Who Buys Code for AI Training: Buyers Compared

Who buys code for AI training

Every row re-read 2026-11-07

Five programs post terms for code or engineering records: micro1, Labelbox Prism, Scale AI, Nyne and SimpleClosure AssetHub. The split is between those that want source itself and those that want the work around it, with staff minimums running from 20 to 40. SimpleClosure alone is designed for closed companies.

The comparison

Code and engineering data buyers compared (fetched 2026-11-07)
Buyer	What its page lists	Published pay	Who qualifies	Personal data	Payment
micro1	Engineering workflows, bug tracking, source code documentation; also product planning and agent trajectories	Entry tier $100K–$500K; $500K–$3M when the dataset is large or spans teams; above $3M if truly unique	Needs 30+ people on full-time payroll and records mostly in English; US, UK and Canada favored	De-identification done by micro1, with sample review offered to the seller	Full payment within a month (30 days) of acceptance
Labelbox Prism	Traces of code being generated and edited, plus reasoning and workflow sets	Roughly $50K at the small end, $1M+ for big enterprise deals; cash at signing, then a share of downstream revenue	No size or age threshold on the page	Labelbox scrubs PII during ingestion. Exclusive is its default license shape	Under 21 days, call to wire, by its own claim
Scale AI	Jira and Confluence records, docs, chat; it will not take credentials or repo access	$10K–$1M+, which Scale calls illustrative value rather than a promised sum	Ideally 40 or more people, though smaller firms are judged individually	Scrubbing runs on your side of the wall; each package is approved before release	Not published
Nyne	Jira and Asana projects, tickets, CRM, and accounting kept in QuickBooks, Xero or NetSuite	Usually $100K and up, scaled by how much data and what kind	Twenty or more US W-2 staff, plus a 3-year trading record	De-identified before it is released; who does it is not stated	Not published
SimpleClosure AssetHub	Repos with their docs and tests, and the workspace around them: Slack, Drive, mail, Notion, Figma	Not published	Founders closing their company with SimpleClosure	Each dataset scrubbed for personal data before licensing	Not published

Disclosure: several of these companies run referral programs and could pay this site after a deal with a company we introduced. Rows are in no paid order, and fees do not shape the wording.

Independence note: none of these five companies owns, sponsors or has approved this site, and we have no affiliation with any of them. The method is reading their public pages; we run no tests or audits.

Which buyers take source code itself?

Read the lists closely; "code" is a slippery word across these pages.

SimpleClosure AssetHub is the literal one: repositories, their docs and their tests, alongside the company's Slack, Drive, mail, Notion and Figma.
Labelbox Prism asks for traces of code being generated and edited, the nearest match to a commit-and-review history.
micro1 names engineering workflows, bug tracking and code documentation, next to product planning and agent trajectories.
Scale AI wants Jira, Confluence and similar work records; raw systems, credentials and source access are refused.
Nyne counts Jira and Asana projects within a broader set of operating records.

Which suits an operating company?

With thirty or more people on payroll, the micro1 and Scale AI minimums are probably within reach; Nyne's bar sits lower, at twenty W-2 staff and three years trading. Prism has no minimum and starts near $50K for a small, focused set, a fit for something narrow like the review history of a single repo.

Which suits a company that has shut down?

Start with SimpleClosure AssetHub, built around founders who are closing down. micro1 has a defunct-company track too, though with no pay band of its own on the page. For price, there is only journalism: one crypto founder cleared about $25,000 for the code of his closed startup, with SimpleClosure arranging the sale (MIT Technology Review, 13 Oct 2026). The wind-down page goes further.

How do they handle personal data?

Every one of the five promises removal; where it happens differs. Scale does the work on your infrastructure, before extraction. micro1 does it in-house and offers you sample review. Prism scrubs during ingestion. Nyne de-identifies ahead of release without saying who performs it. SimpleClosure runs a PII pass per dataset. Committers and reviewers are real people, so make the first sweep yourself.

What changed recently?

Two moves at micro1: a bigger company referral fee than at the June 2026 launch, and a headline band that now runs $100K–$3M+. Dates and details sit in the terms changelog.

Short answers

Which AI companies buy source code?

Of the programs that post terms, SimpleClosure AssetHub is the one that names repositories outright, with their docs and tests, for companies being wound down. Prism wants traces of code being generated and edited. micro1 asks for engineering workflows, bug tracking and code documentation. Scale AI and Nyne want work records such as Jira tickets and refuse repo access, or never mention it.

Do AI labs buy code directly from startups?

Rarely in public. The labs mostly buy through data companies (micro1, Scale AI, Labelbox) and, for closed startups, wind-down services like SimpleClosure. Few labs post any seller-facing terms of their own.

What is the smallest company these buyers work with?

The lowest posted bar is Nyne's (twenty W-2 employees in the US, three years old or more). micro1 wants thirty full-timers; Scale AI would like forty. Prism and SimpleClosure post no minimum.

Is the published pay what we will get?

No. Those figures describe entire data partnerships, Scale's is explicitly illustrative, and real license fees often land far lower. For a closed startup's code, the only press-reported price is about $25,000.

Where is Mercor?

Its company data pages would not load for us on 2026-11-07, and nothing reaches this table from anywhere but the buyer's own site.