Auctions launching soon · waitlist open
AI data auctions for clean, verified training data.
We are building the marketplace where AI labs bid on training data that has been cleaned, deduplicated, PII-scrubbed and rights-checked before it goes on the block. An AI data auction is a competitive sale that sets the price and terms for data used to train or evaluate AI models.
Free to join. We will email you when the first lots open.
Radiology report corpus, cleaned
2.4M records · English · 12-month exclusive license
- Deduplicated98%
- PII scrubbed100%
- Rights verified100%
- Quality sampled94%
- Held-out slice reserved100%
Current bid
$198k
Sealed-bid · Vickrey
Closes in 02:14:36
- Bidder B$198k
- Bidder A$191k
- Bidder C$184k
Illustrative mockup. Not a live auction; figures are examples.
12
providers profiled →
13
data types defined →
11
sourced deals tracked →
35
terms in the glossary →
Planned process
How our AI data auctions will work
Six steps take a dataset from a seller's drive to a licensed delivery. Cleaning and verification come before the auction, so bidders compete on value, not on guesswork.
01Intake
Data owners submit a dataset with its rights documents and a description of how it was collected.
02Clean
We deduplicate, normalise, filter low-quality records and strip personal data and secrets.
03Verify
We check provenance and license terms, sample quality, and reserve a held-out slice.
04List
Each lot gets a data card, a test sample and a reserve price. Buyers see what they are bidding on.
05Auction
Qualified buyers bid in a sealed-bid, English or reverse format chosen for the lot.
06Deliver
The winner receives a license with stated uses, term and exclusivity, and the data through a secure channel.
The standard
What “cleaned” means for a lot
Four checks every lot passes before a buyer can bid. If a lot fails one, it does not list.
- Deduplicated and filtered
- Near-duplicate and low-quality records removed, so buyers pay for signal, not volume. What is training data?
- PII and secrets scrubbed
- Personal information and credentials removed or masked before any buyer sees a sample. What is PII?
- Rights and provenance checked
- Every lot carries a documented chain of rights and an explicit license scope. What is data provenance?
- Tested before you bid
- A representative sample plus a held-out slice lets buyers measure value before they commit. What is a held-out set?
Who the auctions are for
Sell side
Data owners
- Publishers, platforms and creators with unique, rights-cleared data.
- Teams with expert-annotated or domain-specific datasets.
- Anyone who wants competitive pricing instead of a single quote.
Buy side
AI labs and buyers
- Training and evaluation data with a documented rights chain.
- Samples and held-out slices to test value before bidding.
- One process for licensing, price discovery and delivery.
What is an AI data auction?
An AI data auction sells a lot, such as a dataset, a block of expert hours or a licensed archive, to the bidder who values it most, or buys it from the vendor who quotes the lowest price for a fixed specification. The winner receives a license that states permitted uses, term and exclusivity.
Real auctions are rare today because data is not used up when sold and buyers cannot judge its value before seeing it. Cleaning, samples and held-out slices exist to close that gap. Read the full explanation or look up terms in the glossary.
How AI data is priced and auctioned
- English auctionFits one-off, scarce assets where buyers can see each other, such as a unique archive of footage or a rare-language speech corpus.auction format →
- Dutch auctionFits sellers who want a fast sale and have many similar lots, or who expect more demand than supply at a high price.auction format →
- First-price sealed-bid auctionFits private deals where buyers do not want rivals to see their valuation, such as frontier labs bidding for data.auction format →
- Vickrey auctionFits designs where the seller wants truthful valuations, including data markets that must discover price for a one-of-a-kind dataset.auction format →
- Reverse auctionFits a buyer that knows what it needs and wants several vendors to compete, such as sourcing expert data from labs.auction format →
- Posted-price saleFits catalogue datasets that can be sold many times, such as cloud data exchange listings.pricing mechanism →
- Revenue-share licensingFits long-term content licenses where value is uncertain up front and both sides want to share upside.pricing mechanism →
For the economics, read how AI training data is priced.
Who sells AI data today
Providers fall into five groups: expert data labs, expert talent marketplaces, content licensing marketplaces, data marketplaces and cloud data exchanges.
| Provider | Type | Sells to |
|---|---|---|
| Scale AI | Expert data lab | Frontier labs, enterprises, government agencies |
| Surge AI | Expert data lab | Frontier AI labs |
| Mercor | Expert talent marketplace | Frontier labs and AI application companies |
| Turing | Expert data lab | Frontier labs, enterprises |
| Handshake AI | Expert talent marketplace | AI labs |
| micro1 | Expert talent marketplace | AI labs and AI-native companies |
See all AI data providers or start with provider comparisons.
Recent AI data deals
- 2025-10Mercor / Investors$350 million raised (company-announced)
- 2025-09Authors (Bartz class) / Anthropic$1.5 billion, roughly $3,000 per work; final court approval reported July 2026
- 2025-06Meta / Scale AIAbout $14.3 billion (reported)
- 2025-03Turing / Investors$111 million raised (company-announced)
- 2024-05Stack Overflow / OpenAIUndisclosed
Open the full deal tracker for sources and dates.
Guides
- 01Who buys AI training data?Frontier labs, neolabs, enterprises, startups and governments buy AI training data. What each buys, how they pay and where deals are reported.
- 02What is an AI data auction?An AI data auction is a competitive sale that sets the price of AI training data. How the formats work, why they are rare, and where they could fit.
- 03RLHF vs SFT: what is the difference?RLHF ranks model outputs; SFT imitates ideal answers. How the two post-training methods differ in data, cost and when labs use each.
- 04How to sell data to AI companiesSell data to AI companies by proving rights, packaging a sample, choosing a channel and pricing the license. A seven-step process with pitfalls to avoid.
- 05How to evaluate an AI data vendorEvaluate an AI data vendor with ten checks: provenance, quality control, reviewer vetting, security, pricing clarity, pilot results and exit terms.
- 06How to buy AI training dataBuy AI training data by defining the spec, choosing a provider type, running a pilot and enforcing acceptance tests. A step-by-step buyer's guide.
About this reference
No listed provider pays for placement. Learn who we are, how the methodology works, where figures come from on the sources page, or browse everything on the entity map. Last reviewed .
Frequently asked questions
- What is an AI data auction?
- An AI data auction is a competitive sale in which buyers bid, or sellers undercut, to set the price and terms for data used to train or evaluate AI models.
- Are you running auctions now?
- Not yet. We are building a marketplace where cleaned, rights-checked AI data goes up for auction. Join the waitlist and we will email you when the first lots open.
- What does cleaned data mean here?
- Data that has been deduplicated, filtered for quality, scrubbed of personal information and secrets, checked for rights and provenance, and sampled so buyers can test it before bidding.
- How much does AI training data cost?
- There is no standard price. Expert data is priced mainly by the expert's hourly rate, licensed content by negotiated fees, and catalogue datasets by subscription or license. See the deal tracker for reported figures.
- Who buys AI training data?
- Frontier labs, newer well-funded labs, enterprises, startups and government programs. Labs are the largest buyers.
- Is the reference content paid for by providers?
- No. No listed provider pays for placement or has reviewed its entry. Because we are also building an auction marketplace, we will label any commercial relationship with a listed company.
Be first on the block.
Join the waitlist to list a dataset or bid on cleaned, verified AI training data when the first auctions open.
Editorial note. No company listed here pays for placement or has reviewed its entry. AI Data Auction is building its own auction marketplace and will label any commercial relationship with a listed company. Figures are as reported by the named source on the stated date; check the primary source before relying on them. See our methodology.