Startup shutdowns in 2026: what the numbers show and where the records go

What the 2026 startup shutdown numbers measure

There is no official government count of venture-backed startup shutdowns. Federal statistics count every business, and the venture-specific figures that circulate in tech press come from private datasets, usually cap-table platforms and venture databases, each covering only the companies in its own records. Read any headline number with its population, its definition of a shutdown and its reporting period attached.

The federal and research data is still useful context, because it shows how small most firms are and how most of them end.

For a platform team, the practical reading is simple: a shutdown statistic tells you the climate, not which of your companies holds something worth keeping. That answer comes from looking at each company's size, history and systems.

What paths lead a startup to shut down

No single cause explains shutdowns, and no dataset in our sources ranks them. These are the paths platform teams commonly see, and each one shapes who controls the records at the end.

• Runway ends before the next round. A company that raised at a high valuation may face a down round that the board, the cap table or the founders will not accept, and a wind-down becomes cleaner than a recapitalization.
• A loan comes due or a covenant trips. When a venture debt facility matures or a covenant is breached, the lender can set the timetable; see how venture debt lenders approach a startup shutdown.
• The product is overtaken. A competitor or a platform feature makes the product less necessary, and revenue stalls even when the team is strong.
• An acqui-hire takes the people, not the company. The acquirer hires the team and sometimes takes the code, and the entity is left to wind down.
• The company stalls instead of closing. Some end up as zombie startups: alive on paper, shrinking, and still paying for systems nobody has decided about.

What happens to code, CRM and support history when a startup shuts down

Unless someone plans otherwise, records follow the subscriptions. When the company card is cancelled and accounts lapse, history becomes inaccessible and is eventually deleted under each vendor's retention terms. Code often leaves with an acquirer; operational history usually does not.

One rights point catches many startups: work written by contractors. Copyright in work employees create within their jobs generally belongs to the company, but a contractor's work may not unless it was assigned in writing, as the US Copyright Office's Circular 30 on works made for hire explains. Pull the contractor agreements before anyone describes the code base as the company's to license.

Where a records license fits next to an acqui-hire or asset sale

A license is not a replacement for a sale of the business or its code. It covers the operational history that buyers of the company usually leave behind.

Two cautions apply to every row. Deals are typically exclusive for AI training for an agreed term, so a license and any sale of code must be scoped together. And the company keeps ownership: the data is licensed, not sold, and nothing is binding until the company agrees price and terms and signs. Companies that are still operating, acquired or wound down can all qualify if the data still exists, which makes the export decision the one that matters most.

Which shutting-down portfolio companies are worth a screen

Most will not qualify, for reasons the page on why early-stage startups rarely qualify sets out. The ones worth a look are later-stage companies that grew large before they stalled. Use the PHASR screen, and stop at the first clear no.

• Peak: did the company reach 50+ full-time employees at peak (contractors excluded)?
• History: does it have several years of documented operations, spread across many systems (strong companies often run 10-15+)?
• Access: does someone still hold admin credentials, and are the key subscriptions still paid?
• Signer: who can authorize a license today: the CEO and board, an assignee or a trustee?
• Rights: did the company create the records, and do customer contracts and its privacy promises allow licensing?

The full baseline is on who qualifies. A company that is mostly consumer personal data, or whose archives are already gone, is not a fit however large it grew.

What it means for a VC platform team

Platform teams see shutdowns early: the runway slide in a board deck, a request for wind-down counsel, a founder asking for acquirer introductions. That is the moment to add one line to the portfolio wind-down playbook: export before you cancel.

1. Add record preservation to the shutdown checklist; the startup shutdown handoff checklist lists what to keep and who should hold it.
2. Run the PHASR screen, or the company fit checker, a preliminary and non-binding screen that needs no contact details.
3. Introduce the person who can sign: the CEO while the board is in place, otherwise the assignee or trustee.
4. SourceX qualifies the company on size, history, data breadth and rights, and the company completes a data inventory.
5. Price and terms are agreed with the company before any buyer reviews the opportunity.

The platform team never exports, uploads or describes confidential records. De-identification and redaction rules are agreed with the company before work begins, and data moves only after an executed agreement and the company's authorization.

A short note to a founder who is winding down:

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. The reward becomes payable only after the buyer pays and SourceX receives its fee, and no reward is guaranteed. It is a share of SourceX's fee, never a deduction from the company's proceeds. Check your fund's policies on fees connected to portfolio companies; the VC platform team playbook covers portfolio-wide screening.

Limits and open questions

• Shutdown counts are not comparable. Private datasets differ in coverage and definitions, so avoid quoting one provider's figure as the market total.
• AI demand is a forecast, not a promise. Epoch AI projects that, if trends continue, language models will fully use the stock of public human-generated text between 2026 and 2032, with wide uncertainty. That is why permissioned non-public records draw interest, but it does not mean any given dataset will find a buyer.
• Many closing startups fail the baseline. Small teams, short histories and consumer-heavy data rule out most of them.
• Proceeds belong to the company or its estate. How they are distributed follows the wind-down plan and applicable law, which counsel confirms.

This is general information, not legal, tax or financial advice. Confirm rights and authority questions with the company's counsel.

Next step

If a portfolio company is weeks from cancelling its tools, run the screen today and keep the exports. Then register as a partner to make the introduction, or have the CEO apply directly at sourcex.si/apply with your referral link.