Companies close every week, and their records do not close with them on the same day. Slack workspaces, mailboxes, repositories and shared drives sit inside subscriptions that are about to lapse. For a broker or a data company, the useful question is not which startups failed. It is what happens to their records, who may legally sell them, and how quickly the window shuts.
Working with a company that is still running? Qualify it against micro1, Mode and Grepped rules in about 20 seconds.
Most startup graveyards are lists: a name, a logo, a year and a one-line cause of death. They are popular because they are quick to read, and because failure stories are good stories. For anyone who buys or brokers company data, they are close to useless, and some of them are risky to rely on.
A name on a list tells you a company is said to have closed. It does not tell you who controls its records now, whether anyone exported them before the subscriptions lapsed, whether a lender has a claim on them, or whether the customer contracts require them to be deleted. Those are the facts a data deal turns on. A list also goes stale quickly. Some companies on it were acquired rather than closed, some are still winding down, and some came back under a new name.
There is a second reason this page names no closed startups. Calling a specific company failed is a claim about that company, and a wrong one hurts real people who may be trying to sell assets, pay creditors or find new jobs. The only named closure case on this page is the Spirit Airlines bankruptcy data sale, because it ran in public through a court.
Every figure is attributed to whoever published or cited it, as published, checked 7 October 2026. None of it is a price list.
Forbes reported that shut-down startups were selling their old work archives, with Slack and email as the headline, for use as AI training data.
Both outlets covered the same development. A closed company still holds years of records of real work, and AI developers say that is what they want.
Troveo cites about $5,000 per code repository and roughly $10K to $100K per archive deal in the shut-down startup market. One participant's figures, not an average.
Deedy Das mapped 50+ companies that sell data and RL environments to labs, with about $8.5B in revenue, 75% of it held by Scale, Surge, Mercor and Handshake.
Google agreed to pay $10M for the bankrupt airline's internal data. Because the airline was in bankruptcy, the sale needed a court's approval.
A data company bid against a lab for the same archive. Approval was not confirmed as of 7 October 2026. Details are in our Spirit Airlines case file.
Read in order, the year shows two things. First, archive data became a recognized asset in a closure, from five-figure startup archives to an eight-figure sale supervised by a court. Second, the buyers are not only labs. A data company bid against Google for the Spirit records. That is the market brokers work in: data companies that need supply, and companies, some of them closing, that hold it.
The coverage also showed where the friction sits. The people whose messages fill an archive usually have no say once the company is gone, and they notice when it is sold. A deal that ignores them invites objections, and objections slow everything down. For a fuller account of what was reported and what people objected to, read our case notes on shut-down startups selling Slack and email.
Without a plan, most of a shut-down company's records follow the subscriptions that hold them. Here is the usual path for each system, what makes it worth something, and the catch.
A closed AI company may hold things a traditional startup does not: evaluation sets, fine-tuning datasets, labeled examples, prompt libraries, model checkpoints, and logs of how users worked with the product. Some of that is exactly what data companies resell to labs. Much of it comes with strings attached. Training data licensed from third parties may come under terms that end with the company. Data collected from the open web may carry rights nobody checked. Customer prompts and uploads are customer data, under the customer's contract and the privacy policy.
So for an AI startup's archive, the first question is not what it is worth. It is which parts the company had the right to keep, and which it had the right to pass on. The part that is almost always the company's own is the same as at any startup: its code with history, its internal documents, and its work tracking and support records, once personal and client data are set aside.
A closure runs on weeks. A data deal runs on months. If the two calendars are not lined up, the archive is gone before a buyer's review is done.
The board decides. Nothing has been deleted, and the people who know the systems are still on payroll. This is the best moment to make a data decision, and the one most often missed.
Admin rights and system knowledge leave with the people who held them. After this, nobody can say which channel was which, or run a clean export with the links between systems intact.
Software is one of the first costs a closing company cuts. Each vendor's terms decide how long data survives after cancellation, and those windows differ by vendor and by plan.
Workspaces, mailboxes and drives are deleted. What was not exported is gone. What was exported often sits on one person's drive, with no owner and no plan.
Once the company no longer exists, there may be nobody with authority to sign a license, even if the archive survived somewhere.
Four kinds of buyer show up around closures. Figures are as published or as cited, checked 7 October 2026. They are ranges and citations, not offers.
| Buyer type | What it buys | Published or cited figures | Ask first |
|---|---|---|---|
| Closure-market buyers | Archives of shut-down startups: chat, email, documents and code | About $5,000 per code repository and roughly $10K to $100K per archive deal, as cited by Troveo | Which systems and date ranges it takes, and whether it needs anyone available after the closure |
| Data companies with buyer programs | Scoped, de-identified records of running companies | micro1 "$100K-$2M+ for approved data packages"; Mode "$100K-$5M"; Grepped "$20K-$5M" | Whether it reviews a company that is winding down at all. The published rules describe operating companies |
| Codebase and dataset buyers | Private codebases and operating datasets | Miro Advisory: codebases $10K-$1M+, operating datasets $100K-$1M+ | Whether history, reviews and issues are kept, and whether licenses and secrets have been checked |
| Labs and data companies in court-supervised sales | Large internal archives of insolvent companies | Spirit Airlines: Google agreed $10M; micro1 filed a $12.5M competing bid | Who runs the sale process, and what the court has to approve. Approval in the Spirit case was not confirmed as of 7 October 2026 |
Why data companies, not only labs, are natural buyers of closure archives.
Practitioners describe a ladder. Raw data is the cheapest form. Evaluations built from it are worth about 10 times raw, and full environments reach six to eight figures. A closure archive sits on the bottom rung: raw, finished, and with nobody left to grade examples or answer questions about context.
That is why the buyers of archives are increasingly data companies. They have the specialists to turn raw records into evaluations and environments, and the lab relationships to sell the result. For a broker, it means the realistic counterparty for a closing company's archive is usually a data company or a closure-market buyer, not a lab. The Spirit case, with a lab and a data company bidding for the same records, shows how far up the ladder a large archive can reach.
Market figures as published, checked 7 October 2026. Practitioner figures describe the market in general, not any one deal.
The records belong to the company. Who acts for the company changes as a closure moves along, and so does who must agree. General information, not legal advice: insolvency, corporate and privacy rules differ by state and country.
| Situation | Who usually signs | Who else has a say | What a broker should see in writing |
|---|---|---|---|
| Solvent wind-down | The board, acting for the company | Shareholders, and investors with consent rights in the financing documents | A board resolution to explore a data sale, and who is authorized to sign |
| Company acquired | The acquirer, once the deal has closed | The acquirer's legal and privacy teams | Confirmation that the records transferred, and who now controls them |
| Assignment for the benefit of creditors | The assignee that now holds the assets | Creditors, whose claims the proceeds go toward | The assignee's authority and how it runs asset sales |
| Bankruptcy liquidation | The trustee | The court and the creditors | Court filings showing the trustee's authority and any approval the sale needs |
| Sale in bankruptcy by the company | The company, with court approval | The court, creditors and objectors; where personal data is sold, a privacy review may be required | The sale motion and the approval order. The Spirit sale is this kind |
| Secured lender enforcing | The lender, through its enforcement process | The company and other creditors | The security agreement, and whether it covers data and IP |
Referral fees for company data are published and large. micro1 publishes "earn $50,000" per referred company with "no cap", paid after onboarding plus a minimum revenue threshold. Mode publishes "$50K per referral". Grepped publishes "Refer for another $10K". All as published, checked 7 October 2026.
Those programs describe running companies. micro1 asks for 30+ employees and Mode for 20+ full-time US office employees with several years of records. Neither page we checked says how it treats a company in wind-down, so the first question to a program is whether it will review one at all. If not, the realistic counterparties are closure-market and codebase buyers, and you should know before you raise anyone's hopes.
The terms matter as much as the amounts. micro1's terms give it sole discretion over every referred company, allow clawbacks, forbid sharing payouts with the company you refer, and forbid presenting yourself as micro1's partner. A closing company short on cash may ask you for a share of your fee. Under those terms, the answer is no.
Then there is conduct. The people across the table may be losing their company and their jobs the same month. Say plainly what you do and how you are paid, ask once, and accept a no. For the roles in this market, read what is an AI data broker, and for published fee terms, data referral programs.
Every score gives a company a number from 0 to 100, a grade, the data it likely holds, its history and one of four activity statuses. Activity status is one of the nine factor groups, with history, scale, knowledge assets, operational systems, customer systems, organization, industry value and expertise.
The company looks like it is operating.
Next: if a closure rumor brought you here, the rumor may be wrong or early. Talk to the company, not about it, and qualify it like any running business.
The company looks like it is closing or has become part of another company.
Next: find out who controls the records now: the board, an acquirer, an assignee or a trustee.
The domain no longer presents an operating company.
Next: any surviving records sit with whoever controlled the company at the end. Without a person with authority, there is no deal.
The site could not be reached when it was checked.
Next: this proves nothing on its own. Check again later before drawing any conclusion.
# One lookup: score and activity status for one company curl -H "X-API-Key: YOUR_KEY" \ "https://www.selldatatoai.com/api/v1/score?domain=example-startup.com" { "domain": "example-startup.com", "data_asset_score": 57, "grade": "C", "status": "winding_down_or_acquired", "history": { "first_seen_year": 2016, "years_online": 10 }, "likely_data_assets": [ { "type": "code_history", "confidence": "high" }, { "type": "support_tickets", "confidence": "medium" }, { "type": "knowledge_base", "confidence": "medium" } ] }
Searches for companies going out of business in 2026 rise whenever a well-known name closes. For a data buyer, those headlines are the least useful part of the market. By the time a closure is news, the decisions that matter, the exports, the exclusions and who signs, have usually been made or missed. Most closures never make the news at all: a small team winds down, pays what it owes and dissolves.
If a headline does bring a company to your attention, treat it as a prompt to check, not a lead to pitch. Check the activity status. Read what the company itself has said. Find out whether a formal process is running, because then a trustee, assignee or court controls the assets, and the people you would normally call cannot agree to anything.
It tells you a company is closing. It does not say whether the records survive, who controls them, or whether anything can be licensed at all.
Most companies close without coverage. The only reliable way to know where a company stands is to check the company itself, and then ask it.
A broker who adds noise to a closure does not get a second conversation. One plain message to the person with authority is enough.
Usually a list or archive of startups that have shut down, each with a name, a year and a cause. This page uses the phrase for data buyers and brokers instead: it covers what happens to a closed company’s records, who buys them, who may sell them and how fast they disappear. It names no closed startups.
No. We do not publish or sell a list of failed or closing companies. Every Data Asset Score shows a company’s activity status: active, winding down or acquired, parked, or unreachable. You check the companies you are already looking at, one at a time in the demo or through the API.
Without a plan, they follow the subscriptions. Once the workspace and email accounts are cancelled, each vendor’s terms decide how long the data survives before it is deleted. Anything worth licensing has to be exported by an admin before cancellation, with the links between systems kept intact.
Closure-market buyers that purchase archives, with Troveo the participant whose prices were cited in 2026; codebase and dataset buyers such as Miro Advisory; and, in larger court-supervised sales, labs and data companies, as the Spirit Airlines bids showed. Buyer programs such as micro1, Mode and Grepped publish rules for running companies, so ask before assuming they review a closing one.
Troveo cites about $5,000 per code repository and roughly $10K to $100K per archive deal in the shut-down startup market. Those are one participant’s figures, not an average and not an offer. Nobody can price an archive without seeing a manifest and samples.
The company, acting through whoever controls it at that stage: the board in a solvent wind-down, the acquirer after a sale, an assignee in an assignment for the benefit of creditors, or a trustee, or the company with court approval, in bankruptcy. Founders acting personally, former employees and brokers cannot. General information, not legal advice.
In August 2026 Google agreed to pay $10M for the bankrupt airline’s internal data, and micro1 then filed a $12.5M competing bid. Approval was not confirmed as of 7 October 2026. Our Spirit Airlines case file covers the details.
Only if a buyer program accepts the company and the deal meets the program’s conditions. micro1 publishes “earn $50,000” per referred company, paid after onboarding plus a minimum revenue threshold, at its sole discretion and with clawbacks. Published rules describe operating companies, so confirm a program reviews a company in wind-down before you introduce one.
See a company's score, likely data and activity status in the demo, qualify a running company against published buyer rules, or pick a plan for lookups at volume.