Key Takeaways
- OpenAI is in talks with a new AI‑video startup co‑founded by Hollywood mogul Jeffrey Katzenberg, former Sora research lead Bill Peebles, and ex‑Dropbox CFO Sujay Jaswa.
- The startup hopes to obtain access to OpenAI’s Sora model to build and sell video‑creation tools for filmmakers, though it remains unclear whether it will receive the model’s weights or only a deep‑customization layer.
- While OpenAI has begun winding down its broader video‑AI efforts to focus on core enterprise products and prepare for an IPO, it still sees value in collaborating with experts like Peebles who intimately understand Sora’s architecture.
- In contrast, Kaon AI’s CEO Jay Dang advocates a “generative‑first” approach that creates fresh video in real time rather than recommending existing content, arguing that recommendation systems hit a ceiling when no suitable match exists.
- The divergent strategies highlight a growing split in the AI video market: one path leverages powerful pretrained models for customization, the other builds entirely new, user‑driven narratives on the fly.
OpenAI’s Ongoing Dialogue with a New AI Video Startup
According to a report by The Information published on Monday, Oct. 12, OpenAI is currently in discussions with a startup that is developing artificial‑intelligence video tools aimed at filmmakers. The source familiar with the matter told the outlet that the company “wants to access Sora and use it to sell software and other services.” This suggests OpenAI is exploring a licensing or partnership model that would allow an external party to build commercial products on top of its video‑generation model while still benefiting from the underlying technology.
Founders and Background of the Startup
The startup in question was co‑founded by three notable figures: Hollywood mogul Jeffrey Katzenberg, former Sora research lead Bill Peebles, and ex‑Dropbox chief financial officer Sujay Jaswa. Both Katzenberg and Jaswa list themselves as co‑founders of the tech investment firm WndrCo on their LinkedIn profiles, indicating a strong pedigree in both entertainment and finance. The report notes that the exact name of the venture remains unclear, but the combination of Katzenberg’s industry connections, Peebles’ technical expertise, and Jaswa’s operational experience positions the startup to bridge creative storytelling with scalable AI infrastructure.
Uncertainty Around Model Weight Access and Customization Potential
A central point of ambiguity in the negotiations is whether the startup will be granted access to Sora’s model weights—the internal parameters that dictate how the model responds to prompts. The Information added that it is “also unclear whether this startup will be given access to Sora’s weights, or the settings that shape how the model responds to queries.” However, the source indicated that even if the weights are not shared, the startup would still receive “deep enough access to substantially customize it.” This nuance reflects a common industry practice where providers offer fine‑tuning APIs or adapter layers that enable significant model adaptation without exposing the core weights.
OpenAI’s Strategic Shift: Winding Down Sora and Video AI Efforts
Despite these discussions, OpenAI introduced Sora in 2025 but has begun winding down its broader video‑AI initiatives earlier this year. The move is part of a larger effort to scale down what the company calls its “side quests” so it can concentrate on core business lines and prepare for an initial public offering (IPO) tied to enterprise applications. By reducing investment in experimental video products, OpenAI aims to allocate more resources toward its flagship language models and enterprise‑focused services, which it views as higher‑margin and more aligned with its long‑term growth trajectory.
Kaon AI’s Alternative Approach to Video Generation
In a separate development, PYMNTS recently spoke with Jay Dang, CEO and co‑founder of Kaon AI, which takes a markedly different stance on video AI. Rather than relying on recommendation algorithms that surface existing content, Kaon’s platform “generates new video in real time, crafting a story world designed around a user’s interests as they interact with it.” Dang explained the philosophy succinctly: “If the perfect piece of content for the user was never made, the algorithm cannot help the user.” This quote captures the startup’s belief that recommendation systems are fundamentally limited by the bounds of their training libraries.
Industry Implications and Contrast Between Approaches
PYMNTS elaborated on the distinction, noting that “Recommendation systems can only point to the closest available match, and when nothing in the library lines up, the system is out of options.” In contrast, Kaon “removes that ceiling by making the content on the spot, based on what the user chooses in the moment.” This real‑time, generative model allows users to explore narratives that have no pre‑existing analogue, potentially unlocking creative possibilities that recommendation‑driven platforms cannot offer. The tension between these two philosophies—leveraging powerful pretrained models for customization versus building entirely new content on demand—highlights a nascent fault line in the AI video market.
Future Outlook for AI Video Tools and Market Trends
Looking ahead, the outcome of OpenAI’s talks with the Katzenberg‑Peebles‑Jaswa venture could shape how firms balance access to foundational models with the need for proprietary differentiation. If the startup secures substantial—if not full—access to Sora’s capabilities, it may accelerate the availability of high‑quality, filmmaker‑focused video tools that blend Hollywood storytelling expertise with cutting‑edge AI. Simultaneously, companies like Kaon illustrate a growing appetite for systems that generate bespoke video experiences in real time, particularly for interactive media, gaming, and personalized advertising. As OpenAI refocuses on enterprise‑grade offerings and readies for an IPO, the market may see a bifurcation: one track dominated by licensed, customizable large models, and another driven by agile, generative‑first startups that create content dynamically. Both pathways underscore the expanding role of AI in video production, promising creators more tools—and more choices—than ever before.
https://www.pymnts.com/news/artificial-intelligence/2026/openais-video-model-sora-could-get-a-second-act/