Real prices, plainly explained
How Much Do AI Companies Pay for Data
How much do AI companies pay for data? For one established company's old work records, the per-company payouts buyers list publicly start near $10K and climb past $4M. Forbes found roughly 100 startups that had shut down and were each paid between $10K and $100K for their leftover Slack, email and Jira history.
The top of the market goes far higher. Google won an auction for the internal messages of Spirit Airlines with a $10M bid. Your own number depends mostly on headcount, years of records, the systems you used and where your team works. A free estimate turns those four facts into a range for your company.
- Only public price points
- Ranges, never promises
- $0 cost to owners
The short answer
Public prices paid for business and content data
Every number below was reported in public. DillisData took no part in any of these deals. Together they show the low end, the middle and the very top of the market.
| Who was paid | What the buyer got | Reported price |
|---|---|---|
| About 100 shut-down startups | Old Slack, email and Jira archives | $10K to $100K each (Forbes) |
| Cielo24, a software company | Thirteen years of Slack, email and Jira history | Hundreds of thousands of dollars (Gizmodo) |
| Established companies, per buyers' own price lists | A license to one company's old work records | About $10K up to more than $4M per company |
| Spirit Airlines, in bankruptcy | Roughly 600 million messages: 100 million emails, the rest from Teams | $10M winning bid from Google, with rival bids of $7.5M and $12.5M (Axios) |
| Shutterstock | Licenses to its image and media library | $104M from AI developers in 2023 (PetaPixel via Bloomberg) |
| Licenses to its platform data for AI use | $203M across its AI contracts (SEC Form S-1) |
Reading the range
Why one archive brings $10K and another brings $10M
The gap comes down to volume and depth. Spirit was a large airline with decades of history, and its archive held hundreds of millions of messages. Many of the startups in the Forbes story were small teams that lasted a few years. A buyer pays for how much real work the records show, so a bigger and older archive earns more.
Reddit and Shutterstock belong in a separate group. They are media platforms that license huge libraries to many AI developers. Their totals show how much money is flowing into training data overall. They are not a fair guide to what one private business will be offered.
Most established businesses land between those two ends. A company with a few dozen people and ten years of email, files and tickets holds far more than a two-year startup, yet far less than an airline. Our overview of licensing company records to AI companies explains which records count and how these deals are set up.
One more thing shapes the answer. In these deals a business licenses its records. It grants permission to use a copy and keeps ownership of the originals. That is why the same archive can sometimes earn money from more than one buyer, and why the terms of each deal matter as much as the headline price.
Price drivers
Six things that raise or lower an offer
Buyers ask the same handful of questions about every company. Here is what each answer does to the price.
Headcount
Each person who wrote email, filed tickets or kept notes added to the archive. Published buyer formulas add money for every employee, so a 150-person firm is priced well above a 25-person one.
Years of history
A long record shows how the work changed through good years and bad ones. Buyers add value for each year, and most want at least 3 years before they make an offer.
Types of records
Written back-and-forth between people is the core of most deals. Email, team chat and shared documents count the most. A company with only one system, or with no documents at all, gets a much lower price.
Exclusive or non-exclusive
An exclusive license means no other buyer can get these records from you, so it should pay more up front. A non-exclusive license leaves you free to license again later. We point out these terms in every offer you see.
Language and location
English-language writing from US-based teams draws the highest prices. Companies in Canada, the UK and Europe still qualify, and our estimate counts them at about three quarters of the US figure.
How many buyers compete
When several buyers want the same archive, the price climbs. In the Spirit auction, bids ran from $7.5M to $12.5M for one set of records. We ask several buyers for offers, so your price is never set by a single bidder.
Record types
Which records carry the most value, and where they live
Buyers want records that show people solving real problems in writing or out loud. Here is how the common business systems compare.
| Record type | Where it usually lives | Why buyers pay for it |
|---|---|---|
| Microsoft 365, Outlook, Gmail, PST backups | Years of real requests, answers and follow-ups between people | |
| Team chat | Slack, Microsoft Teams, Google Chat | Fast, informal problem solving that public text rarely shows |
| Documents and contracts | SharePoint, Google Drive, Dropbox, Box | Long, careful writing such as proposals, policies and plans |
| Support tickets | Zendesk, Freshdesk, Jira Service Management | A clear question, the fixes tried and the answer that worked |
| Sales and CRM notes | Salesforce, HubSpot, Zoho CRM | How a lead becomes a customer, one note at a time |
| Projects and work orders | Jira, Asana, ERP and field service tools | How tasks get planned, handed off and closed out |
| Call recordings | RingCentral, Zoom, Dialpad | Spoken conversations, which are scarce in public data |
| Source code | GitHub, GitLab, Bitbucket | Working code along with the reviews and comments around it |
Get a price range built for your company
You tell us your team size, your company's age, which software holds your records and your country or region. The whole thing takes about two minutes, and you share no files and no logins.
What stays out
Records that add risk without adding much to the price
Holding back the wrong records protects you and rarely moves the offer much. Every exclusion you pick is written into the contract.
Personal and HR files
Payroll, medical notes, ID scans and performance reviews carry risk and teach a model very little. Exclude them on day one. The buyer's own scrubbing of names and personal details helps, yet no scrub catches every last detail.
Regulated records
Patient charts, privileged legal files and similar records come with legal duties. They generally stay out unless your lawyer says a version of them can go in.
Content made by someone else
Purchased reports, vendor manuals and stock media belong to the people who made them. Only records your company created or owns can be licensed.
Material covered by an NDA
Customer and partner agreements may forbid sharing certain information. Read them, or exclude the related folders and accounts.
Anything on legal hold
If a lawsuit or audit requires you to preserve records, ask your lawyer before any copy goes out.
From estimate to payment
How a price range becomes money in the bank
Your estimate is the starting point. Real numbers come from buyers, and the owner spends about an hour on the whole process.
See your range
Four questions about size, age, systems and location. There is nothing to upload and no account to connect.
2 minutesWe gather bids
We pitch a one-page summary of your business to AI buyers hunting for your kind of records, then round up their bids. Your records stay where they are.
1 to 2 weeksCompare and sign
On one call we walk through each price, the exclusivity terms and your list of exclusions. You pick the buyer and sign with them directly.
One callReceive payment
Only the records you signed off on are copied. The buyer scrubs personal details, then sends the money on its own schedule, which buyers publish as 7 to 60 days.
7 to 60 daysWhere the money comes from
Why buyers have this much to spend
These public figures show the size of the market behind every offer. DillisData had no role in any of them.
What Amazon, Microsoft, Alphabet and Meta together plan to put into AI during 2026. Training data is one of the things that budget pays for.
Financial Times via AI Weekly, Jul 2026Meta's price for just under half (49%) of a firm that supplies training data to AI labs.
TechCrunch, Jun 2025What Shutterstock earned in a single year, 2023, by licensing its library to AI developers.
PetaPixel via Bloomberg, Jun 2024Reddit's IPO filing listed AI data deals worth this much in all.
SEC Form S-1, Feb 2024Timing
Why current prices may not last
Nobody can say how many more years prices will hold near these figures. Here is why waiting has a cost.
Public text is running short
A 2024 forecast by Epoch AI found that AI could exhaust the stock of public text written by people at some point from 2026 to 2032. OpenAI co-founder Ilya Sutskever went further. Speaking to AI researchers in late 2024, he described the field as having hit "peak data".
Early sellers get the best prices
A buyer's first archives from a given field fill a real gap, so they command the strongest prices. Later archives of the same kind add less, and offers shrink to match.
Labs will need less over time
As labs fill their gaps and find other ways to train, demand for business records should fall. Nobody can say exactly when that will happen.
Idle archives pay nothing
Old mailboxes and drives cost money to store and earn nothing while they sit. A license puts them to work without deleting anything.
Our monthly spots are limited
Each month we accept only so many new companies, which keeps every deal properly staffed. When a month fills up, the next company starts the month after.
Questions
Questions owners ask
Is there a standard price per email or per gigabyte?
No single rate exists. Some published formulas add a small amount for data volume, but most of the price comes from headcount and years of history. Two archives of the same size can draw very different offers.
Does the estimate guarantee that price?
No. It is a careful range built from public buyer pricing formulas and rounded down, not an offer. Real offers come from buyers after they see your company profile, and they can land above or below it. You can read how our company data estimate is calculated, step by step.
Can a small company still get paid?
Buyers usually look for at least 20 people on the payroll and three or more years of steady operation. Below that, offers are less common but not ruled out. Companies that have closed or are closing can qualify too.
Does DillisData take a cut of the payment?
No. Our fee is billed to the buyer as a separate charge, apart from your payment, so your share is never reduced. When no deal closes, no one owes us a cent.
Can I be paid more than once for the same records?
Yes, if your first deal is non-exclusive. You can license the same archive to another buyer later. An exclusive deal closes that door, so its price should make up for what you give up.
Will leaving some records out cut my price?
Leaving out a whole system, such as all of your email, lowers the price a lot. Leaving out a few sensitive folders or customer accounts usually matters far less than headcount and years of history.
How soon does the money arrive?
Payment times that buyers publish run from 7 to 60 days after the records are delivered and checked. We confirm the timing for each offer before you sign.
Do buyers pay less for companies outside the US?
Usually, yes. US teams writing in English get the top prices. Companies in Canada, the UK, Europe and elsewhere can still qualify at lower prices.
Your number
Find out where your records fall in the range
Answer four quick questions and see a cautious price range for your company on the spot. Nothing to upload, nothing to sign and nothing owed.
Free. Two minutes. A limited number of companies each month.