CuriosityStream Q2 2026 press release discloses $14.1M licensing revenue driven by AI training partnerships
CuriosityStream turned a rights-cleared factual-media archive into a sell-route licensing business generating $14.1M in a single quarter at 73% gross margin, by curating video, audio and code corpora into training-ready datasets before they leave the perimeter.
What happened
- Q2 2026 licensing revenue of $14.1M, up 48% year-on-year, disclosed August 12, 2026.
- Gross margin expanded to 72.8% in Q2 2026 versus 53.4% in Q2 2025.
- Licensing sits on a corpus of 3M+ hours of rights-cleared video and audio plus an 880 billion token private code corpus.
- Management frames the licensing opportunity as three pillars: video licensing to traditional media; audio and video licensing for AI training; and private code licensing for AI training.
- Specific Q2 deals included licensing millions of tokens of code for AI training/RL/evaluation, a 40,000-segment HDR dataset, and synchronized multi-camera action sequences to a video research lab.
- Record Q2 net income of $8.9M (up 1,033%) and Adjusted EBITDA of $11.4M; full-year revenue guidance raised to $77-82M.
Who is involved
US factual-media streaming company that has repositioned as a provider of AI model training datasets leveraging a rights-cleared media corpus spanning millions of hours of premium video and audio, 850 billion tokens of production-grade code, and bespoke datasets
Nasdaq: CURI; founded 2015; headquartered Silver Spring, Maryland; Q2 FY2026 licensing revenue of $14.1M per the cited press release
The reading
CuriosityStream itself rights-clears, curates and packages its video, audio and code corpora into training-ready datasets before licensing them, per its own about-blurb describing millions of hours of premium video and audio, 850 billion tokens of production-grade code, and dozens of bespoke datasets created with proprietary content intelligence tools. The buyers (AI labs) then train on that packaged substrate.
CuriosityStream keeps the licensing fee at high gross margin (management flagged 73% margin on this line) and holds the buyer relationship directly; because it did the rights-clearing and curation, the rake is owned rather than on loan to a partner.
To an AI-lab buyer, a rights-cleared factual video/audio corpus plus production-grade code is legally defensible training data — a scarce alternative to scraped web content, and structured around real-world domains (science, nature, history) that models otherwise learn poorly.
Licensed use of the corpora and datasets crosses to the AI-lab buyer under contract; underlying copyright and the SVOD/linear consumer business remain with CuriosityStream.
Not determinable — $14.1M in a single quarter at 73% margin is a strong early print, but public disclosures do not name the counterparties, term length or per-token pricing, so whether CuriosityStream is capturing full market rate is unknown.
Why it matters
CuriosityStream is a clean public example of a company whose substrate already earned inside a wrap (the SVOD subscription) and that deliberately built a sell line on top of the same asset — with rights-clearing and curation done in-house so the rake stays owned rather than shared with a data broker. The interesting move is that the same underlying asset (factual video/audio) is now priced against a foundation-model buyer's P&L rather than a consumer subscription P&L, and the code corpus was assembled specifically for that buyer — suggesting the substrate was extended, not just repackaged. Worth using to illustrate how enrichability for an AI-lab buyer (rights-cleared, domain-structured, production-grade) creates a price surface a scraped-web substitute cannot match.
Sources
- nasdaq.com — primary
- en.wikipedia.org — party background
- sec.gov — party background
Announced 2026-08-12 · Added to the register 2026-08-25