Insights · 6 min read

Is My Startup’s Data Worth Anything to AI Labs? A Frank Guide

Sometimes yes, often not much, and occasionally a surprise. Here is how to tell which you are.

Bearish AI · October 21, 2026

Founders ask this question in two moods: hopeful, because the company has years of work stored, or resigned, because the company is winding down and the data is the last asset. In both cases the useful answer is a clear picture of what the public record shows and what it does not.

What has been reported

• Shut-down startups. Nearly 100 deals in a year at $10,000 to $100,000 per company, according to Forbes as relayed by Gizmodo.
• One founder. About $25,000 for a codebase, with workspace data pending at around $15,000, per MIT Technology Review.
• Use. Gizmodo reports the data trains agentic AI models in simulated workplace environments, sometimes called reinforcement learning gyms.

Treat them as reference points, not promises. They come from reports about sellers who chose to come forward, often during a wind-down. We did not find a published distribution showing how many startups try to sell and get nothing.

Three honest tiers

These tiers are our rough framing from the reported data points, not a market standard. Where your company falls depends on facts only you know.

What buyers look at

MIT Technology Review quotes a wind-down firm’s chief executive saying valuation rests on headcount, years in operation, data volume and how the team used its tools. A buyer in the same piece values data by the tasks it can yield. A young startup with a tiny team scores low on almost every one of these, which is why the honest answer for many is “not much yet.”

Why the answer changes with time

A startup’s archive is a growing asset. Each year adds history, each hire adds contributors, and each new tool adds another kind of record that can be linked to the rest. A company that would be told “not much” at two years may be in a different position at five, as long as it kept the records and kept the rights clean. That is a reason to avoid selling in a hurry, unless a wind-down forces the timing.

Conversely, buyer interest is not guaranteed to last. MIT Technology Review says prices will likely stabilize over the coming year unless an AI bubble pops first, so waiting has a risk too. The sensible approach is to know your number now, so you can decide on facts instead of fear or hope.

What can make it worth nothing

• No rights. If you cannot show that the company owns or can license the data, buyers typically want a warranty that you do. Without it, there is no sale.
• Customer restrictions. Customer contracts may forbid reuse of data derived from their activity.
• Personal data you cannot remove. Even emails and calendars can contain private details such as medical appointments, an attorney told MIT Technology Review, and suppliers there leave out HR records.
• Nothing linked. Isolated documents with no code or discussion around them yield few tasks.
• Nobody can export it. If it sits in tools whose terms or formats block a clean export, the preparation cost may exceed the price.

A five-step check before you sell

• Inventory. List each source (code, reviews, tickets, chat, design files, recordings, support conversations, business documents), the years it covers and how many people produced it.
• Rights review. Ask a lawyer who owns what, which contractor agreements apply and what customer contracts say. This is the step most likely to change the answer.
• Privacy pass. Identify personal data, secrets and customer details, and decide what must be removed. Pseudonymizing people is not the same as clearing rights.
• Independent range. Get a conservative estimate before you talk to a buyer. The free one at /estimate asks for the inputs above and does not ask for any of your actual data.
• Read the license. Scope, term, exclusivity and sublicensing matter as much as the price. See /insights/exclusive-vs-non-exclusive-data-licensing.

Habits that raise the value of what you will have later

If your data is worth little today, you can still change what it is worth in a year or two. Keep the full history rather than squashing it. Keep discussion and tickets in systems you can export. Put clear assignment terms in contractor agreements, so ownership is not in doubt. Avoid pasting secrets into chat and code. Write down decisions where the team will find them. None of this is done for a buyer. It is ordinary good practice, and it happens to be exactly what makes a body of work easier to use as training material.

If the answer is “not much”

That is a real outcome. You can keep the archive and wait until it has more history and links, bundle it with other material, or use it internally. A small archive that is sold cheaply, with broad rights, can cost more in lost options than it earns. For how prices build up in general, see /insights/how-much-is-my-company-data-worth-to-ai-companies.

This article is general information, not legal or financial advice. Before sharing any company data, speak to a lawyer.