# micro1 Says It Will Spend $1 Billion on Company Data. Its Price List Is the Floor

Source: https://sellyouraidata.com/blog/micro1-1-billion-company-data/
Updated: 2026-10-22

> On October 9, 2026 micro1, an AI training company, said it will spend $1 billion over 12 months to acquire and license de-identified company operational data, with capital from Citi and Hercules Capital. Its own page lists three tiers: more than $100,000, more than $500,000 and more than $1 million. Those tiers are the floor, not the price: they are one buyer's published numbers, and the page does not name buyers, exclusivity terms or liability terms. Taking the first tier means accepting a single bidder's view of what the data is worth.

## Key takeaways

- micro1 announced on October 9, 2026 that it will spend $1 billion over 12 months through its Company Data Partnerships program, financed by Citi and Hercules Capital.
- The published tiers are more than $100,000 for a qualified partnership, more than $500,000 for large datasets or work across several teams, and more than $1 million for highly unique proprietary data.
- The tiers are the floor, not the price. They are published by the only bidder on the page, and nothing in them shows what a second buyer would pay. Taking the first tier means giving up competition, and possibly exclusivity and scope.
- micro1 wants SOPs, knowledge bases, internal documents, CRM data, project histories and QA processes, plus records of team decisions and human feedback on AI outputs.
- micro1's own illustrative payout estimator shows $50,284 to $92,776, which is below its lowest published tier, according to one analysis. The tiers are floors for deals the buyer judges to qualify, not a quote.
- A $1 billion commitment is a ceiling on spending, not proof that companies are being paid that much. We found no public record of a closed micro1 purchase from an ordinary company.

## The short version

On Friday, October 9, 2026, micro1's founder and CEO, Ali Ansari, posted on the company's blog that micro1 will spend $1 billion over the next 12 months on "acquiring and licensing enterprise operational data" through its Company Data Partnerships program. The post says the money funds a new revenue stream for businesses from the operational knowledge they have built up, and that Citi and Hercules Capital are providing capital.[^1]

The Next Web, reporting the same day, says micro1 builds reinforcement learning environments from de-identified company data. In those environments AI models and agents practice workflows, decisions and complex tasks. micro1 says real operations involve "incomplete information, competing priorities and exceptions that require judgment," and the billion dollars will widen the industries and workflows it covers.[^2]

micro1 also publishes a price list. This page walks through it, what it does and does not say, and how it compares with other public numbers. We have no relationship with micro1, and nothing here is a quote for your data.

## The price list

micro1's data partnerships page lists three tiers of compensation.[^3]

**Table 1.** micro1's published compensation tiers, as of October 2026.

| Tier | What it covers | Published payment |
|---|---|---|
| Qualified enterprise data partnership | Operationally mature companies with high-quality documentation | More than $100,000 |
| Large-scale datasets | Large operational datasets, or ongoing participation across multiple teams and business functions | More than $500,000 |
| Highly unique proprietary data | Operational data with significant value for frontier AI development | More than $1 million |

The page adds that "compensation is based on the quality, uniqueness and value of the data provided," and lists dataset size, workflow complexity, data quality, domain expertise and uniqueness as factors.[^3]

Business Insider, relayed by AI Weekly on October 19, reports micro1 pays $100,000 to $2 million for operational data such as Slack logs, emails, CRM records and project histories, and that offers often arrive as six figure "exploding offers" that expire in 48 hours.[^4] That claim is secondhand and the "48 hours" detail is not on micro1's page, so treat it as reported.

### The tiers are the floor, not the price

The tiers use "more than." They are the lowest number at which a deal in each tier would start, set by the buyer, and not what an average company receives. They are also one buyer's published numbers, so they say nothing about what a second buyer would pay for the same records. One analysis of the program, by Pebblous, notes that micro1's own illustrative payout estimator shows a range of $50,284 to $92,776, which is below the $100,000 floor of the lowest tier, and that micro1 says the actual amount depends on sources, size, quality, access terms and due diligence results.[^5] We have not run the estimator ourselves, so that figure is as reported by Pebblous.

Put simply, a company that the buyer judges a strong fit might see six figures, and a company that does not pass the buyer's due diligence might see nothing. In both cases the buyer is the only one naming the price.

### What $1 billion implies

A little arithmetic helps keep the headline in proportion. If every deal sat at the $100,000 floor, $1 billion would cover up to 10,000 deals. If every deal were at the $1 million tier, it would cover about 1,000. Nobody has said how many deals micro1 expects, so this is just the shape of the number. And a commitment to spend is not evidence that the money has gone out: we found no public record of a closed micro1 data purchase from an ordinary operating company. Its late $12.5 million bid for Spirit Airlines' archive, reported by Troveo and others, is a different kind of deal, a bankruptcy auction, and the ombudsman has since backed Google's $10 million deal, according to The Next Web.[^6][^7] It is separate from the announced program.

## What micro1 says it wants

micro1's data partnership page lists the sort of material it is after: standard operating procedures, knowledge bases, CRM data, project histories, internal processes and quality assurance processes.[^3] The Next Web adds records of team decision-making and human feedback on AI outputs, plus internal documents.[^2]

That fits what the company does with the data. The goal is to teach an agent how a particular kind of business runs, including exceptions and judgment calls, not just to show it finished documents. MIT Technology Review reports the same pattern at another buyer, Surge AI, which wants varied archives that include disagreements and judgment calls.[^8]

For the data types we see sellers hold, see [customer service data](/data/customer-service/) and the other pages under [data types](/data/).

## What micro1 says about handling

Here is what the program page says about handling, retention and ownership.[^3]

- Personally identifiable information is removed "when relevant."
- Some datasets go through synthetic rewrites that preserve workflow structure while reducing direct ties to original records.
- Access is limited to a small number of authorized team members, and data is processed in isolated pipelines that are not commingled with other engagements.
- Employees who handle the data use company-managed devices and follow security and training requirements.
- Companies can review representative samples before prepared data is used.
- Anonymization and redaction requirements and the approved scope are defined before participation.
- "Companies retain ownership of their underlying data." Data is kept only for agreed periods and deleted on request or at the end of the engagement.

That is more detail than most buyers publish. It is still a marketing page, and several words ("when relevant," "some datasets," "reduce direct ties") leave room. The Pebblous analysis points out that synthetic rewriting keeps the structure of a workflow and mainly removes names, and cites a 2019 study that estimated 15 demographic attributes could single out 99.98% of Americans in a dataset.[^5] The study's authors describe that as a model-based estimate, and the body of the paper ties the figure to uniqueness in Massachusetts, so it is better read as a warning about weak de-identification than as a measurement of any specific program.[^9]

## What is not disclosed

Reading the page, the announcement and the independent coverage together, these points are missing:

- **The buyers.** None of the reporting names the labs or companies that receive micro1's data, as the Business Insider summary notes for micro1 and its peers.[^4]
- **Exclusivity and scope.** Pebblous notes the end buyer, exclusivity wording and liability or indemnity clauses are not published.[^5] Neither is the field of use, meaning which models may be trained and for how long.
- **Who checks permissions.** Participation requires "internal approval processes and permissions." Pebblous reads this as putting the burden on the seller, because micro1 does not itself require consent from customers or trading partners whose information appears in the records.[^5]
- **Payment timing.** The page text we read does not say when payment arrives. Mercor's page does, as covered below.
- **The referral reward.** The homepage banner offers "refer and earn up to $50k," tied to a referred company reaching revenue milestones, not to completing legal review.[^3][^5]

> **Watch out:** "You keep ownership" is not the same as "you have the right to sell." Confidentiality clauses in customer contracts, privacy promises to staff and customers, and license terms from your software vendors can all limit resale. Pebblous lists vendor no-redistribution terms and the end of the CCPA's business-to-business exemption on December 31, 2022 among the reasons.[^5] California's attorney general confirms those exemptions expired on that date.[^10]

## How the numbers compare

Different buyers publish different things.

**Table 2.** Public figures from buyers and intermediaries, as of October 22, 2026.

| Source | What it publishes | Amounts |
|---|---|---|
| micro1 | Three tiers plus a $1 billion, 12 month commitment | More than $100,000, $500,000 and $1 million[^3][^1] |
| Mercor | Named industries, tools, pay basis; no amounts | "Based on data volume, the tools you connect, and the depth of your data"[^11] |
| SimpleClosure (shut-down startups) | Typical payouts reported by Forbes | $10,000 to $100,000[^12] |
| Google at Spirit Airlines | Winning auction bid for a large archive | $10 million, pending court approval[^8] |

Mercor's program is the closest peer. Its page names eight industries (legal, healthcare, finance and insurance, professional services, manufacturing, retail and e-commerce, real estate, and logistics and supply chain) and says its platform extracts and de-identifies data, that no AI lab buyer or third-party vendor sees raw data, that it covers extraction, anonymization and transfer costs, and that payment is by wire after delivery, typically two to four weeks after signing.[^11] It publishes no payout amounts. Mercor also confirmed a March 2026 supply-chain attack that led to class actions, which is relevant if you are weighing who holds your data.[^13][^14] For a side by side, see [Mercor vs micro1](/compare/mercor-vs-micro1/).

The wind-down route is smaller. SimpleClosure's Forbes-reported range, $10,000 to $100,000, sits below micro1's lowest tier.[^12] The [SimpleClosure deal page](/deals/simpleclosure-shutdown-data-sales/) and [our piece on shutdown data sales](/blog/shutdown-data-sales/) cover that. And the [Spirit Airlines page](/deals/spirit-airlines-google/) covers the large end.

The Spirit case also shows micro1 under scrutiny. The court's consumer privacy ombudsman wanted to review micro1's plan to de-identify the data itself, not through an independent party.[^6] That does not say anything about the Company Data Partnerships program, but it is a reminder that who de-identifies data is a live question.

## What a seller gives up by taking the first tier

A published tier is an invitation to deal with one buyer. Here is what that costs a seller, using only what the sources show.

- **Competition.** The one large documented auction shows how far a contested price can move. Google opened at $5 million for Spirit Airlines' data and the sale closed at $10 million, with Mercor countering.[^15] Time reported a $7.5 million backup bid, and Time and CSO Online reported that micro1 itself made a late $12.5 million offer.[^16][^17] That was a bankruptcy sale, not a forecast for an operating company, but it shows that a first number and a contested number can be far apart.
- **Exclusivity.** micro1's page does not state whether a deal is exclusive.[^3] Two founders told Epoch AI that exclusive deals cost roughly four to five times as much as non-exclusive ones.[^18] A seller who grants exclusivity without a price for it gives away that premium, along with the option to license the same records to other buyers. A Malaysian law firm advises licensors to keep the right to license to third parties.[^19]
- **Scope and term.** The page does not state which models may be trained, whether fine-tuned derivatives are covered or how long the license lasts. Sidley notes that a trained model generally keeps working after a license expires, so the scope you agree to matters as much as the number.[^20]
- **Buyers and recipients.** The page does not say who receives the data, including whether a competitor of yours could.[^4][^5]
- **Terms that can change the number.** The page ties the final amount to due diligence, so the tier is not a promise of that figure.[^3][^5]

Before any data moves, settle the basics: list what you hold, check your customer contracts, NDAs, vendor licenses and privacy notices for limits on sharing, put the missing terms in writing, and have the de-identification tested by someone other than the buyer. SOPs and internal wikis are generally lower risk than chat logs, CRM exports and support tickets that contain personal details.

A single published tier gives you one reference point. More than one offer, on comparable terms, is what shows whether it is a fair price. Our [guide to selling direct versus through a broker](/guides/selling-direct-vs-broker/) covers the trade-offs, and if you are already talking to a buyer, **we review offers free** on our [offer review](/offer-review/) page. We are a broker with no deals or partnerships to report, so weigh our view accordingly. Our [assessment](/assessment/) is a way to sort your data by type and risk first.

## Bottom line

micro1 has put a number on the table, $1 billion, and a floor for each tier of deal. That is real information: it shows that operational data has buyers and that at least one is publishing a price list. It is not evidence that any particular company will be paid, or that the typical payment will be in the six or seven figure range, and it is not the price your data would reach with more than one buyer competing for it. The program is days old as of this writing, its buyers are unnamed, and the legal weight of any sale falls on the company selling.

We will update this page as micro1 or its partners publish more.

[^1]: [micro1 commits $1B to enterprise data acquisition](https://www.micro1.ai/blog/micro1-commits-1b-to-enterprise-data-acquisition) (micro1, October 9, 2026)
[^2]: [micro1 commits $1bn to buy company data for training AI agents](https://thenextweb.com/news/micro1-1bn-enterprise-data-ai-training-citi-hercules) (The Next Web, October 2026)
[^3]: [micro1 Data Partnerships](https://www.micro1.ai/data) (micro1, accessed October 2026)
[^4]: [Micro1 pays up to $2M for employees' Slack DMs and emails](https://aiweekly.co/alerts/micro1-pays-up-to-2m-for-employees-slack-dms-and-emails) (AI Weekly, October 2026, summarizing Business Insider and Fast Company)
[^5]: [micro1 Company Data Partnerships: rights analysis](https://blog.pebblous.ai/blog/micro1-company-data-partnerships-rights/en/) (Pebblous, October 2026)
[^6]: [Spirit's privacy ombudsman backs Google's $10m data deal](https://thenextweb.com/news/google-spirit-airlines-data-ombudsman-approval) (The Next Web, October 2026)
[^7]: [Google's $10M Spirit Airlines Data Deal, Explained](https://www.troveo.ai/resources/google-spirit-airlines-data) (Troveo, August 2026)
[^8]: [AI labs are buying data from dead startups to train their models](https://www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models) (MIT Technology Review, October 2026)
[^9]: [Estimating the success of re-identifications in incomplete datasets using generative models](https://pmc.ncbi.nlm.nih.gov/articles/PMC6650473) (Nature Communications, July 2019)
[^10]: [California Consumer Privacy Act (CCPA)](https://oag.ca.gov/privacy/ccpa) (California Attorney General, accessed October 2026)
[^11]: [Your enterprise data can advance frontier AI](https://www.mercor.com/data/) (Mercor, accessed October 2026)
[^12]: [Mercor chases shuttered startups' Slack and Jira archives](https://aiweekly.co/alerts/mercor-chases-shuttered-startups-slack-and-jira-archives) (AI Weekly, August 2026, summarizing reporting by The Information and Forbes)
[^13]: [Mercor says it was hit by cyberattack tied to compromise of open source LiteLLM project](https://techcrunch.com/2026/03/31/mercor-says-it-was-hit-by-cyberattack-tied-to-compromise-of-open-source-litellm-project/) (TechCrunch, March 2026)
[^14]: [AI industry recruiting platform faces multiple lawsuits over data breach](https://www.hrdive.com/news/ai-industry-recruiting-platform-faces-multiple-lawsuits-data-breach/817319/) (HR Dive, April 2026)
[^15]: [Google picked and paid for the firm anonymising the Spirit Airlines data](https://thenextweb.com/news/google-spirit-airlines-data-10m-bankruptcy-auction-mercor) (The Next Web, August 2026)
[^16]: [Google's $10 million bid for Spirit Airlines' data reveals AI's next frontier](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/) (Time, August 2026)
[^17]: [Bidding war for defunct Spirit Airlines' employee data will not die](https://www.csoonline.com/article/4218769/bidding-war-for-defunct-spirit-airlines-employee-data-will-not-die.html) (CSO Online, September 2026)
[^18]: [Gradient Updates: the state of RL environments](https://epoch.ai/gradient-updates/state-of-rl-envs) (Epoch AI, January 2026)
[^19]: [Licensing of data for AI model training: things to take note of](https://hhq.com.my/posts/licensing-of-data-for-ai-model-training-things-to-take-note-of/) (Halim Hong & Quek, February 2024)
[^20]: [Legal implications in AI development and deployment: training data and grounding data](https://www.sidley.com/en/insights/newsupdates/2026/07/legal-implications-in-ai-development-and-deployment-training-data-and-grounding-data) (Sidley Austin, July 2026)

## FAQ

**What did micro1 announce?**

On October 9, 2026, micro1 said it will spend $1 billion over the next 12 months acquiring and licensing enterprise operational data through its Company Data Partnerships program. Citi and Hercules Capital are providing the capital. The company uses de-identified data to build reinforcement learning environments that mirror real business operations.

**How much does micro1 pay?**

Its page lists three tiers: more than $100,000 for qualified enterprise data partnerships, more than $500,000 for large-scale datasets or ongoing participation across teams, and more than $1 million for highly unique proprietary data. Business Insider, as relayed by AI Weekly, reports offers of $100,000 to $2 million. The page says pay is based on quality, uniqueness and value.

**Is the first published tier what a company should expect to be paid?**

It is the starting point of one buyer's range, and the page says the actual amount depends on quality, uniqueness, due diligence and other factors the buyer judges. It is a floor for deals the buyer judges to be in that tier, not a quote, and it says nothing about what other buyers would pay. See [what a seller gives up by taking the first tier](#what-a-seller-gives-up-by-taking-the-first-tier).

**Do I keep ownership of my data?**

micro1 says companies retain ownership of their underlying data, that data is kept only for agreed periods, and that it is deleted on request or at the end of the engagement. Ownership is not the same as the right to sell it, so contracts with customers and staff still matter. See [the risk checklist](/blog/legally-radioactive-data-sale-risks/).

**Does micro1 remove personal information?**

It says it removes personally identifiable information when relevant, and that some datasets get synthetic rewrites that loosen direct ties to the original records. It also says companies can review representative samples before use. The wording is looser than a defined standard.

**Is micro1 the same as Mercor?**

They are similar but separate companies. Both supply expert human data to AI labs and both now run company data programs. Mercor publishes no payout amounts; micro1 publishes tiers. See [our Mercor vs micro1 comparison](/compare/mercor-vs-micro1/).

**What does a seller give up by taking the first published tier?**

Competition, mainly. A single buyer sets the number, and the page does not state exclusivity, usage scope or who receives the data. See [selling direct versus through a broker](/guides/selling-direct-vs-broker/), and if you already have an offer, [we review offers free](/offer-review/).

**Is SellYourAIData working with micro1?**

No. SellYourAIData has no relationship, deal or partnership with micro1 or any company named on this page. This is independent analysis of public sources.