Can a company sell its source code to AI companies, and what do buyers want?

The short answer: companies license code history, they rarely sell it

A company can earn money from its source code with AI developers, but the usual route is a license, not a sale. The company grants an AI lab or data buyer the right to use a copy of selected repositories for AI training, typically on an exclusive basis for that purpose for an agreed term, and keeps ownership of the code and the product built on it.

That distinction answers the first objection most founders raise. Licensing the history of a codebase does not hand over the product, transfer the intellectual property or stop the company from shipping, selling or being acquired. The company sets the scope, agrees one all-in price and signs only if the terms work.

What AI buyers want from a private codebase

The final snapshot of the code is the least interesting part. Coding agents learn from how software changed and why.

A company with 50+ full-time employees at peak, a long-lived product and disciplined review habits usually holds far more of this than a young startup. The GitHub system brief describes what a well-used organization account tends to contain.

What has to be screened out first

Most of the work before a code license is exclusion. The company and its counsel decide each item; a partner never reviews or forwards code.

• Open-source and vendored libraries, which carry their own license terms and are excluded or handled as counsel directs.
• Client-owned code: agencies and consultancies often assign the code they write to clients under statements of work.
• Contractor code without a written assignment. The Copyright Office's circular on works made for hire explains that work an employee prepares within the scope of employment belongs to the employer, while commissioned work qualifies only in listed categories with a signed written agreement, so contractor-written code may need an assignment on file.
• Secrets in history: API keys, passwords, certificates and production data in test fixtures, removed across every commit rather than only the latest one.
• Personal data in commit metadata, such as names and email addresses, handled under the de-identification rules agreed before work begins.
• Third-party SDKs or code received under a nondisclosure agreement.

Ownership can also be divided. Under 17 U.S.C. 201, any exclusive right in a copyrighted work can be transferred and owned separately, which is why a company can grant AI-training rights in code it owns while keeping everything else. This is general information, not legal, tax or financial advice. Confirm with your own counsel before acting.

How a fractional CTO can raise it

Fractional CTOs see the moments when code history is at risk: a monolith being retired, repositories moving between hosts, history being squashed to tidy up a migration, or a product line being sunset. Those are the moments to ask.

The referral overview for fractional CTOs covers the rest of the workflow, and an engineering lead can use the data inventory builder to list repositories, date ranges and linked systems as metadata only. If the company also holds support tickets, documents or other records worth licensing, see how to sell data to AI companies.

When the objection is right

Sometimes no is the correct answer:

• The company is below the size bar of 50+ full-time employees at peak (contractors excluded); the who qualifies page lists the complete criteria.
• Most of the code was written for clients who own it.
• History was lost or squashed in an old migration, leaving only a recent snapshot.
• The repositories are mostly forked open-source projects with light changes.
• The code was generated with AI tools specifically to sell it.
• The same repositories were already licensed for AI training.
• Nobody with authority can approve a license, for example because a lender or trustee has taken control of the assets and is not yet part of the conversation.

Next step

If a company you advise passes these checks, register as a partner and introduce its CEO or CTO, or have the company apply at sourcex.si/apply with your referral link. Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company, and rewards become payable only after the buyer pays and SourceX receives its fee.