OpenAI Kills Sora: $15M/Day Costs, $2.1M in Revenue, and a Collapsed Disney Deal
OpenAI announced on March 24 it is shutting down Sora, its AI video generation app, after the product burned through an estimated $15 million per day while generating just $2.1 million in total lifetime revenue. Disney simultaneously cancelled a planned $1 billion investment.
OpenAI is shutting down Sora. The company announced on March 24 that it would discontinue the Sora app, API, and all video generation capabilities inside ChatGPT. The Sora team posted a brief notice on X with no firm shutdown date and no detailed explanation. The economics tell the real story.
At peak usage, Sora’s inference costs ran approximately $15 million per day. That estimate — based on roughly $1.30 per 10-second clip multiplied by millions of daily generations — gives an annualized burn of $5.4 billion. Lifetime revenue from in-app purchases: $2.1 million. The math required no further analysis.
Sora 2 launched as a standalone app in September 2025 with a TikTok-style social feed and audio generation. It hit the top of the iOS App Store’s Photo and Video category within 24 hours. Downloads peaked at roughly 3.3 million in November 2025, then fell 66% to 1.1 million by February 2026. The audience found Sora interesting; they were not willing to pay for it.
The Disney collapse made the shutdown impossible to hide. In December 2025, Disney announced a $1 billion investment in OpenAI that included licensing 200+ characters across Disney, Marvel, Pixar, and Star Wars for use in Sora. No money changed hands. The announcement unraveled within 90 days. When OpenAI disclosed the shutdown, Disney exited the deal entirely.
OpenAI’s stated reason for the closure is a strategic pivot: the Sora team will redirect toward “world simulation research to advance robotics.” This is not spin — the underlying technology that made Sora possible, training a model to understand physics, motion, and causality from video data, is genuinely more valuable for robotics than for social video generation. The company is choosing the application with defensible long-term leverage over the one that looks impressive in demos.
What this means in practice: OpenAI is exiting the consumer video market and betting that the physics-grounded representations Sora learned will give its robotics efforts a head start. Google’s Veo, Runway, and Kling will absorb the video generation user base. OpenAI seems fine with that.
The timeline for app closure and API shutdown will be communicated separately. Users who created content on Sora have been told the company will provide instructions on preserving their work before access ends.
The broader lesson here is about unit economics at the frontier. Consumer AI applications that require heavy compute per interaction face a structural problem: the generation cost scales with usage, but users will not pay proportionally. Sora charged nothing at the base tier and couldn’t convert enough users to paid plans to offset costs. OpenAI reached the same conclusion every other compute-intensive consumer product eventually reaches — subsidizing usage indefinitely is not a business.
What remains unclear is whether OpenAI’s robotics pivot is a genuine strategic reorientation or a face-saving narrative for killing a product that never worked financially. The two things can be true simultaneously.