Reported AI Data Licensing Deals, 2023 to 2026: A Review

1. Introduction#

AI developers need large volumes of text, images, audio, video and code, and the free supply of public web data has become contested. Rights holders have sued, restricted crawlers and, increasingly, signed contracts. Since mid-2023 a series of announced agreements has put commercial terms around data that was previously collected without them.¹²

The reporting is scattered and uneven. Some deals come with press releases that say little about price. Others surface through a securities filing, a leaked figure or an analyst's inference from a revenue line. A reader therefore faces two problems: finding the deals, and judging how much weight each reported number can bear.

This review addresses both for the period January 2023 to October 2026. It records for each verifiable arrangement what was licensed and what value was reported, states the basis of that value, and groups the deals by data type and seller. It does not estimate market size, forecast prices or assess the legality of training on copyrighted works; other papers in this series cover those subjects. The deals are also not representative: those disclosed are, by construction, those that someone chose to disclose or that leaked.

Section 2 sets out sources and inclusion criteria. Section 3 presents the findings, starting with a consolidated table. Section 4 discusses patterns, Section 5 lists limitations, and Section 6 concludes.

2. Methods and sources#

2.1 Inclusion criteria#

An arrangement was included if it met all of the following conditions:

1. It involved a license, sale, subscription, data-access agreement or purchase of data or content for use by an AI developer or AI product, including training, evaluation, retrieval or display in an AI assistant.
2. It was publicly reported, with at least one named source describing the parties and the nature of the arrangement.
3. It was announced or first reported between January 1, 2023 and October 20, 2026.
4. At least one source was a primary document (company announcement, securities filing) or a reputable secondary source such as a major news organization or a trade publication with named reporting, and we read the relevant portion before including the item.

We excluded rumors, social media posts and vendor marketing claims about unnamed customers. A vendor's announcement of its own program was included. Lawsuit outcomes that did not create a license were excluded from the table and discussed only as reference points.

2.2 Source handling and units#

Where possible we used the primary announcement. Several primary pages, including some company press rooms, could not be retrieved with our tooling, so for those deals we relied on secondary coverage that summarizes the announcement (see Section 5). For each arrangement we recorded the date, licensor, licensee, data type, any reported value and the basis of that value: a company filing or announcement, a named outlet citing unnamed sources, a figure derived from filings, a negotiating range or projection, or a vendor or interview claim.

Reported values use incompatible units: annual payments, multi-year totals, per-item prices, equity investments and auction bids. We did not convert them to a common unit, because the sources do not support the assumptions that would require. Where this paper states the amount of a deal it keeps the source's qualifiers, such as "could be" and "at least." Dates are as reported, and facts about ongoing matters are stated as of October 2026.

3. Findings#

3.1 Overview of the reported record#

Table 1 lists the 33 arrangements that met our criteria, spanning platforms, publishers, music and entertainment companies, stock-media libraries, a nonprofit encyclopedia, failing companies and expert-data programs. Arrangements covered by one source are listed in a single row.

Table 1. Publicly reported AI data licensing and data-sale arrangements, 2023 to October 2026, in approximate chronological order. "Undisclosed" means that no source we read reported a financial term. Values are reproduced as reported, with their basis stated in the text.

Two features of the table stand out. First, 18 of the 33 rows state outright that no financial term was disclosed, and others give only a negotiating range or a program structure. Second, among the 15 rows that report a value, the basis varies. Only a few come directly from a company filing or announcement (Reddit's aggregate, Perplexity's program terms and Disney's equity investment). Several come from named outlets citing unnamed people (Reddit and Google, News Corp, the New York Times and Amazon), one is derived from filings (Dotdash Meredith), and the remainder are negotiating ranges, projections, vendor or interview figures, or bid prices. Dates reflect when an arrangement was announced or first reported, and the table is shaped by source access as much as by the underlying market.

3.2 Platforms and community content#

Reddit. Reddit offers the best-documented example because its 2024 IPO forced disclosure. Its registration statement, as reported by TechCrunch, disclosed data licensing arrangements entered in January 2024 with an aggregate value of $203.0 million and terms of two to three years. Reddit expected to recognize at least $66.4 million of that in 2024. The buyers were not named.⁶ Reuters separately reported, citing three people familiar with the matter, that the deal with Google was worth about $60 million per year.⁷ One account is a company disclosure of an aggregate across unnamed counterparties, the other an anonymous-source estimate for one counterparty, so they are compatible but not identical.

In May 2024 Reddit and OpenAI announced that OpenAI would get access to Reddit's Data API, with OpenAI also becoming an advertising partner. Financial terms were not disclosed.¹² A later Sherwood News article, citing Bloomberg, said Reddit had struck deals worth $60 million per year with each of Google and OpenAI and was in early talks about dynamic pricing.³⁵ We could not corroborate the OpenAI figure in another source and do not tabulate it. Reddit's fiscal 2025 annual report states that other revenues rose as a result of content licensing agreements executed in 2024 and 2025, that substantially all licensing contract value comes from two partners, and that agreements are medium-term and may not be renewed or may be renewed on less favorable terms.³⁶ This is the most direct public statement we found about concentration and renewal risk.

Stack Overflow. Stack Overflow announced a partnership with Google Cloud on February 29, 2024. Gemini for Google Cloud would use Stack Overflow content through its OverflowAPI, developers could reach Stack Overflow from the Google Cloud console, and Stack Overflow would use Google Cloud for hosting. No financial terms were disclosed.⁸ On May 6, 2024 Stack Overflow announced a partnership with OpenAI: OpenAI would use OverflowAPI and show validated Stack Overflow knowledge in ChatGPT with attribution, while Stack Overflow would use OpenAI models for its own OverflowAI product. Again, the page states no financial terms.¹¹

Wikimedia. On January 15, 2026, the Wikimedia Foundation named Amazon, Meta, Microsoft, Mistral AI and Perplexity as Wikimedia Enterprise partners, alongside existing partners including Google. The product is described as a commercial service for large-scale reusers, delivered through on-demand, snapshot and realtime APIs. The announcement gives no fees.³⁰

3.3 News and magazine publishers#

News publishers account for the largest group of named agreements, and OpenAI was the first and most frequent counterparty.

The OpenAI wave. In July 2023 OpenAI and the Associated Press announced that OpenAI would license part of AP's text archive, with stories back to 1985. The price was not disclosed.² In December 2023 OpenAI and Axel Springer announced a multi-year partnership covering model training and ChatGPT summaries of selected articles, including paywalled ones. Neither side committed to exclusivity and the amount was not disclosed.¹ The Financial Times later reported that the deal was worth "tens of millions of euros" per year, with a one-time payment for historical content and recurring payments for current content, according to The Decoder.⁵ Le Monde and Prisa Media followed on March 13, 2024,⁹ and the Financial Times on April 29, 2024, both without disclosed terms.¹⁰

News Corp announced a deal on May 22, 2024 covering current and archived content from titles including The Wall Street Journal, Barron's, MarketWatch and the New York Post. Neither company disclosed the value. The Wall Street Journal reported that it "could be" close to $250 million over five years, and VentureBeat noted that OpenAI had reportedly paid other publishers $10 million or more per year.¹³ The Atlantic and Vox Media announced deals on May 29, 2024 without full financial terms.¹⁴ The Washington Post announced a partnership on April 22, 2025 and declined to share terms.²²

For Dotdash Meredith, Adweek reported that OpenAI pays a minimum of about $16 million per year, based on filings of its parent IAC. It is a minimum because it reflects only the "fixed" component, as an IAC executive described on an earnings call, with a variable component calculated separately. The article cites Q3 2024 licensing revenue as up about $4.1 million year over year but does not show the arithmetic.¹⁵ We treat the number as a derived minimum, not a contract term.

Other buyers. Meta announced a deal with Reuters in October 2024, giving Meta AI real-time news with citations and links; the amount was not disclosed, and the companies did not say whether the deal covered training Llama.¹⁸ In December 2025 Meta announced commercial AI data agreements with USA Today, People Inc, CNN, Fox News, The Daily Caller, the Washington Examiner and Le Monde, with terms not disclosed.²⁹ Google signed its first reported deal with a news publisher, the Associated Press, on January 15, 2025, to provide a real-time feed for Gemini. Neither company disclosed the payment, and Google would not say whether it would credit AP.²⁰

Amazon's reported deals include the clearest published price outside OpenAI. TheWrap, citing the Wall Street Journal, reported on July 30, 2025 that Amazon will pay the New York Times between $20 million and $25 million per year for news, NYT Cooking and The Athletic, for AI training and for summaries in products including Alexa. The deal had been announced in May and was described only as multiyear.²³ Glossy reported that Condé Nast and Hearst signed multi-year agreements with Amazon for the Rufus shopping assistant, with terms undisclosed. Hearst confirmed its deal, while Condé Nast's confirmation came from unnamed sources at the time.²⁴

Marketplaces and revenue-share programs. Perplexity announced Comet Plus in August 2025 as a $5 subscription tier with a $42.5 million pool for publishers and 80% of pooled subscription revenue going to participating publishers, divided across visits, citations and task use. Perplexity declined to say which publishers had joined.²⁵ Microsoft's Publisher Content Marketplace, expanded in February 2026, lets publishers set their own terms and be paid based on the value their content delivers. Named co-designers include Business Insider, Condé Nast, Hearst, People Inc, the Associated Press, USA Today Co. and Vox Media. No rates were reported.³¹ These programs shift pricing from negotiated lump sums toward usage.

3.4 Books, academic publishers, music and entertainment#

Books. Publishers Weekly reported on November 19, 2024 that HarperCollins offered authors an opt-in deal to license select nonfiction backlist titles to an AI company for three years, at $5,000 per title split evenly between author and publisher, not counted against royalties. HarperCollins confirmed only an agreement with an unnamed company, and Bloomberg, citing an anonymous source, reported the counterparty as Microsoft. The AI company agreed to limit verbatim reproduction and not to train on pirated content.¹⁹ This is one of very few per-title prices in the record, and it is a negotiated price for a restricted, time-limited use.

Academic publishers. Times Higher Education reported in July 2024 that Informa, parent of Taylor & Francis, had a deal with Microsoft worth more than $10 million (£7.8 million) in its first year and running until 2027, under which Microsoft may use content to improve its AI systems. Academics criticized the lack of publicity and the absence of author payment, while Taylor & Francis said the deal limits verbatim reproduction and preserves royalties.¹⁶ Separately, infoDOCKET, relaying The Bookseller, reported in August 2024 that Wiley expected about $44 million from AI partnerships, of which about $23 million had already been earned and about $21 million more was expected in the fiscal year. The report did not name the counterparties, and said authors had no opt-out.¹⁷ The Wiley figure is a company-side projection described secondhand and should be read with that limit.

Music and entertainment. On October 29, 2025, Universal Music Group settled its copyright suit against Udio and announced a partnership on a licensed platform planned for 2026, with new license agreements covering recorded music and publishing. The article reports a compensatory payment but gives no amount.²⁷ Disney and OpenAI announced on December 11, 2025 a three-year agreement under which Sora and ChatGPT Images can use more than 200 Disney characters, excluding talent likenesses and voices, with Disney taking a $1 billion equity stake in OpenAI. Disney's CEO described exclusivity "at the beginning" of the term without explanation.²⁸ The $1 billion is an investment by the licensor in the licensee, not a license fee, and the sources do not say how the license is priced.

Litigation reference point. The class settlement in the authors' case against Anthropic is not a license, but it is often cited as a price signal. It was announced in September 2025 at $1.5 billion for an estimated 500,000 books, about $3,000 per book, after a judge found that training on copyrighted books was fair use but that Anthropic had taken millions of books from pirate sites.³⁷ Final approval was granted on July 20, 2026.³⁸ Because it compensates for acquisition from pirate sources, we do not place it in Table 1.

3.5 Stock media, images and video#

Shutterstock. Shutterstock and OpenAI announced a six-year agreement in July 2023 under which OpenAI gets a license to additional image, video and music libraries and associated metadata, and Shutterstock gets priority access to OpenAI technology and continued use of DALL-E on its platform. The report discloses no payment amounts.³ Reuters, as relayed by The Decoder in April 2024, reported that Shutterstock had deals with Apple, Meta, Google and Amazon covering hundreds of millions of items, with initial deals of $25 million to $50 million that have since been expanded.⁴ Shutterstock's annual report for fiscal 2024 shows a "Data, Distribution, and Services" revenue line of $175.3 million in 2024, up from $137.3 million in 2023 and $38.5 million in 2022. The filing says there is increased demand for its metadata for machine learning and that revenue timing depends on delivery of large metadata sets. That line is broader than AI training licensing, so it is an upper bound on AI-data revenue, not a measure of it.³⁹

Photobucket. Reuters, via The Decoder, reported in April 2024 that Photobucket's chief executive was negotiating with several technology companies to license 13 billion photos and videos at prices of 5 cents to $1 per photo and more than $1 per video. One prospective buyer wanted billions of videos. The executive pointed to a terms-of-service update granting the company an "unrestricted right" to sell uploaded content for AI training, and the article noted that the FTC had warned in February 2024 against retroactively changing terms to allow AI training.⁴ These were prices under discussion, and we found no source reporting a completed Photobucket deal.

Other image and video licensing. The same Reuters report said Freepik had licensed most of its 200 million image archive to two major technology companies at 2 to 4 cents per image, and listed Defined.ai prices of $1 to $2 per image and $100 to $300 per hour for longer films.⁴ Bloomberg, relayed by The Decoder in January 2025, reported that companies including OpenAI and Google pay $1 to $4 per minute for unpublished creator video, through intermediaries such as Troveo AI.²¹ Getty Images and Perplexity announced a multi-year license for image display on October 31, 2025. No payment was reported, and a source said it was not a traditional lump-sum deal because Perplexity does not train its own foundation models.²⁶

3.6 Bankruptcy and wind-down sales#

A newer category involves companies selling internal data because they are failing. Forbes, relayed by Gizmodo in April 2026, reported that SimpleClosure had launched Asset Hub to help startups license code and workplace data such as Slack messages and emails, with nearly 100 deals in a year and payouts of $10,000 to $100,000 per company.³² MIT Technology Review reported in October 2026 that SimpleClosure had handled more than 150 deals, that 47% of its deals now include code or data sales, and that it expected to return more than $10 million to startups by the end of 2026. Buyers included foundation labs, vendors building simulated workspaces and established tech companies. One crypto startup sold its code base for roughly $25,000, with about $15,000 more pending for workspace data. The article said buyer quotes vary widely because valuations rest on estimates of how many useful tasks a dataset could yield.⁴⁰

Spirit Airlines. The largest reported example is Spirit Airlines, which stopped flying in May 2026 and sold corporate data in its bankruptcy case. Time reported on August 25, 2026 that Google won the auction with a $10 million bid for data including about 100 million emails, 500 million Microsoft Teams messages and 30 million lines of code, with Mercor bidding $7.5 million. A third party would scrub personal information, Google's spokesperson said, and the dataset would include no customer data.³³ The Next Web gave the bidding sequence: Google opened at $5 million, Mercor countered at $5.2 million and then offered $7 million if it could take the raw data first and anonymize it itself, and Google closed at $10 million as reported by Reuters on August 17. The sale agreement was filed on August 14 and the sale hearing was initially set for August 19. The article itemized the included data, which spans email and collaboration systems, engineering repositories, operations and booking records, corporate material and employee records, and listed as excluded customer records such as 97.5 million customer profiles, call recordings and chat sessions.³⁴

Subsequent reporting shows the matter was contested. The Association of Flight Attendants-CWA objected that the sale agreement preserves links between datasets that could allow reconstruction of information about individuals or small groups, and the hearing was postponed to September 9, 2026.⁴¹ On September 4, AI Weekly, citing a Bloomberg-reported court filing, reported that micro1 had filed a late $12.5 million bid.⁴² CSO Online reported that the sale's fate remained unresolved and described the dataset differently, including about 600 million email and chat records from 17,000 employees and over 30 million recorded customer service calls.⁴³ That conflicts with The Next Web's itemization, which lists call recordings as excluded and gives 175,658 employee records. We could not reconcile the two, and we found no report of a final ruling as of October 2026. MIT Technology Review quotes an attorney calling such data "legally radioactive" and notes that no federal law bans such sales.⁴⁰

3.7 Expert-labor and data-buying programs#

A different group of entities does not license a stock of existing content. They hire people to produce data, or they have recently started buying operational data from companies. The public record here consists mostly of funding announcements and company statements, with a few reported prices.

Mercor. TechCrunch reported on October 27, 2025 that Mercor raised $350 million at a $10 billion valuation, up from $2 billion in February 2025. It pays more than $1.5 million per day to contractors, whose more than 30,000 experts earn over $85 per hour on average, and it told investors it was on track for $500 million in annual recurring revenue, a projection.⁴⁴ A later TechCrunch article reported that Mercor pays experts up to $200 an hour to fill out forms and write reports, that customers include OpenAI, Anthropic and Meta, and that most revenue comes from a few labs, all from the chief executive's statements and not independently verified. It also reported that Mercor had sought a startup CTO who could authorize access to a production codebase, and that many labs reportedly stopped working with Scale AI after Meta's investment.⁴⁵

Turing. TechCrunch reported in March 2025 that Turing raised $111 million at a $2.2 billion valuation, with annualized revenue of about $300 million, up from $167 million when the round was priced. It supports AI labs such as OpenAI with engineering work and says it works with about 4 million coders.⁴⁶

Scale AI. NBC DFW reported on June 12, 2025 that Meta was investing $14.3 billion in Scale AI for a 49% stake, with no voting power per a Scale spokesperson, and that Scale supplies data for AI models to customers including Google, Microsoft and OpenAI.⁴⁷ This is an equity transaction with a data supplier, not a license.

micro1. On October 9, 2026, micro1 announced a commitment of $1 billion over 12 months to acquire and license enterprise operational data through a Company Data Partnerships program, financed by Citi and Hercules Capital, to build reinforcement learning environments from de-identified company data.⁴⁸ Its program page lists payment tiers of $100,000 or more for qualified enterprise data partnerships, $500,000 or more for large operational datasets or ongoing participation across teams, and $1 million or more for highly unique proprietary data. It says compensation depends on quality, uniqueness and value, and that eligibility favors operationally mature companies with at least 30 employees.⁴⁹ These are a buyer's own published tiers and a stated budget. We found no independent report of completed purchases under the program, apart from its bid for the Spirit data.

Market context. Epoch AI, from interviews, reported that contracts for reinforcement learning environments are often six to seven figures per quarter, typical task prices are $200 to $2,000, exclusive deals cost roughly four to five times as much as non-exclusive ones, and high-fidelity replicas of complex products such as Slack can cost around $300,000.⁵⁰ TechCrunch, citing The Information, reported that Anthropic leaders had discussed spending more than $1 billion on such environments over a year, without confirmation from Anthropic.⁵¹ These are interview estimates and a second-hand report of a discussion, not contract disclosures.

3.8 Reported values and what they imply#

Table 2 groups the reported figures by category and unit. It keeps each figure in the unit the source used.

Table 2. Reported values by category, with the unit and basis of each figure. Figures are not directly comparable across rows.

Three observations follow, stated cautiously. First, the numbers describe four different economic objects: a recurring fee for access to an updated corpus (Reddit, Axel Springer, the New York Times), a one-time dataset purchase (Spirit, startup wind-downs), a per-item price (Photobucket, Freepik, HarperCollins, video minutes), and payment for newly produced labor (Mercor, Turing, environment vendors). Comparing a $60 million annual fee with a $10 million auction price says little without knowing what each bought.

Second, per-item prices are low, and aggregate values are high only when the item count is very large. Photobucket's range was a posture in negotiations, not a completed deal.⁴ The $5,000 per-title book price sits at the other end, for curated long-form text under a restricted license.¹⁹

Third, the highest values attach to rights holders with distinctive, continuously updated content that also competes with AI answers: News Corp, Reddit, the New York Times and Axel Springer. This is an observation about a sample skewed toward newsworthy deals, not a causal finding.

4. Discussion#

4.1 Disclosure is the exception#

About half of the arrangements in Table 1 state that no financial term was disclosed, including the AP, Financial Times, Le Monde, Washington Post and Stack Overflow agreements. Numbers usually appear because a public company had to disclose them (Reddit, IAC, Shutterstock), because a reporter obtained them from people familiar with a deal (News Corp, Reddit and Google, the New York Times and Amazon), or because a participant chose to publicize them (micro1, Mercor). Each route selects for a type of deal, so the visible distribution of values is probably skewed upward relative to all contracts. We cannot estimate the size of that skew.

4.2 From training rights to access and display#

Early agreements, such as the AP and Axel Springer deals, named model training alongside display in a chatbot.²¹ Later reports increasingly describe real-time feeds, grounding and attributed summaries: Google's deal with AP, Meta's with Reuters, Perplexity's with Getty, and Microsoft's marketplace.^20182631 The Perplexity and Microsoft programs tie payment to usage.²⁵³¹ This suggests some current deals buy ongoing access and attribution, not a one-time right to train. The sources do not say how large usage-based payments are, and we make no estimate.

4.3 Structure behind the headline number#

Where structure is described, it is rarely a single number. Axel Springer was reported to combine a payment for historical content with recurring payments for current content.⁵ Dotdash Meredith's reporting describes fixed and variable components, and a cited pitch deck says most OpenAI publisher deals have both.¹⁵ Vox Media's statement mentions attribution, limits on summarizing, links back and early access to OpenAI products,¹⁴ and Reddit's annual report notes API performance requirements and medium-term durations.³⁶ Headline figures therefore leave out terms that affect value: non-cash consideration, attribution, volume limits and renewal.

4.4 Concentration and renewal risk#

Reddit's filing says substantially all licensing contract value comes from two partners and that renewals are uncertain.³⁶ Shutterstock's says data revenue fluctuates with the timing of large deliveries.³⁹ Mercor's chief executive said most revenue comes from a few labs.⁴⁵ Sellers' AI revenue appears concentrated and lumpy, though the record does not show whether that holds for smaller sellers.

4.5 Rights, privacy and contested sales#

The Spirit case shows that sales of internal corporate data can be contested on privacy grounds even after an auction. The union objected that preserved links between datasets could allow re-identification, while Spirit said records would be de-identified and Google said a third party would scrub personal information.⁴¹³³ The sale included employee, payroll and tax records and excluded customer records.³⁴ MIT Technology Review quoted a law professor noting that proceeds typically go to creditors and investors, not employees.⁴⁰ Photobucket's reliance on a terms-of-service change shows that sellers' rights may depend on terms users accepted before AI licensing existed, and the FTC had warned against retroactive changes.⁴ We do not analyze the law here.

4.6 Reading reported values#

Four cautions follow. Identify the unit before the amount. Identify the basis, because a filing, an anonymous-source report and a vendor statement carry different weight. Note qualifiers such as "could be," "at least" and "about." And do not transfer a figure to a dissimilar dataset: the large figures belong to parties with large, distinctive archives, and micro1, the one program publishing tiers for operational data, says compensation depends on quality, uniqueness and value.⁴⁹ Epoch's task prices and exclusivity premiums refer to reinforcement learning environments, not all data.⁵⁰

5. Limitations#

Selection. The record is built from public reports. Confidential deals are invisible, and reported deals are probably larger or more newsworthy. We cannot measure representativeness.

Source access. Several primary announcements, including pages on some company and press-agency websites, could not be retrieved. For those deals we relied on secondary reports, which may omit details or contain errors the originals do not.

Secondhand figures. Many values are relayed through a second outlet: News Corp through VentureBeat citing the Wall Street Journal, the New York Times through TheWrap citing the Wall Street Journal, Axel Springer through The Decoder citing the Financial Times, Wiley through infoDOCKET citing The Bookseller, and Shutterstock and Photobucket prices through The Decoder citing Reuters. We did not open the original articles and cannot exclude relaying errors.

Conflicting accounts. The Spirit dataset is described differently by two sources. The Sherwood News statement about an OpenAI deal with Reddit was not corroborated. Some dates are report dates, not signature dates.

Changing facts. Several matters were unresolved as of October 2026, including the court outcome for the Spirit sale and Reddit's reported renegotiation talks.

Heterogeneity and scope. We did not normalize values or compute totals across categories. We did not systematically review non-English reporting or specialist sources on healthcare, financial or scientific data licensing.

6. Conclusion#

Between 2023 and October 2026, publicly reported AI data licensing arrangements grew from a few news-archive deals into a varied set that includes community platforms, book and journal publishers, music and entertainment companies, stock-media libraries, bankrupt companies and programs that buy operational data. Most terms remain confidential, and the values that were reported differ in unit, term and evidential basis, so they cannot be ranked or averaged without strong assumptions.

Three findings are best supported. Financial terms are the exception, not the rule. Reported values describe different things (access fees, one-time dataset sales, per-item prices, usage-based shares and newly produced labor), and the largest figures belong to parties with large, distinctive archives. And contract structures often include fixed and variable components, attribution and renewal terms that a headline number omits.

Open questions include how usage-based programs pay out in practice, how courts will treat sales of internal corporate data, and whether the operational-data programs launched in 2026 complete deals at their published tiers.