OpenAI announced the Sora shutdown on March 24, 2026, via an X post that read simply: “We’re saying goodbye to the Sora app.” The consumer video generation product, launched in late 2024 and expanded with a standalone app in September 2025, is being discontinued in two stages: the web and mobile app go dark on April 26, 2026, with the API following on September 24, 2026.
The financial picture explains the urgency. According to multiple reports, Sora’s inference costs ran approximately $15 million per day at peak usage, while lifetime in-app revenue totaled just $2.1 million. Downloads peaked at 3.33 million in November 2025 and fell 66 percent to 1.13 million by February 2026. A reported $1 billion Disney investment and licensing deal, which would have brought character content from Disney, Star Wars, Pixar, and Marvel properties to the platform, collapsed without money changing hands.
But the shutdown isn’t the whole story. According to Computerworld’s reporting, the Sora research team continues as a unit, redirected from consumer video generation to world simulation for robotics. Bill Peebles, the Sora team lead, described the new mission as “building systems that deeply understand the world by learning to simulate arbitrary environments at high fidelity.” He called the ultimate prize “automating the physical economy.”
That redirect makes technical sense. Sora was always, at its core, a physics simulator. OpenAI’s own technical paper, still live, is titled “Video generation models as world simulators.” The model learned to represent how objects move through space, how light behaves, how physical environments operate. Consumer video was one output of that understanding. Robotics navigation, manufacturing automation, and warehouse logistics are others, and they come with enterprise contracts instead of consumer download numbers.
The freed compute isn’t all going to robotics. Sam Altman has indicated that some of it is being redirected toward “Spud,” OpenAI’s next major model, and the company’s broader AGI roadmap. OpenAI is also doubling its workforce to 8,000 employees by end of 2026, with a focus on enterprise deployment roles.
For the AI video market, Sora’s exit raises a structural question. If the company with arguably the most compute and capital in the industry couldn’t make consumer AI video work at $15 million a day in inference costs, who can? Runway, Pika, and Luma AI are still in the space, but none operate at Sora’s scale. The economics may simply not work for consumer-facing video generation at current model sizes.
What to watch: whether the world-simulation research produces visible robotics demos in Q2 2026, and whether the “Spud” model incorporates spatial reasoning capabilities inherited from Sora’s architecture. The consumer product is dead. The technology underneath it may be more valuable than what it was being used for.