Dark data is the information a company collects and stores while doing business but never uses again, such as old email, chat history, shared drives, closed tickets and finished job records. Since 2025, AI companies have paid for de-identified copies of these archives: one buyer's published case studies for operating companies run from $110,000 to $450,000 (Polyshares, checked October 2026). Shut-down startups have typically received $10,000 to $100,000 (Forbes via Fast Company, April 2026).
On this page
Most companies have plenty of it. A 2019 Splunk survey found that on average 55% of an organization's data is dark (Splunk, 2019).
What counts as dark data in a small company?
Gartner describes dark data as information assets an organization "collects, processes and stores in the course of its regular business activity, but generally fails to use for other purposes" (Gartner, quoted by TechTarget, 2021). For a 20 to 500 person company, it is mostly the everyday paperwork of running the business, kept because deleting it feels risky:
| Where it lives | What is in it | Why it sits unused |
|---|---|---|
| Email (Gmail, Outlook) | Quotes, approvals, customer back-and-forth, vendor threads | Nobody searches mail older than a year or two |
| Chat (Slack, Teams) | Daily coordination, decisions, troubleshooting | Old channels are archived and forgotten |
| Shared drives (Google Drive, SharePoint, Dropbox) | SOPs, proposals, spreadsheets, drawings | Finished files are rarely reopened |
| Tickets and projects (Jira, Zendesk, Asana) | Requests, steps taken, fixes, outcomes | Closed tickets are kept only for reference |
| Industry systems (ServiceTitan, Procore, an ERP) | Jobs, dispatch notes, change orders, quality checks | Used for the current job, then left |
| Code (GitHub, GitLab) | Commits, reviews, issues | Old projects and retired products |
| Call recordings and meeting notes | How problems were explained and solved | Recorded once, never replayed |
How much of a company's data is dark?
Estimates vary because "dark" is hard to measure. Splunk's survey of more than 1,300 business managers and IT leaders worldwide found that 55% of data was dark on average, that 60% of respondents said more than half their organization's data was dark, and that one third said more than 75% was (Splunk survey, reported by Datacentre Solutions, May 2019). It is a vendor-sponsored survey from 2019, so treat it as a rough guide. Most owners know the pattern without a survey: years of files nobody opens.
Why do AI labs now pay for dark data?
Three changes in 2025 and 2026 gave these archives buyers.
Public text is finite. Fortune, citing a 2025 Epoch AI report, wrote that the supply of public human-written text could be fully used before 2030 under current trends (Fortune, September 2026).
AI agents need records of real work. Labs are building agents that take actions, not just write text. An agent has to learn how to choose steps, use tools, respond to results and recover from mistakes, and AI companies call the record of those steps a trajectory (Fortune, September 2026). An old support ticket linked to the chat thread that diagnosed it and the fix that closed it is one such record. Labelbox Prism lists human decisions and escalation paths, expert reviews and corrections, and edge cases among the most valuable data (Labelbox, checked October 2026).
Budgets grew fast. Gartner estimates annual spending on data for AI will grow from less than $1 billion in 2025 to almost $6.5 billion in 2027 (Business Insider, October 2026). Brendan Mahony, who runs the shutdown firm Sunset, compared these records to wood shavings that everyone thought had no value, the raw material of particle board (MIT Technology Review, October 2026).
What have companies been paid for dark data?
| Company | Records licensed or sold | Amount | Kind of number and source |
|---|---|---|---|
| cielo24, a transcription company | 13 years of Slack, email, Jira tickets and Drive files | "Hundreds of thousands of dollars" | Seller's account, Fast Company, April 2026, and Fortune, September 2026 |
| About 100 shut-down startups | Slack, email, documents, code | $10,000 to $100,000 each | SimpleClosure to Forbes, via Fast Company, April 2026 |
| HVAC companies | Operating records | $150,000 | Press report (headline only), The Information, July 2026 |
| Global marketing agency | Full operating record | $110,000 | Buyer case study, Polyshares, checked October 2026 |
| Roofing company | Full operating record | $450,000 | Buyer case study, Polyshares, checked October 2026 |
| Spirit Airlines, in bankruptcy | 100M emails across 80,000 accounts, 500M Teams messages, 17M OneDrive files | Google's $10 million winning auction bid; sale not final | Axios and The Next Web, August 2026 |
Spirit is the largest example, and it is still open. Google won the August 2026 auction with a $10 million bid, and Mercor was the next-highest bidder at $7.5 million (Axios, August 2026). Mercor had also offered $7 million if it could take the raw data and de-identify it itself (The Next Web, August 2026). In September, micro1 offered $12.5 million (Fortune, September 2026), and the court hearing on the sale was moved from October 14 to November 6, 2026 (TravelPulse, October 2026). As of October 29, 2026, no sale has been approved.
All of these are in our deal ledger with links. For what drives the price, see how much company data is worth.
Which dark data is worth the most?
Not the biggest pile. Buyers pay for connected records with visible decisions and outcomes, over years. SimpleClosure found that among workspaces it evaluated, 95% used Slack, 87% Google Drive and 82% GitHub, and that buyers increasingly want to follow one project across all of them, from idea to decision to final output (SimpleClosure, September 2026). A Surge AI lead told MIT Technology Review (October 2026) that one company's Slack history, if it has variety and depth, can be worth more than another's Slack, email and calendars combined.
SimpleClosure also reports demand for fields where operating data is scarce, such as manufacturing, energy, professional services and logistics (SimpleClosure, September 2026). Our guides cover email archives and Microsoft 365 data in more detail.
Is dark data a risk as well as an asset?
Yes. Old archives hold personal details about employees and customers, client material under contract, and sometimes passwords. Storing it has a cost, and a breach exposes all of it. Before licensing any of it:
- Check you own it and no client contract forbids its use.
- Leave out HR, payroll, health and privileged legal records.
- Make sure names, contacts and IDs are replaced before anything leaves. See how de-identification works.
Nyne notes that being able to export a record does not by itself give you permission to license it (Nyne, checked October 2026). This page is general information, not legal advice.
What should I do with my company's dark data?
- Do not delete it on impulse. Follow your retention policy, any legal holds, and any privacy rules that require you to delete certain records. Keep paid plans active where you need history: Slack deletes messages and files older than one year on free workspaces (Slack, 2024).
- Take inventory. List each system, its years of history and how many people used it.
- Decide what stays out. Mark client files, HR and anything sensitive.
- Find out what it is worth. Our estimate asks about your team, tools and years of records. Data Licensors, the company that publishes this page, also licenses de-identified copies of company records, and other buyers are compared in our guide to selling company data to AI.
If your company is closing, act before wiping servers. Our guide to selling startup data before shutdown covers timing.
