What can recent AI data deals teach a business owner?
That the price is only one part of an AI data licensing deal. The terms that limit how your data can be used, how and when you get paid, and what happens when the project ends matter just as much. Several deals disclosed in November 2026 show those terms in unusual detail.
Most AI licensing deals stay private. When Adweek reported on November 4, 2026 that Anthropic had struck its first content licensing deal, with the Associated Press, neither company disclosed the financial terms or the timeframe. Public company filings are one of the few places where the structure of a data deal is written down, which makes them worth reading even if your business is much smaller.
How are large data licensing deals structured?
They usually combine a narrow permitted use, a schedule of payments and rights that flow back to the data owner. Tempus AI's quarterly report, filed November 2, 2026, describes a deal in which Tempus licenses a de-identified, multimodal dataset to an AI company called Pathos "for the sole purpose" of helping develop and train a specific AI model. Pathos agreed to pay $200 million in data license fees over three years, including $50 million paid upfront as of April 2025. Tempus also receives a license to use the finished model, with certain restrictions.
The same filing shows that payment does not have to be all cash. Pathos may pay up to 50% of the data license fees in its own preferred stock, and it paid $8.8 million that way in both June and August 2026.
A second filing shows how buyers can commit to minimums. In an 8-K filed November 18, 2026, Tempus disclosed that AstraZeneca committed to $200 million in fees from 2027 through 2031 across Tempus's products and services, not only data. The amendment sets a $40 million annual target and a $32 million minimum each year, adds a year if the total falls short, and makes AstraZeneca pay any shortfall if it terminates early.
These are large healthcare deals, and your numbers will be different. The lesson is in the structure, not the size.
Which terms protect your data after you hand it over?
The terms that limit use, block resale and require the data back at the end. On November 10, 2026, the Korean Publishers Association signed an agreement with five AI data companies to supply publishing data for AI. As reported by Seoul Economic Daily, the partner firms may use the data only within the agreed business purpose, must get the association's prior consent before providing or selling it to third parties, and must return or destroy the data and any copies when a project ends. Rights in the works stay with the rights holders, and individual supply terms depend on their consent.
Those protections answer questions every data owner should ask:
- What exactly may the buyer do with the data? Training a model, answering questions in real time and building a product are different uses. The Internet Licensing Collective, for example, says AI companies license only what they need, such as training or retrieval. Our guide to AI search access versus training permission explains the difference.
- Can the buyer pass it on? Without a consent requirement, your data could end up with companies you never chose.
- What happens at the end? Return or deletion of the data and copies should be written into the agreement, with a way to confirm it happened.
- Who still owns it? A license should grant permission to use the data, not transfer ownership.
How can you get paid for your data?
Through upfront fees, guaranteed minimums, usage-based payments, equity or a mix. The Tempus filings show upfront payments, annual minimums and payment partly in stock. The Internet Licensing Collective, which launched on November 12, 2026 with Yahoo Scout as its first commercial licensee, pays participating publishers based on how their content is used. We covered how that collective works in can your business get paid when AI uses your content?
Each option carries a different risk. Upfront fees and minimums give you certainty. Usage-based payments can grow, or shrink, with the buyer's product. Stock in a private company is not cash, and what it is worth depends on that company's future.
What makes data valuable to an AI company?
Data that is hard to get elsewhere, accurate, current and legally clean. Adweek's reporting argues that the AP deal gave Claude timely, attributed election information that it could not get from the open web on election night. Tempus's dataset is valuable partly because it is large, multimodal and de-identified.
Not every deal is about training. On November 18, 2026, OpenAI and Le Groupe Figaro announced a partnership that makes Le Figaro's current reporting and archive discoverable in ChatGPT through summaries and excerpts with links to the original stories. For some businesses, attribution and links in AI answers may matter as much as a fee.
What should you do before you talk to an AI company?
Know what you own, what you are allowed to license and what you want in return. A practical starting point:
- Inventory your data. List the content, records and datasets you hold, and note which are unique to your business.
- Check your rights and obligations. Customer and personal information can carry privacy and contract limits. Our guide to what business data you can put into AI covers the basic boundaries.
- Decide the uses you will allow. Training, real-time answers, or both, and for how long.
- Choose a payment structure. Weigh upfront fees and minimums against usage-based or equity payments.
- Write down your exit terms. Onward-transfer limits, return or deletion, reporting and audit rights.
- Get legal advice before you sign. Licensing terms are contracts, and the details matter.
How DigitalOS can help
DigitalOS helps business owners understand what their data may be worth before they license it to AI companies. We can help you inventory your data and content, identify what is genuinely distinctive, think through permitted uses and payment structures, and prepare questions for your legal team before you talk to a buyer. Learn more about our AI Data Advisory work, or start with the DigitalOS AI diagnostic to see where your business stands today.
Sources and limitations
Reviewed November 19, 2026. The Tempus deals are large healthcare agreements described in public SEC filings; the Korean agreement terms come from a news report, not the agreement itself; the Anthropic and AP terms were not disclosed. Deal terms vary widely. This is general information, not legal or financial advice, and external deals are not DigitalOS results or guarantees.
- Tempus AI — Form 10-Q for the quarter ended September 30, 2026 (filed November 2, 2026)
The Pathos data license: sole-purpose use, $200 million in fees over three years, $50 million upfront, up to 50% payable in preferred stock, and Tempus's license to the finished model.
- Tempus AI — Form 8-K (filed November 18, 2026)
AstraZeneca's $200 million commitment for 2027 to 2031, $40 million annual target, $32 million annual minimum, shortfall extension and termination terms.
- Seoul Economic Daily — "Korean Publishers Build Licensed Data Pipeline for AI Training" (November 10, 2026)
Purpose limits, consent before third-party transfer, return or destruction of data, and rights retained by rights holders.
- Adweek — "Anthropic Strikes Its First Licensing Deal, With the AP, for Election Coverage" (November 4, 2026)
Anthropic's first publisher licensing deal; financial terms and timeframe not disclosed.
- Internet Licensing Collective — "The Internet Licensing Collective (ILC) Launches with Yahoo Scout as First Commercial Partner" (November 12, 2026)
Collective licensing, license types such as training and retrieval, and usage-based compensation.
- OpenAI — "Le Groupe Figaro and OpenAI partner to bring Le Figaro journalism to ChatGPT" (November 18, 2026)
A partnership built on discovery, summaries, excerpts and links in ChatGPT.




