making that data available to a third party, such as an AI developer, outside that contract.

The rule applies from September 12, 2025, and Art. 3(1) requires access by design for products placed on the market after September 12, 2026.

The Act's full reach, including access rights and contract terms, is covered under EU Data Act and AI training data.

Outside the EU, the customer or service contract is the first document to check. Mercor lists data restricted by customer contracts as not a fit, and telemetry held on a device platform can be bound by the maker's terms. This section and the next are information, not legal advice.

Location Data in Telemetry: Privacy Limits Before Licensing

Telemetry that records where a vehicle, device or worker was is usually personal data, and several laws limit selling precise location even when the rest of the set is cleared.

The table compares six location rules as of November 16, 2026, with the instrument, section and date for each, EU first.

Jurisdiction Instrument and section Rule for location data Date Source
EU Regulation (EU) 2016/679 (GDPR) personal data, including location tied to a person, needs a legal basis applies from 2018-05-25 eur-lex.europa.eu
EU Directive 2002/58/EC, Art. 9(1) location data other than traffic data processed only anonymised or with consent in force 2002-07-31 (national transpositions) publications.europa.eu
US federal 28 CFR 202.205 precise geolocation of more than 1,000 US devices in 12 months is bulk; no licensing to countries of concern or covered persons, even de-identified effective 2025-04-08 ecfr.gov
Connecticut Conn. Gen. Stat. 42-520(a)(3), 42-521(a)(2) as amended by P.A. 26-64 precise geolocation sale ban effective 2026-10-01 cga.ct.gov
Oregon OCPA (ORS 646A.570 to 646A.589), per Oregon DOJ FAQ DOJ states HB 2008 bans sale of precise geolocation from 2026-01-01; OCPA applies to all motor vehicle manufacturers for vehicle-use data as of September 2025 FAQ checked 2026-11-16 doj.state.or.us
Virginia Va. Code 59.1-578(A)(6); 59.1-575 controller may not sell or offer for sale precise geolocation data; "sale" means exchange for monetary consideration text read 2026-11-16; VCDPA in force 2023-01-01 law.lis.virginia.gov

A license for money counts as a "sale" under Virginia's definition in 59.1-575.

EU: Personal Data and the ePrivacy Rule on Location

In the EU, location that can be tied to a person, including through a vehicle or device, falls under the GDPR, and telecom location data may be processed only anonymised or with consent.

The GDPR, Regulation (EU) 2016/679, applies from May 25, 2018, and Directive 2002/58/EC, Art. 9(1), has been in force since July 31, 2002 through national transpositions. The legal basis a set needs is explained under GDPR and AI training data.

US: State Limits on Selling Precise Geolocation and Federal Thresholds for Foreign Buyers

Several US states bar selling precise geolocation data, and a federal rule bars licensing it in bulk to buyers linked to countries of concern.

Va. Code 59.1-578(A)(6) says a controller shall "Not sell or offer for sale precise geolocation data concerning a consumer" (text read November 16, 2026). Connecticut's P.A. 26-64 adds a precise geolocation sale ban in 42-520(a)(3) and 42-521(a)(2), effective October 1, 2026.

Oregon's DOJ FAQ states that HB 2008 bans the sale of precise geolocation data from January 1, 2026, and that the OCPA applies to all motor vehicle manufacturers for vehicle-use data as of September 2025 (written as the DOJ's statement, checked November 16, 2026).

Federally, 28 CFR Part 202, effective April 8, 2025, treats precise geolocation of more than 1,000 US devices in 12 months as bulk data under 202.205.

Licensing it to buyers in or linked to China, Cuba, Iran, North Korea, Russia or Venezuela is prohibited even when anonymized or de-identified, as set out under the bulk sensitive data rule for foreign AI buyers.

Preparing a Telemetry Set for Review: Exports, Timestamps and Context

A telemetry set is reviewed by what it lets a buyer learn, so it travels with the context that makes readings usable: what each sensor measured, when, on which machine, and what happened next.

Mercor's listings tie value to history and linkage: several years of records, process settings linked to measured performance, and failed runs kept with dates (checked November 9, 2026). The tools they name point to historians and plant systems as the export source.

No acquirer page on record states a required file format, sampling rate or schema for sensor data, so none is stated here. Formats, transfer and acceptance are set out under delivering licensed data to an AI buyer: formats.

Licensing Sensor Data: Benefits, Risks, Downsides and Alternatives, Including Not Licensing

Licensing can earn from machine records that already sit in historians and fleet platforms, but location and customer contracts can block the set, and delivered data is hard to call back.

Benefits

  • The records already exist, because Mercor's listings ask for logs and histories a plant keeps anyway.
  • Stated ranges exist as claims: Mercor USD 100,000 to 1,500,000 for manufacturing records and Prism around USD 50,000 and up, both checked November 9, 2026.
  • Sets without personal data avoid the location rules in the table above.

Risks

  • Passing product data to a third party outside the user contract breaches the Data Act rule in the EU.
  • Selling precise geolocation falls inside the Virginia and Connecticut bans and the Oregon DOJ's stated ban.
  • A bulk US location set licensed to a covered buyer is prohibited under 28 CFR Part 202.

Downsides

  • Prism states its license is exclusive by default and that trained models cannot be unlearned.
  • Human Archive states approved submissions are not recalled on account deletion.

Alternatives

  • License only non-personal, owned records.
  • License aggregates without location.
  • Keep the records for an in-house or private model, as The General Data Company describes for its industrial customers.
  • Not licensing at all: see whether to sell or license your data at all.

Neighboring Data Types and the Seller Steps for Telemetry

This page covers records from machines, devices and fleets that a company already holds; robot logs, vehicle footage, aerial imagery, motion capture and medical device data each have their own page.

The steps for a telemetry set run in this order:

  1. Read how to sell or license company data, the complete seller guide.
  2. Check who owns the machine records through a rights audit before licensing.
  3. Remove identifiers and location under de-identification before licensing.
  4. Agree the license clause by clause, using AI data licensing agreements.
  5. Check the rules in each country through AI training data law by jurisdiction.

Frequently Asked Questions

Can a company license its IoT data to AI companies?

Yes, where the company controls the records under its user and customer contracts and has removed or cleared location. Mercor's manufacturing listing names machine alarm logs and process data (checked November 9, 2026).

How much do AI companies pay for sensor data?

No per-unit price for sensor data is published. Mercor states USD 100,000 to 1,500,000 for manufacturing records and Prism states that targeted datasets typically start around USD 50,000 and that large-scale partnerships can reach USD 1,000,000 or more, all checked November 9, 2026. Valuation methods and public price points are covered under what company data is worth to AI companies.

Is telemetry with GPS coordinates personal data?

Usually yes, when the coordinates can be tied to a person, driver or device. Anonymization changes that only if it is real. What it achieves legally is set out under anonymization vs pseudonymization for AI training data.

Do robotics buyers want machine telemetry?

Some do. Prism names robotics telemetry among the data it wants, and robot logs and teleoperation are covered with robotics data (checked November 9, 2026).

Advisers and agents can use the introducer form.