OpenAI Negotiates to Restart Sora Video Model Collaboration
OpenAI is exploring the possibility of reintroducing the Sora video generation model and is in talks with a stealth startup focused on AI video technology in the film production sector.
According to The Information, the startup seeks access to Sora for the sale of software and other services. The company was co-founded by Hollywood producer Jeffrey Katzenberg, former Sora research lead Bill Peebles, and former Dropbox CFO Sujay Jaswa, and its name has not been disclosed. Katzenberg and Jaswa were previously associated with the tech investment firm WndrCo. It is unclear whether Sora model weights will be made available to the startup, but if the deal progresses, they will gain sufficient access for substantial customization.
OpenAI previously paused the Sora model and its associated TikTok-like video application development as part of a plan to reduce high server consumption, reallocating computing resources and talent towards enterprise and coding tools. Meanwhile, competitors like xAI and Google continue to advance their own AI video generation technologies. After its launch in 2024, Sora gained significant attention for its text-to-video capabilities, and subsequent standalone applications and APIs briefly topped download charts. However, due to high computational costs and limited monetization, the company announced in March 2026 the closure of consumer applications, developer APIs, and the termination of a nearly $1 billion collaboration with Disney.
If this potential collaboration moves forward, it could reintroduce Sora-related technology to the market through professional film production tools. Sources indicate that given Peebles' leadership in Sora's development and familiarity with the model's internal mechanisms, collaborating with him is reasonable. OpenAI has previously redirected some of Sora's team efforts towards world simulation research to support robotics development, and this negotiation suggests the company may be reconfiguring video generation capabilities in professional verticals rather than fully restoring consumer products.
The company had previously faced issues with Sora consuming substantial computational resources, affecting power allocation for other teams, and prioritized revenue-generating enterprise tools in preparation for a potential IPO. This engagement with the film AI startup comes months after Sora's closure and reflects OpenAI's reassessment of the commercialization path for video generation technology.
Market Mechanism: OpenAI, as the model holder, is negotiating technology licensing or deep customization collaboration with the film AI startup. The event is driven by the previous closure of Sora due to computational costs and strategic focus, alongside competitive pressure from rivals advancing video generation. Funds and technology are flowing from OpenAI to professional film tool developers, benefiting startups with Hollywood resources and model experience (gaining customized video capabilities), while competitors relying on closed self-developed video models face challenges in re-entering the professional field.
Source: Public Information
ABAB AI Insight
After launching Sora in 2024, OpenAI attempted to position it as a consumer-facing social video platform and reached a nearly $1 billion character licensing and investment framework with Disney by the end of 2025. However, due to the model's inference computational costs far exceeding monetization capabilities, the company fully shut down the application and API in March 2026 and redirected the team towards robotics world simulation research while focusing resources ahead of an IPO.
In terms of capital strategy, OpenAI is now negotiating with a startup co-founded by Katzenberg, former Sora lead Peebles, and Jaswa to direct existing model weights or deep access permissions towards professional film production tools, rather than reinvesting in high-cost consumer applications. The motivation is to leverage the external team's industry resources and familiarity with the model to monetize video generation capabilities in vertical scenarios while avoiding ongoing server consumption burdens.
This is analogous to Google transforming some research models into enterprise products through partners or independent video generation companies like Runway and Luma establishing ecosystems in professional film tools ahead of larger firms. Currently, OpenAI is transitioning from broad consumer expansion to a focus on enterprise and professional verticals, with video generation capabilities shifting from public products to customizable collaborative assets.
Essentially, this represents a shift in pricing power through technological substitution: high-computational video generation models struggle to cover costs on the consumer side, forcing capabilities from closed self-use to scene-based pricing and distribution through professional partners. The mechanism involves the model holder outsourcing inference costs and customization demands to external entities with industry clients, thereby retaining technological influence without increasing its own server burden.
ABAB News · Cognitive Law
High-computational models first fail in monetization, then thrive in vertical collaboration.
Shutting down consumer products does not equal abandoning technology; it merely changes the billing target.
Computational costs determine whether a model is a self-use asset or a licensable tool.