Nearly Half of SimpleClosure's Deals Now Include Data. Why Wait?

Where the 47% comes from#

On October 13, 2026, MIT Technology Review published a piece titled "AI labs are buying data from dead startups to train their models." In it, Dori Yona, CEO of SimpleClosure, said 47% of the firm's deals now include a sale of code or workplace data, up from something that was rare and, in his words, "somewhat taboo."¹

It helps to be precise about what that means. SimpleClosure is a company that helps startups dissolve. In April 2026 it added data sales through a marketplace called Asset Hub, which the magazine says has handled more than 150 deals, and the company expects to return more than $10 million to startups through it by the end of 2026.¹ The magazine does not say whether the 47% covers all of SimpleClosure's work or only Asset Hub, and Business Insider's version, relayed by AI Weekly, attaches it to Asset Hub's 150 deals.² If it is the latter, 47% of about 150 is roughly 70 deals, by our arithmetic. Either way it is one firm's customers, who chose to use a marketplace for selling what is left.

So the headline stat is real, but it is not "47% of startup shutdowns." We changed our title for that reason. The wider shutdown trend is also real: Carta and AngelList estimates cited by the magazine show shutdowns rising 26% to 56% a year in recent years.¹ SimpleClosure's own data, quoted by Crunchbase News, shows closures in the first quarter of 2026 at 2.6 times the first quarter of 2025.³

What gets sold#

SimpleClosure describes two categories on its Asset Hub.⁴

• Source code. Git repositories, ticketing systems, product documentation and production codebases.
• Workspace data. Operational data such as communications, workflows and documents that show how teams worked and made decisions. It launched in beta in April 2026 and moved to general availability in September.

AI buyers want the second category because of how they train agents. AI Weekly, summarizing The Information, describes "RL gyms," simulated workplaces stocked with a real company's Slack threads, Jira tickets and internal emails, so an agent can practice being an employee.⁵ Nick Heiner of Surge AI told MIT Technology Review that he looks for varied archives with disagreements and judgment calls.¹ The Information reported that Anthropic leaders have discussed spending more than $1 billion on reinforcement learning environments over the next year.⁵ That is a discussion reported from sources, not a budget Anthropic has announced.

The magazine adds a caution. No company has published tests comparing real-world data with synthetic alternatives, so whether these archives improve models is unproven.¹

Who is buying#

The named buyers in these reports are a short list, and none of them publishes a rate card for shut-down companies.

• Mercor offered to buy or license Warmly's code base and task records, and, per The Information, is one of several data vendors chasing shuttered startups' Slack threads and tickets.⁵ It was also the backup bidder at $7.5 million for Spirit Airlines' archive.⁶
• Surge AI buys data to build reinforcement learning gyms, according to MIT Technology Review.¹
• SpaceX's AI unit was reported by Bloomberg in September 2026 to be considering buying data from struggling startups, a report the magazine relays and we have not seen directly.¹
• micro1 runs a program for operating companies and filed a competing bid for the Spirit archive.⁷⁶

SimpleClosure says its own buyer network includes "leading AI foundation labs" and companies that build post-training and reinforcement learning environments, but does not name them.⁴ That is typical of this market. Sellers learn little about who ends up with the data, and the Spirit Airlines file shows how contested that becomes when the seller is large.

What it pays#

The public figures are thin and they are mostly from sellers or intermediaries.

Table 1. Reported payouts for shut-down company code and data, as of October 22, 2026.

Miskell had run his company for about two and a half years before winding down.¹ Cielo24's CEO, Shanna Johnson, is named in the Forbes report as the seller.⁵ See the deal pages for Miskell, Cielo24, SimpleClosure and Spirit Airlines.

The pattern: a typical small startup gets low five to low six figures, Cielo24's long archive reportedly got more, and a very large estate drew $10 million. Valuation, the magazine says, depends on speculative "research theses." Sunset, a rival wind-down firm, values assets by headcount, years in operation, data volume and how the team used its tools.¹ That is a seller's rule of thumb, not a published price list.

What the sellers are not asked#

Both the magazine and AI Weekly flag questions that the deals leave open.

• Employee consent. Whether employees consent to the sale of records tied to their work is unresolved, according to RuntimeWire.⁸
• Who gets the money. MIT Technology Review reports that proceeds typically go to senior creditors and investors, not employees. Ifeoma Ajunwa, a law professor at Emory, called this "captured capital" and noted no federal law bans such sales.¹
• How filtering works. AI Weekly says reporting does not establish how the gyms scrub employee names and customer conversations, and that coverage leans heavily toward the sellers' side because the labs say little about their filtering.⁵
• Deal terms. AI Weekly asks whether an acquirer's deal terms allow such a sale. In its example, Warmly agreed in late June to be acquired by HubSpot, and eight days later Mercor emailed Warmly's CEO offering to buy or license its code and task records. The reporting does not say whether Warmly accepted.⁵

SimpleClosure says assets go through multi-pass de-identification to replace names, emails, credentials, keys and customer identifiers, that it verifies chain of title, and that sensitive employee and health records are not sold, nor is anything the seller marks out of scope.⁴ That is the company's own description. MIT Technology Review reports a lawyer's view of the risk in plain terms: Greg Ewing, an attorney who advises AI startups, called this kind of data "legally radioactive."¹ The risk checklist covers how to approach that.

In bankruptcy there is a specific check. A company that promised customers it would not share their personal data with outsiders generally cannot sell it unless the sale fits that promise or a consumer privacy ombudsman is appointed and the court approves.⁹ Outside bankruptcy we did not find a source that says what rules apply to a plain dissolution, so ask counsel.

So why wait?#

The title asks it because the practical case for planning early is stronger than the case for any particular price. These are our reasons. They are inferences from the sources, not guarantees.

1. Workspace data can disappear#

Slack's help center says free plan workspaces can keep data for 90 days or one year, that data older than one year is deleted on a rolling basis, and that deletion is permanent.¹⁰ If a company is on a free plan, or on a paid plan with a short retention setting, the older history may already be gone by the time a sale is considered. A 13 year archive like Cielo24's implies someone kept the history on purpose.⁵ The point is to know what you keep, and for how long, before you need it.

2. Rights are easier to sort out while the company is running#

Customer contracts, privacy notices, staff agreements and vendor licenses decide what can be sold. Those documents are easiest to find, and the people who understand them are easiest to ask, while the team is still around. Founders in a wind-down are short on money, time and staff. AI Weekly's editor's note makes a similar point: operational data is now a priced line item at exit, and founders and acquirers should put explicit data-disposition terms in deal documents.⁵

3. Buyers and channels are opening to operating companies#

SimpleClosure said in September 2026 that Asset Hub is now open to companies at any stage, including operating, pivoting and winding-down ones, and that it is free for dissolution customers.⁴ Mercor's program is for operating companies in eight industries and pays based on data volume, tools connected and depth of data.¹¹ micro1 publishes tiers starting above $100,000 for companies with more than 30 employees.⁷ The shutdown route is not the only one.

4. The wind-down route pays little and offers little control#

By the figures above, a shut-down startup typically gets tens of thousands of dollars, and founders have little visibility into buyers' plans, a point the magazine makes directly.¹ Selling while you can negotiate, with a team to answer questions, is not guaranteed to be better. It is at least a position with options.

A short checklist for founders planning an exit#

1. Inventory your tools and retention settings. Slack, email, Jira, Notion, Drive, CRM. Note how far back each goes.
2. Collect the paper. Customer contracts, NDAs, staff agreements, privacy notices and vendor licenses.
3. Separate the risky parts. Customer conversations, payroll, health and HR material are the hardest to clear. SimpleClosure says it does not sell sensitive employee and health records,⁴ and lawmakers pressed Google to exclude Spirit's staff records from its purchase.¹²
4. Ask any intermediary how it is paid. SimpleClosure's Asset Hub page does not state fees or how proceeds are split.¹³
5. Get a lawyer to review the terms. Exclusivity, deletion, use limits and indemnity.

Our assessment can help sort data types and rights questions, and our main guide has a section on selling during a bankruptcy or wind-down. SellYourAIData has no deals to report. Take our view as that of a broker with an interest in the market.

Bottom line#

The 47% statistic comes from one firm, and the payouts behind it are small. The more useful part of the story is what sits behind the number: buyers now want workplace records, shutdowns are rising, and the questions about employees and rights are unresolved. If you run a company, the sensible move is to know what you hold, what you are allowed to sell and what is being deleted, whichever way you eventually go.