If you're shutting down a startup, you still own something AI developers want. The code, the pull requests, the tickets, the Slack threads and the docs are a record of how a real software team built and ran a product, and labs pay for that record because it doesn't exist on the public web. Most founders wipe it during the wind-down. Read on to find out who buys it, what it has paid, what to export before you cancel anything, and how to sell or license it without creating a problem for yourself later.
Who buys code and data from companies that are closing
AI labs are the end buyers. They're training coding agents on how real teams write, review and fix software, and training work agents on how real companies handle a support ticket or close a deal. A finished startup gives them something an operating company rarely will. They get the whole history, start to end, with nothing held back.
Three kinds of buyers sit between you and the labs. Closure platforms broker the assets of companies that are shutting down (code, workspace archives, sometimes the whole data room) and take a cut. Lab-run programs buy directly, and at least one large lab runs a program that takes full private repositories together with design docs, testing logs and wikis from companies that have closed. Data marketplaces like Troveo license company data to many buyers over time rather than selling it once, and we'll come back to why that difference matters. Our guide to selling source code to AI labs has the fuller picture of who's buying code and what they do with it.
What it has paid
There's no price list, but there are public reference points, and they cluster by what was sold.
| What was sold or licensed | Reported value | What it tells you |
|---|---|---|
| A single repository through a closure marketplace | About 5,000 dollars per repo | The baseline for a small, clean, self-contained codebase |
| A startup's code and workspace archive, sold on wind-down | Roughly 10,000 to 100,000 dollars per deal | The typical range for a few years of real engineering history |
| A larger company's complete archive, sold on wind-down | Hundreds of thousands of dollars in at least one reported case | What a bigger team's history with full context has fetched |
| A full company's operating data, licensed while operating | Typical deals start at six figures | The floor when the data is licensed rather than sold |
| Spirit Airlines' operational archive in bankruptcy court | 10 million dollars (Google's bid) | What a large company's complete history drew as one purchase |
What code and company data from startups has sold or licensed for, from a single repository to a full archive.
The closure-market figures come from reporting on the platforms that broker these deals, and the six-figure floor comes from our business data page. The Spirit number reflects an airline with decades of operations, so don't read it as a target; read it as proof that operating history has a market value even in a liquidation, and our breakdown of the Spirit Airlines data deal explains why the archive drew a bid at all. Where your company would land depends on the same six factors that price any company's data, and our guide to how much your company's data is worth to AI labs walks through each one.
Sell it once or license it more than once
You have two ways to turn the assets into money, and they pay differently.
Selling is what closure platforms do. The buyer takes ownership of the code and the archive, you get one payment, and it's done. It's simple, it's fast, and it's the right answer when the company is dissolving and nobody will be around to hold the rights.
Licensing keeps ownership with the company (or with whoever takes the rights on when it closes) and grants a buyer the right to use a copy, with names and personal details removed, for defined purposes. The same archive can then be licensed to more than one buyer. It can also become more than one product (a historical export, an enriched dataset, evaluation data built on it), and you're paid on each sale. That's why we treat data licensing as revenue you earn repeatedly, and it's why our business data page puts typical full-company deals at six figures with every later product a new sale. The catch for a closing company is that someone has to keep holding the rights and signing the agreements, so if licensing is the better fit, set it up before the entity dissolves. Our guide to exclusive vs non-exclusive data licenses explains what each option pays over time.
If you're still deciding whether to close, the calculation changes again. An operating company can license selectively (a retired product line, a replaced system, an internal tool with years of tickets behind it) and keep running on the data, and our guide to how tech startups and SaaS companies license their data to AI shows what's in a typical stack and what has to come out.
Export everything before you cancel anything
Do this first, before the sale conversation, because the wind-down checklist works against you. Standard practice is to cancel the SaaS subscriptions and wipe the servers, and once Slack, Notion, Jira, HubSpot or Zendesk lapses, the history in them is gone or locked behind a reactivation you may not be able to afford.
Export the code with its full history (every branch, pull request, review and issue, not a zip of the latest commit). Export the project tool, the support desk, the CRM, the wiki and the chat workspace. Pull the data warehouse and any internal databases. Keep the exports together with a short note of what each system was used for and over what dates. Buyers pay for the connections between systems, and the glossary entry on workflow trajectories explains why a ticket linked to its Slack thread, its Jira issue and the pull request that closed it is worth more than any of those on its own.
Don't clean anything up yet. The messy parts (the bug that took three tries, the escalation, the reverted release) are what make the record useful. Removing names and personal details comes later, and it's done to a documented standard, not by deleting history.
What a buyer checks before paying
Buyers of a closing company's assets ask the same questions a licensing buyer would, and a company that can't answer them sells for less or not at all.
They'll check the chain of ownership. Every contractor and early employee should have signed over the IP they wrote, and if someone didn't, fix it before you list the assets. They'll check third-party code, and anything under a restrictive open-source license, or copied from a client, comes out or gets disclosed. They'll check for secrets (API keys, credentials, tokens) in the repo history and expect them removed. They'll check what customer data is in the archive and whether your customer contracts allow any of it to be transferred, and most of the time the answer is that customer data stays out entirely. And they'll check that whoever signs can sign, which means board or investor approval for an asset sale, no liens on the IP, and a company that still exists at the moment of transfer.
The documented history of where the data came from and what rights attach to it is the thing buyers pay a premium for, and our guide to AI data provenance describes what that documentation looks like.
What has to come out
Customer personal information, employee personal information, credentials and secrets, third-party confidential material, and anything a customer contract restricts all come out before delivery. For most startups, that means the product's user data and the customer records are out of scope, and what sells is the record of the company's own work (engineering, internal support, sales process and decision-making). That's usually the more valuable part anyway.
Names, contact details and client identities in Slack, tickets and email get scrubbed. When we license company data, that scrubbing is done to a documented standard before anything reaches a buyer, and your company isn't named publicly; buyers are introduced confidentially after an agreement is signed. If you sell through a closure platform instead, ask who removes the personal details and to what standard, because it's your liability if it's done badly. The full list of what to exclude and how the rights review runs is in our owner's guide to licensing company data for AI.
How long it takes
Weeks, not days, and the rights work sets the pace. The exports take a few hours if the subscriptions are still live. The ownership check takes as long as it takes to find the contractor agreements. A buyer will want to evaluate a sample before committing, and a clean, documented repo with its history moves faster than a large archive that needs sorting. If you're running against a dissolution date, start the exports and the ownership check now and let the buyer conversation run in parallel.
Where to start
Run our free data value assessment first. It takes about five minutes, asks ten questions about your systems, history and industry, and gives you an estimate of what the data could license for. It'll tell you whether the archive is worth a licensing conversation or whether a quick sale through a closure platform is the better use of your remaining time.
Where Troveo fits
Troveo helps companies understand what proprietary data they hold, protect what matters, and selectively license what's valuable. We've paid more than 20 million dollars to rights holders across video, audio, gaming, robotics and business data, we work with more than 40 active buyers, and we charge no fees and take no deductions from your payouts. For a company that's closing, we can scope the archive, remove the personal details, package it, and license it to buyers on terms you approve, and you're paid on every sale. Start with the data value assessment, or talk to us before the subscriptions lapse.