MarTech Verdict

Reviews · AI data licensing

How Much Is Company Data Worth to AI in 2026? Price Evidence

Dated, sourced prices for company data sold to AI buyers, from a $25K startup code base to Spirit's $10M archive, and what moves value, checked October 21, 2026

By the MarTech Verdict Editorial Team · Updated October 25, 2026

How we put this together: We read the official pages of micro1, Scale AI, Mercor, Handshake AI, Labelbox Prism, Polyshares, Endless Data, Troveo, SimpleClosure AssetHub and AIData.Broker on October 21, 2026, and re-read micro1's pages on October 25, 2026 after it rewrote its data page, plus reporting from MIT Technology Review, Business Insider, The Next Web, iTnews, Upstarts Media and The Information, Epoch AI's January 12, 2026 interview-based report and a Pebblous blog analysis. We haven't sold data to any of these companies, requested an offer or spoken to them for this article. Figures are the companies' own unless attributed, and any arithmetic is labelled as ours.

Published evidence puts company data sold for AI between tens of thousands of dollars and low millions, with one $10 million exception for Spirit Airlines’ archive. Buyers advertise $10K to $4M+, but the few checkable deals are smaller: about $25,000 for one startup’s code base and $110,000 to $450,000 in Polyshares’ unnamed case studies, as of October 21, 2026.

Onfolio may earn a referral fee if you sign up to Polyshares through our links. This does not affect our editorial assessment.

This page collects every dated price point we could source, then explains which characteristics of your systems move the number. It’s written for the person who would run the export: an ops, RevOps or IT lead who needs to know what a buyer is paying for before deciding what to put in scope.

The price evidence

Every row is either a company’s own published figure or a named publication’s report. Advertised ranges are what buyers say they pay, not completed deals.

Data pointAmountSourceDate
Polyshares stated typical deal$100K to $2M+polyshares.comRead October 21, 2026
Polyshares published case studies (four, unnamed, perpetual exclusive)$110,000 (marketing agency) to $450,000 (roofing)polyshares.com/track-recordRead October 21, 2026
micro1 partnership tiers$100K-500K, $500K-2M, $2M+; payments “typically ranging from $150K to over $2M”micro1.ai/data-partnershipsRead October 25, 2026
micro1 payout estimator, illustrative (secondhand)$50,284 to $92,776Pebblous blogOctober 2026
Scale AI$10K to $1M+ “illustrative value per data partnership”scale.com/data-partnershipRead October 21, 2026
Handshake AI“$100K to $4M+”joinhandshake.comRead October 21, 2026
Labelbox PrismAbout $50K for targeted datasets to $1M+; upfront fee plus revenue shareprism.labelbox.com/faqRead October 21, 2026
Endless Data“Up to $4M” (no typical range)endless-data.comRead October 21, 2026
TroveoFull-company deals “start at six figures”troveo.ai/business-dataRead October 21, 2026
SimpleClosure AssetHub“Tens to hundreds of thousands of dollars”simpleclosure.comRead October 21, 2026
SimpleClosure test marketplace, prior year$1 million across nearly 100 dealsSimpleClosure to Forbes, via Upstarts MediaSeptember 3, 2026
Closed crypto startupAbout $25,000 for code, about $15,000 pending for workspace dataMIT Technology ReviewOctober 13, 2026
HVAC firm (14 staff), workflow partnershipAbout $150,000The Information (headline)2026
Spirit Airlines archive (bankrupt)$10M agreed with Google, court approval pendingiTnews; MIT Technology Review; The Next WebAugust 2026; October 13, 2026; October 9, 2026
Single code repository, closure marketplacesAbout $5,000Troveo’s value article, citing reportsRead October 21, 2026
Startup code and workspace archiveAbout $10,000 to $100,000Troveo’s value article, citing reportsRead October 21, 2026
Reddit data access (Google)About $60M a yearTroveo’s value article, citing reportsRead October 21, 2026
News Corp catalog (OpenAI)$250M+ over five yearsTroveo’s value article, citing reportsRead October 21, 2026
Medians compiled by AIData.Broker, a sell-side broker$31.6K for “the few public sales”; $333K across 103 public company-record prices that mix sales, offers and estimatesaidata.broker (the broker’s own compilation)Read October 21, 2026 (page updated October 20, 2026)

Our arithmetic from these figures: Polyshares’ four case studies total $1,135,000, an average of about $284,000 per license. SimpleClosure’s $1 million across nearly 100 deals averages about $10,000 per deal. The MIT Technology Review founder’s two amounts add up to about $40,000, of which about $15,000 was still pending.

The Reddit and News Corp rows are publisher and platform deals, not operating-company records. They show the upper end of the disclosed market, not what a 50-person firm’s Slack and CRM would fetch.

What labs pay vendors

Epoch AI’s January 12, 2026 report on reinforcement learning environments is interview-based, so treat these as reported ranges rather than audited prices. They describe what AI labs pay the companies that build training environments and tasks. micro1, for one, uses de-identified company data to build reinforcement learning environments (The Next Web, October 9, 2026).

ItemReported price (Epoch AI, interviews)
One RL training task$200 to $2,000
Website replica (“UI gym”)About $20,000 each
High-quality replica of a complex product like SlackAbout $300,000
Environment contracts$300,000 to $500,000 per quarter (one researcher’s view)
Exclusive vs non-exclusiveExclusive deals roughly 4 to 5 times the price

These are prices for finished training products, one step removed from your records. They don’t tell you what a vendor will pay you for the raw material.

What moves value in your systems

Buyers publish the factors they price on. Mapped to the systems an ops team runs, they look like this.

FactorWho says it mattersWhere it shows up in your stack
Connected context across systemsMercor: connecting more tools “generally results in a higher payout”; micro1’s CRO told Business Insider that without Slack, CRM or email records “the rest of the corpus is less valuable”The same project traceable across Slack or Teams, email, Jira or Asana, and the CRM
Intact conversation structureScale keeps “full conversation histories, threads, participants, and reply structure”; Polyshares keeps “roles, timestamps, and the work itself”Threaded Slack channels and email chains, not flattened exports
Decisions and workflow complexitymicro1 says the most valuable datasets capture “decision-making processes” and prices on “workflow complexity”Tickets with resolutions, CRM deal stages, approval workflows
Length of historyPolyshares: “the length and texture of the record”; Sunset values by years in operation (MIT Technology Review)Years of retained data in each system
Volume and team sizemicro1 and Mercor price on volume; Sunset on headcount and data volume; micro1’s $500K-2M tier covers multiple teamsNumber of active users and records
Uniqueness and domain depthmicro1’s $2M+ tier is for “highly unique, proprietary operational data”Specialist SOPs, niche workflows, regulated-industry processes
Clean rightsTroveo calls customer and product-usage data “frequently the least licensable”; Surge AI’s Nick Heiner told Business Insider a group tried to sell a codebase it had no rights toWhether records are yours or your clients’
ExclusivityEpoch: exclusive 4 to 5 times non-exclusive; Prism prices extra exclusivity to the buyerExclusive by default at Polyshares and Prism; non-exclusive at Endless

Why published ranges and estimates differ

Headline ranges and estimator outputs measure different things. A tier like micro1’s “$2M+” is a floor for partnerships that qualify, and “up to $4M” is a ceiling. The Pebblous analysis found micro1’s illustrative estimator figure, $50,284 to $92,776, sat below its own lowest tier.

Estimators are built on a few inputs. Handshake’s uses employees, years and country; Endless needs a work email before showing a range; Troveo labels its estimate “directional.” None of them has seen your data, and Polyshares says it doesn’t pay for a record it hasn’t seen.

Buyers also price on different theories. MIT Technology Review (October 13, 2026) reported that buyers quote “wildly different numbers” because valuations rest on “research theses” about how many useful tasks a dataset will produce. SimpleClosure’s October 1, 2026 blog warned about new buyers advertising $1M to $5M.

Payment structures

How you’re paid changes the real value as much as the headline number. The same figure can arrive at signing, after delivery, or only if the data keeps selling.

StructureWho uses it (published)What it means for timing
Fixed license fee after deliveryPolyshares (net 30 to 60 after approval and delivery); micro1 (30 days to approve, then paid within 30 days of invoicing)Export work first, payment one to two months later
Fixed sum agreed before sharingEndless DataNumber fixed early; payment date after signing not published
Payment on delivery to buyerMercor (wire, typically 2 to 4 weeks after signing)Tied to extraction and processing time
Upfront fee plus revenue shareLabelbox PrismSome cash at signing, more if the data keeps selling
Revenue share per saleTroveoDepends on sales happening; timing not published
Recurring partnershipScale AI (“scaling with cadence”)Ongoing deliveries, ongoing payments
Success percentage on a sale or licenseSimpleClosure AssetHub (percentage not published)Net figure known only at the offer

In wind-downs, MIT Technology Review notes that proceeds typically flow to senior creditors and investors, not employees. Prism says wind-down packages are possible, including inside bankruptcy proceedings.

The bottom line

The checkable figures we found sit between about $10,000 per deal (SimpleClosure’s average, by our arithmetic) and $450,000 per company, with Spirit Airlines’ $10M as the one large exception for a major airline’s archive. Advertised ranges run much higher, and most buyers don’t publish completed deals. Business Insider cites Gartner estimating that annual spending on data for AI will rise from under $1 billion last year to almost $6.5 billion in 2027.

There’s no published proof yet of how much real company data improves models. MIT Technology Review reported that no company has released tests comparing real-world data training with the alternatives, and that prices will likely stabilize in the coming year. Until then, a written offer on your own scoped data is the only number that applies to you. Our step-by-step guide to selling company data to AI covers how to scope the data and collect two or three offers.

What data you can and can’t license

Value only counts if you have the rights. Your own SOPs, templates, internal tickets, project history and de-identified internal chat are generally licensable; client deliverables, client ad accounts and ESP lists, client Slack Connect channels, contact-level CRM data, HR records and call recordings generally aren’t. CCPA/CPRA has covered employee and B2B contact data since January 1, 2023, GDPR covers EU and UK people, and Slack’s API terms (August 2025) bar bulk export and LLM training via the API, per Hunton. The buyer comparison covers each buyer’s published exclusions, and the Polyshares review and micro1 data partnership review go deeper on two programs.

Deciding whether to sell at all? onfolio.com looks at how much company data is worth to AI from the owner’s side.

Sources

Considering licensing your company’s data? Apply to Polyshares for an assessment.

FAQ

How much is my company data worth?
There's no public rate card. Buyers advertise ranges from $10K to $4M+, but the checkable deals we found on October 21, 2026 ran from about $25,000 for one closed startup's code base to $450,000 in Polyshares' largest published case study, with Spirit Airlines' $10M archive as the large exception. Value depends on history, connected systems, uniqueness, rights and exclusivity.
How much do AI companies pay for data?
Published figures vary widely. Scale AI shows an illustrative $10K to $1M+ per partnership, micro1 lists tiers from $100K-500K to $2M+, Labelbox Prism says about $50K to $1M+, and Handshake AI advertises $100K to $4M+. Reported deals include about $150,000 for an HVAC firm (The Information) and $10M for Spirit Airlines' archive. AI labs also pay vendors $200 to $2,000 per training task, per Epoch AI's interviews.
How much do companies make by selling data?
Reported examples are mostly in the tens to hundreds of thousands. SimpleClosure told Forbes its test version recovered $1 million across nearly 100 deals, about $10,000 per deal on average by our arithmetic. Polyshares' four published licenses ran $110,000 to $450,000. AIData.Broker puts the median of the few public company-data sales at $31.6K, or $333K when offers and estimates are counted.
Is there a calculator for how much my company data is worth?
Several buyers run estimators, but they're indicative. Troveo calls its estimate "directional," Endless Data needs a work email to show a range, and Handshake AI shows ranges by employees and years. A Pebblous analysis reported micro1's illustrative estimator figure of $50,284 to $92,776, below micro1's own lowest $100K tier.