Market facts quoted on this site
| Claim | Where it comes from |
|---|---|
| +25% from one competing bid | Spirit Airlines' 2026 bankruptcy auction of its business records. Google won the August 14 auction at $10M; Mercor was the $7.5M backup bidder; Micro1 later filed a $12.5M offer. $10M to $12.5M is a 25% increase. Reported by CSO Online, TechRadar and AI Weekly. The final court outcome was not confirmed in the sources we reviewed. |
| $7.5M–$12.5M, large enterprises | The range of bids in the same Spirit Airlines auction, as above. |
| Spirit's dataset: ~100M emails, ~500M Teams messages, ~30M lines of code | Court filings as reported by CSO Online and TechRadar (links above). Passenger profiles, loyalty data and personal information were excluded from the sale. |
| $10K–$100K, closing startups | Deal sizes reported for shutting-down startups' workspace and code archives sold through wind-down services, as covered by Fast Company, Gizmodo and MIT Technology Review. These figures are largely self-reported by the intermediaries involved. |
| $100K–$4M per license, operating companies with 20–500 staff | Bright Info's working range for operating companies, drawn from partnership-program ranges AI data buyers published in 2026 and from the disclosed deals above. We revise it as new deals are disclosed. It is a range, not a quote; your estimate depends on the factors below. |
| Who buys: AI data vendors, agent-training environment builders, data marketplaces, lab partnership programs | The four categories of buyer we invite. The named public bidders for Spirit Airlines' records (Google, Mercor, Micro1) span three of them. Background on the agent-training environment market: TechCrunch. |
What is illustrative
- The bid-round card on the homepage ("Bid round 2 of 2") is a worked example of what a results screen looks like. It is not a real deal and the buyers in it are categories, not companies.
- The timings on the four process steps are typical planning estimates, not commitments. Every engagement letter states its own schedule.
How we estimate a range
Before any data moves, we estimate from metadata only: which systems you run, how many years each covers, record and message volumes, and how much visible reasoning the record is likely to contain (decisions, escalations, handoffs rather than notifications). We then check rights status: employee notices, customer and vendor contracts, and software terms. The range we send back reflects those factors against the disclosed deals above. Years of history and visible reasoning move the number more than raw volume.
Corrections
If a figure on this site is out of date or you have a better source, write to hello@brightinfo.com. We update this page when a claim changes.