OpenAI, the Microsoft-backed artificial intelligence developer, is reportedly in talks to license its shelved video generation model, Sora, to a newly formed stealth startup. According to a person with knowledge of the discussions cited by The Information, the unverified startup aims to access and modify Sora to sell specialized software and services directly to filmmakers. The venture was co-founded in recent months by Hollywood executive Jeffrey Katzenberg, former Sora research lead Bill Peebles, and Sujay Jaswa, the former chief financial officer of cloud storage pioneer Dropbox.

The potential partnership marks a significant shift in how OpenAI approaches the commercialization of generative video. Earlier this year, the company halted development on Sora and an associated TikTok-style consumer application, reportedly as part of a broader effort to cull server-intensive side projects. While competitors such as Google and Elon Musk’s xAI continued to advance their own video models, OpenAI appeared to step back from the compute-heavy consumer video space. These early-stage talks suggest the technology may instead find its route to market through specialized enterprise licensing.

Offloading the capital expenditure of generative video

The structural logic behind the reported talks highlights the immense infrastructure costs associated with video generation. Training and running models like Sora requires vast amounts of computing power, a reality that forced OpenAI to prioritize its core text and reasoning models over consumer video applications. By licensing the underlying technology to a third party, OpenAI could effectively offload the capital expenditure and distribution risks associated with bringing a high-fidelity video product to market.

For the stealth startup, the arrangement offers immediate access to a foundation model that previously captured widespread industry attention. The combination of Peebles’ technical familiarity with the Sora architecture and Jaswa’s experience scaling enterprise software at Dropbox suggests a focus on building professional-grade workflow tools rather than consumer novelties. If the deal materializes, the startup would take on the burden of refining the model for specific commercial use cases, allowing OpenAI to capture value from its initial research without bearing the ongoing operational costs of a specialized product.

Hollywood's integration of foundation models

The involvement of Katzenberg points to a highly targeted go-to-market strategy focused on the film and television industry. Hollywood has historically maintained a cautious stance toward generative artificial intelligence, balancing the promise of production efficiencies against concerns over copyright and labor displacement. A specialized software provider, backed by an established industry figure, may be better positioned to navigate these institutional sensitivities than a broad consumer technology company.

This specialized approach contrasts with the broader strategies of OpenAI’s primary competitors. Google, the search and cloud infrastructure giant, continues to leverage its massive proprietary compute resources to develop in-house video capabilities, while xAI is similarly pushing forward with integrated models. The broader technology sector remains highly optimistic about artificial intelligence's trajectory—with TikTok's chief executive recently projecting major scientific breakthroughs driven by AI within the next three years—but the immediate commercialization of video generation requires navigating steep technical and financial barriers. The reported talks indicate that strategic partnerships may become a preferred vehicle for deploying resource-intensive models.

The final structure of any agreement between OpenAI and the Katzenberg-backed startup remains unverified, and the talks could still evolve or collapse. However, the negotiations illustrate a maturing ecosystem where foundation model developers increasingly rely on domain-specific partners to commercialize their most demanding technologies. The trajectory of these discussions will offer a clearer view of how the next generation of generative video reaches professional markets.

With reporting from The Information, CNBC Technology

Source · The Information