SMWB and CURI Show Two Models for AI Data Licensing

From August 12 through August 26, 2026, Similarweb Ltd. (SMWB), CuriosityStream Inc. (CURI), Zhihu Inc. (ZH), and Nine Entertainment (NNMTF) described different purchasing and payment models for static archives, continuously refreshed data, and content used to ground AI answers.[1][2][3][4]

---

Ongoing data utility determines whether customers pay again

An AI model can ingest an existing collection of images, videos, or articles once. A multiyear contract for that archive may therefore recognize much of its revenue near the beginning. In production, however, agents need current information about web traffic, markets, and professional subjects. Enterprise assistants also need content they can retrieve and attribute. Customers are more likely to pay repeatedly when they are buying updates and recurring access.

CuriosityStream shows that an archive owner can raise the value of static material by reorganizing it. The company said it is packaging intellectual property around AI developers' requirements instead of simply offering large quantities of video, including high-dynamic-range footage covering different lighting, angles, and partial obstructions.[2] That approach still depends on whether the existing archive contains the conditions a buyer needs, so it does not automatically create subscription revenue.

---

AI licensing contracts already have different revenue patterns

Similarweb sells changing digital-world data to model developers and AI agents. Management explicitly described its large deals as annual recurring revenue contracts with an immaterial one-time component, if any. Remaining performance obligations reached $345 million at the end of the second quarter, up 26% year over year.[1] Because websites, apps, and consumer behavior keep changing, customers need continuing updates.

CuriosityStream reported $8.5 million of licensing revenue related to the disposal of a subsidiary in the second quarter. It also recorded $5.3 million of barter revenue, up from $1.9 million a year earlier.[2] Those components show why its licensing line can be affected by specific transactions and noncash arrangements and should not be treated automatically as recurring revenue.

Zhihu and Nine described another repeat-use case. Zhihu's open data API had attracted more than 17,600 professional developers, and clients for its AI content-asset offering increased 50% sequentially. Management nevertheless said the business did not yet represent stable or scalable revenue.[3] Nine signed licenses that let corporate customers use its journalism to ground large language models and also reached a news-content agreement with Microsoft Copilot. Those transactions remained below the A$25 million or 5% revenue materiality threshold discussed on the call.[4]

---

Control shifts from archive size to freshness, attribution, and production

The disclosures support a bounded conclusion: AI data licensing is splitting according to data utility. Finite archives can still generate large but concentrated recognition. Refreshed data, attributable content, and data produced for new tasks have a stronger basis for renewals, usage fees, or successive projects. A multiyear label alone says little about revenue quality.

Useful follow-up measures include renewal rates, remaining performance obligations, API calls, customer counts, and the cash composition of licensing revenue. The limits are equally important. Similarweb has identified recurring contracts, while Zhihu still calls commercialization early, Nine's agreements remain below its materiality threshold, and CuriosityStream's figures include special transactions. The evidence supports a bifurcation in business models, not a claim that every content owner will earn stable AI revenue.

---

Companies exposed to the same change

• Reddit (RDDT): Its community corpus updates continuously and already generates data-licensing revenue. Renewals can test whether live content is more likely to support repeated payment when customers expand from training to retrieval.
• S&P Global (SPGI): Its proprietary financial data changes continuously and reaches customer workflows through LLM-ready APIs and connectors, exposing it to a mix of consumption and dataset licensing.
• Innodata (INOD): It produces new reasoning, evaluation, and embodied-AI data for frontier model labs. This work may capture spending when existing archives do not meet a task, although project continuity still requires confirmation.

---

Sources

[1] Drillr · Similarweb Ltd. · August 12, 2026 · FY2026 Q2 earnings call

Hey, Scott. Thanks for the question. Those deals are ARR deals. And if there is a one-time element, it's really quite negligent and immaterial. And usually it's recognized on the initiation of the deal or a couple of months later at max. But those deals are pure ARR deals with, again, a small and immaterial amount of one-time, if at all.

[2] Drillr · CuriosityStream Inc. · August 12-13, 2026 · FY2026 Q2 earnings call and Form 10-Q

[3] Drillr · Zhihu Inc. · August 26, 2026 · FY2026 Q2 earnings call

[4] Drillr · Nine Entertainment · August 25, 2026 · FY2026 earnings call

This article identifies potentially overlooked industry changes and companies. It is not investment advice.