DataPacts

Data Warehouses (Snowflake, BigQuery, Databricks) as AI Training Data Sources

Checked

At a glance

A company can license what sits in its Snowflake, BigQuery or Databricks warehouse only as far as the data is its own to license, a buyer program or channel takes it, and the people named in it can be cleared.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

Six programs name a warehouse product on their own pages (checked November 9, 2026), and the three vendors document sharing data without copying it (documentation read November 17, 2026). A share is access, not a license, and the wider context is AI training data licensing, where not licensing is a valid result.

This page is written by the DataPacts Editorial Team, and every term is the program's or the vendor's own statement with its date; see how we source, date and check every figure.

What a Warehouse Offers an AI Data Seller

A cloud data warehouse is where a company's tables from its other systems already sit together, which is why programs name it as one source among many. Centific states that its Enterprise Data program covers 188 supported applications and lists "BigQuery, Snowflake, Databricks" among the data wanted (checked November 9, 2026).

Mercor's SaaS & Subscription listing names the same three products among ten tools, alongside dbt, Airflow, Fivetran, Looker, Salesforce, NetSuite and Stripe (Mercor states, checked November 9, 2026).

Two routes lead out of a warehouse. One is a records license to a program that connects or receives an export, the subject of this page; the other is a listing on a vendor's marketplace.

The warehouse is a store, not a source of rights, and the export rights of each source system sit on separate pages.

Whose Data Sits in a Warehouse: Own, Customer and Third-Party Licensed

Warehouse tables fall into three groups with different answers: the company's own records, data it holds for customers, and data it subscribed to from other providers.

  • Own records: tables the company produced in its operations and holds the rights to license.
  • Held for customers: tables that contain customers' content or data processed on their behalf, where the customer contract decides.
  • Subscribed from third parties: datasets obtained through a marketplace or a data provider, where the provider's license decides.

Mercor states that its SaaS & Subscription listing requires "the company's own production databases and a cloud warehouse" and names "agencies holding their clients' databases" as not a fit (checked November 9, 2026).

A Snowflake Marketplace listing carries the provider's own listing terms under the Snowflake Provider and Consumer Terms, and a Databricks Marketplace provider profile must link the provider's own terms of service (both checked November 9, 2026).

A dataset subscribed to carries the provider's license, and whether that license allows onward licensing or training is read from the listing terms, not assumed.

A derived table, such as a join of own and licensed data, takes the strictest of its inputs, which is why a rights audit for data you want to license covers each warehouse table. No vendor term on a customer's own data used for AI training is on record, so none is stated here.

Programs That Name Snowflake, BigQuery or Databricks

Six programs name a warehouse product on their own pages, each listed here with its role, the products named and the date its page was checked.

The table compares the six programs as of 2026-11-09, in alphabetical order, with no payout column; roles are as each company describes itself, and no program on the list is ranked or recommended.

Program Role Warehouse products it names (own words) One published term relevant to warehouse data Checked Source
Avelence seller advisory adviser "GitHub/Snowflake/GitLab" among eight record groups de-identification on a separate copy; no default system access 2026-11-09 avelence.com
Centific Enterprise Data data vendor "BigQuery, Snowflake, Databricks" read-only connection to in-scope systems; exclusions enforced in extraction code 2026-11-09 centific.com/enterprise-data
Mercor SaaS & Subscription listing human-data platform tools named: Snowflake, BigQuery, Databricks requires "the company's own production databases and a cloud warehouse"; agencies holding clients' databases not a fit 2026-11-09 api.data.mercor.com listings /saas-data
Scale Data Partnerships human-data platform "data warehouses (BigQuery, Snowflake, Databricks)" anonymizes inside the company's environment before extraction 2026-11-09 scale.com/data-partnership
Troveo business data marketplace Snowflake and Databricks among logos labeled on the page seller retains ownership; Troveo licenses on the seller's behalf 2026-11-09 (page re-read 2026-11-16) troveo.ai/business-data
Turing Project Lazarus human-data platform "databases (Postgres, Snowflake)" license to Turing exclusive for the first 6 months; Turing may license the record to more than one lab in that term 2026-11-17 lazarus.turing.com

Terms differ: Scale anonymizes inside the company's environment, Centific connects read-only, and Turing states that the license to Turing is exclusive for the first 6 months after signing, during which it may license the record to more than one lab (checked November 17, 2026).

Labelbox's FAQ also lists data warehouses among sources, but it is not a row because its program record does not name a warehouse product. Payouts and the other terms sit in AI data partnership payouts and program terms compared, each dated.

Sharing Without a Copy: What Each Vendor Documents

Snowflake, Google and Databricks each document a way to give another party access to warehouse data without copying it, and none of the three pages treats AI training as a use.

The table compares the three vendors' own documentation as of 2026-11-17, in alphabetical order, quoted verbatim from the pages read that day.

Vendor and mechanism What the documentation states Marketplace link stated Source
Databricks: OpenSharing "an open protocol developed by Databricks for secure data sharing with other organizations"; providers can share "with any user, whether or not they have access to Databricks" the basis for Databricks Marketplace docs.databricks.com delta-sharing page
Google BigQuery: sharing (formerly Analytics Hub) "securely share, discover, and access data across organizational boundaries without replicating data"; a linked dataset is "a read-only BigQuery dataset that serves as a pointer or reference to a shared dataset" publishers can "monetize listings on Google Cloud Marketplace or through your own channels" docs.cloud.google.com bigquery analytics-hub-introduction
Snowflake: Secure Data Sharing "no actual data is copied or transferred between accounts"; the sharing account is the provider, the recipient the consumer, and shared objects are read-only a listing can "offer data publicly on the Snowflake Marketplace" docs.snowflake.com user-guide data-sharing-intro

A share is access, not a license. The Databricks documentation uses the vendor's own term OpenSharing and does not use the name Delta Sharing on that page, while the Marketplace product page mentions Delta Sharing only in a customer example (checked November 16, 2026), so this page writes OpenSharing.

A read-only share gives access, and whether the other party may copy results out, train on them or must delete them is set by the license, not by the share. Formats, transfer and acceptance are covered under delivering licensed data to an AI buyer.

Collaboration Environments Inside Warehouses and What They Prove

Both Databricks and Snowflake document environments where parties work on combined data without seeing each other's raw tables, which suits proving quality but is not a license.

Databricks states that "Collaborators cannot see the data in other collaborators' tables, views, or volumes", that notebooks need approval by all collaborators before running, and that each clean room is limited to ten collaborators (documentation read November 17, 2026).

Snowflake states that a data offering's specification "controls which columns are exposed and what policies apply", that an analysis runner runs permitted templates on permitted data offerings, and that results can be returned or activated to another collaborator (read November 17, 2026).

BigQuery's sharing page mentions clean rooms only as a limitation, so no BigQuery clean-room claim is made here, and this page does not state that any AI data program uses such an environment. For the role of these environments in showing a buyer what the data is, see samples, clean rooms and proving data quality.

Marketplace Listings Versus a Records License

A marketplace listing offers a dataset to any eligible consumer on the provider's own terms, while a records license gives one program defined rights to a company's internal tables.

Snowflake states that listings run on the provider's own terms, that Stripe pays providers after receiving payment from consumers, and that paid listings need a billing address in one of 20 listed countries; no Snowflake fee is stated in its provider documentation (checked November 9, 2026).

Databricks requires a Premium plan or above and a Unity Catalog workspace, a provider profile that links the provider's own terms, and states no payout terms (checked November 9, 2026).

Fees, payouts and eligibility are covered on the profile pages and in the comparison of AI training data marketplaces.

Personal Data and Customer Data in Warehouse Tables

Warehouse tables often join personal data from several systems, so a warehouse export is personal data before it is training data. Methods for PII removal and de-identification before licensing apply to the export, and the tables most likely to need them are these:

  • customer and contact tables
  • event and usage logs with user identifiers
  • support and sales tables with free text
  • finance and order tables with named parties

Scale states that it anonymizes inside the company's environment, with approval gates for each package, and Centific states that exclusions are enforced in extraction code (both checked November 9, 2026). No vendor term on stored customer data and AI training is on record, so none is stated.

Law differs by jurisdiction, and the comparison sits under privacy law and AI training data by jurisdiction. This is information, not legal advice.

Benefits, Risks, Downsides and Alternatives of Licensing Warehouse Data

Licensing warehouse data can earn license income from tables the company already maintains, and it can place customers' and contacts' data in a model that cannot be untrained.

Benefits

  • Tables the company already maintains in one place can match a program's named tools, as Mercor's listing and Centific's page show.
  • A read-only connection, as Centific states, leaves the source system unchanged.

Risks

  • Tables can carry third-party license terms that bar onward licensing.
  • Joined tables carry personal data across systems.
  • No vendor term on stored customer data and AI training is on record.

Downsides

  • Turing states that the license to Turing is exclusive for the first 6 months (checked November 17, 2026).
  • A program may ask for exclusivity, which limits other licenses.

Alternatives

  • Not licensing at all, which is a valid result; see whether to sell or license your data.
  • Licensing a de-identified extract of own tables only.
  • A marketplace listing on the vendor's channel.
  • A program that connects read-only.

Where Warehouses Fit Among Source Systems and the Seller Process

A warehouse is a store of records that other systems produced, so its export comes after the inventory and rights checks and before de-identification and the choice of buyer.

The parent page compares export rights and mechanics by source system, and a warehouse often holds tables from Salesforce data export and licensing CRM records, SAP data export for licensing and QuickBooks data export for licensing; NetSuite data export for licensing is not yet published.

A program that connects to a warehouse is a case of buyer connectors vs native exports.

By data type, the tables correspond to CRM records for AI training, accounting, ERP and financial records for AI training and order, booking and transaction records. The full sequence to sell or license company data to AI companies runs in this order:

  1. Data sale readiness
  2. Data inventory for an AI data sale
  3. Customer consent and confidentiality clauses
  4. The data manifest
  5. Choosing a buyer for your data
  6. AI data licensing agreement

Warehouses of a Closed or Insolvent Company

A closed company's warehouse can be licensed only while someone with authority still controls the account and the vendor contract. Turing's intake form lists the statuses "Shut down ≤90 days ago" and ">90 days ago" (checked November 9, 2026), and no vendor retention period after termination is stated here.

The page on selling data from a shut-down or defunct company covers the rest, and exporting company data before shutting down is covered under insolvency exports.

Program and Marketplace Pages Named on This Page

The program terms above come from each program's own page, checked November 9, 2026, with Turing's license term re-checked November 17, 2026.

Frequently Asked Questions

Can a company sell the data in its Snowflake, BigQuery or Databricks warehouse to AI companies?

Yes, a company can license warehouse records that are its own where its customer contracts, any third-party license terms and privacy law allow, usually as a license rather than a sale. Mercor states that its SaaS & Subscription listing requires the company's own production databases and a cloud warehouse, and rules out agencies holding clients' databases (checked November 9, 2026).

Does sharing data through a warehouse mean licensing it for AI training?

No: a share gives access, and the license decides use. Snowflake, Google and Databricks document sharing without copying, and none of those pages treats AI training as a use (read November 17, 2026).

Can data a company bought on a marketplace be licensed on to an AI developer?

Only if the provider's listing terms allow it; the terms on the listing decide. Snowflake and Databricks providers each set their own terms, so the company reads them before any onward license.

What is warehouse data worth to an AI buyer?

No public price exists for warehouse data as such. Mercor states USD 55,000 to 350,000 for a SaaS company's whole dataset (checked November 9, 2026), with its own production databases and a cloud warehouse among the requirements. For whole-dataset figures, see how much company data is worth to AI companies.