OpenAI Killed Sora. It Was Burning $1M a Day and Disney Found Out an Hour Before the Public.
OpenAI's AI video generator lasted six months before Sam Altman pulled the plug. Internal reporting reveals Sora was costing $1 million daily in compute, user counts had collapsed below 500,000, and Disney — which had committed $1 billion to the partnership — learned of the shutdown less than an hour before the announcement.
OpenAI shut down Sora. The video generation app, released to the public in September 2025, lasted six months before Sam Altman made the call to kill it. Wall Street Journal reporting confirms what the numbers already implied: the product was hemorrhaging compute at roughly $1 million per day while user adoption had cratered from a peak of one million to below 500,000.
That is not a viable unit economics story at any scale.
The shutdown announcement, made March 24, gave users 30 days before service ends. Existing subscribers see a “Service Sunset” banner. The iOS and Android apps are still functional but have been pulled from the App Store and Google Play for new downloads. The compute freed up is being redirected — to what, exactly, OpenAI has not specified publicly.
The Disney detail is the most revealing part of the reporting. Disney had committed approximately $1 billion to a creative partnership with OpenAI built around Sora. According to WSJ sources, Disney executives learned the product was being shut down less than an hour before the public announcement. The deal is now dead. Whatever OpenAI’s internal logic for that particular timeline, informing a billion-dollar partner with 60 minutes’ notice is a significant relationship failure.
The competitive backdrop matters here. While Sora was struggling to hold users, Anthropic’s Claude Code was quietly winning over the software engineers and enterprise customers that drive OpenAI’s revenue. Sora was not losing to a comparable video product — it was losing in a resource allocation fight against a different product category entirely. Keeping Sora running meant competing for compute against models that were actually growing.
This is the first major product shutdown in OpenAI’s commercial history. The company had previously discontinued API endpoints and deprecated model versions, but Sora was a consumer product with branding, marketing, and a dedicated app. Killing it changes how the market reads OpenAI’s strategic discipline.
The harder question is whether video generation as a product category was ever viable at Sora’s price point, or whether it required fundamentally cheaper inference to work as a business. The cost-per-generated-minute problem is not unique to OpenAI — every video AI provider faces it. Runway, Pika, and Kling all operate in the same constraint. But none of them burned compute at the scale OpenAI was running to keep Sora free and fast for consumers.
The compute now liberated from Sora is substantial. Where it goes next will say more about OpenAI’s actual priorities than any product announcement.