The Lion King of AI Video: Katzenberg, Ex-Sora Chief Peebles and Jaswa Team Up for a Filmmaker-First Startup

One of the most consequential talent pairings of the AI era was reported on the evening of September 9 by The Information: Jeffrey Katzenberg — the DreamWorks co-founder behind Shrek and Kung Fu Panda, former chairman of Walt Disney Studios and principal of investment firm WndrCo — is teaming up with Bill Peebles, the researcher who led OpenAI’s Sora video generation and world simulation effort, and Sujay Jaswa, the former CFO of Dropbox and Katzenberg’s co-founder at WndrCo. Their goal: a new startup that trains its own video generation models, purpose-built for professional filmmakers.

The company does not yet have a name. Terms, funding size and lead investors were not disclosed in the initial report. But according to follow-up coverage, the venture is already in talks with potential backers — reportedly including venture capital giant Andreessen Horowitz — on a large funding round. Stephanie Palazzolo and Cory Weinberg, who broke the story at The Information, framed it simply: Hollywood’s most famous traditionalist is now building AI video from the inside.

Why this matters

To understand the significance, you have to understand who is doing this — and what each brings.

Katzenberg is Hollywood royalty. He ran Paramount Pictures in its 1980s heyday, became chairman of Walt Disney Studios and presided over its golden animation era (The Little Mermaid, The Lion King, Beauty and the Beast), then co-founded DreamWorks with Steven Spielberg and David Geffen. Shrek, released in 2001, became the first feature-length animated film to win an Academy Award in its category. When Katzenberg pivoted to investing through WndrCo — where he and Jaswa raised $460 million for seed and venture funds — he remained one of the loudest establishment voices on the collision course between entertainment and technology.

Peebles is one of the most credentialed video-generation researchers alive. A Berkeley PhD (with stints at FAIR and Adobe Research), he co-authored the DiT paper — “Scalable Diffusion Models with Transformers” — which became the architectural backbone of modern latent video diffusion, Sora included. At OpenAI he was Head of Sora, the video generation and world simulation effort that at its peak served nearly 2 million weekly active users, with a stunning 70% of them actively creating content. He left OpenAI on April 17, 2026, in the same leadership shakeup that saw apps chief Kevin Weil exit — the day OpenAI shut the Sora project down entirely as part of its “shedding side-quests” consolidation. The Sora web and app experiences were discontinued on April 26; the Sora API is scheduled to sunset on September 24, 2026.

Jaswa is the operator-investor bridge. As Dropbox’s former CFO he has seen a consumer-technology platform scale to hundreds of millions of users, and at WndrCo he has spent years investing at the intersection of media and enterprise AI.

Put together: academy-level creative judgment, frontier-level model architecture, and scaling experience. That combination does not exist anywhere else in the AI video landscape.

The context: Hollywood’s cold war with AI video

The timing is what makes this a story about an industry, not just a startup.

Hollywood has spent two years fighting generative video tooth and nail — arguing in strikes, lawsuits and lobbying that AI-generated footage infringes copyright and stifles creativity. The flashpoint came in February 2026, when ByteDance’s Seedance drew global attention for instantly producing footage that looked like film scenes featuring real Hollywood actors from a short text prompt. Seedance suspended its planned March release after a series of copyright disputes with major studios and streaming platforms, and eventually launched in global markets excluding the United States.

Against that backdrop, the entry of Katzenberg — the ultimate Hollywood insider — into video generation AI signals how much the industry’s mood has changed. As Seoul Economic Daily put it, that a producer who ran major studios, rather than an engineer who climbed up through ML, is now building generative video reflects “a reality the industry can no longer ignore.”

There is also a commercial logic: with OpenAI having abandoned Sora over its enormous compute burden and strategy shift, and ByteDance locked out of the US market, the position of “trusted AI video partner for Western studios” is effectively vacant. Someone with Katzenberg’s relationships and Peebles’ technical pedigree is the most credible bid yet to fill it.

The challenges ahead

The path is not easy, and the reporting is candid about that:

• Compute costs. Video generation is the most compute-hungry modality in AI. Training competitive models requires enormous capital and infrastructure — precisely the burden that helped kill Sora, despite OpenAI’s resources. A startup, even with a large a16z-led round, will need unusual efficiency (or unusual partnerships) to compete.
• Differentiation. General-purpose video models (Runway, Pika, Kling, Veo, and whatever Meta ships next) are improving monthly. A filmmaker-first model must translate creative intent — consistent characters, controllable camera language, editable scenes — into measurable training objectives. Whether a studio veteran’s product judgment converts into model architecture advantage is the central open question.
• Trust and copyright. The startup’s biggest asset may also be its constraint: to win Hollywood’s data and licenses, it must be demonstrably cleaner on provenance than its rivals. Katzenberg’s insider status makes him uniquely positioned to strike licensing deals — and uniquely exposed if the venture is seen as legitimizing the technology that Hollywood’s guilds fought.

What to watch

The immediate signals to track: the company’s name and official launch, the size and composition of the funding round (whether a16z leads, and whether strategic studio investors participate), any licensing agreements announced with film libraries, and the first technical disclosures from Peebles’ team — particularly whether they iterate on DiT-style architectures or pursue a new paradigm for long-form, controllable generation.

One thing is already clear: the era of Hollywood treating AI video purely as an external threat is over. The next phase will be built by people from inside both worlds at once — and this founding trio is the strongest proof yet that such people exist.