OpenAI killed Sora after burning approximately $15 million per day in inference costs while generating only $2.1 million in lifetime revenue, according to reports from The Hollywood Reporter. Disney, which had been in discussions for a $1 billion partnership, was informed the product was dead just 30 minutes after a working meeting concluded. OpenAI's President Fidji Simo reportedly called the video generation tool a "side quest" as the company pivots toward enterprise offerings ahead of a planned Q4 IPO.
The Sora shutdown reveals the brutal economics of AI inference at scale. Video generation requires exponentially more compute than text or image models, and OpenAI's decision suggests they couldn't find a path to profitability despite the technology's impressive demos. This mirrors broader industry struggles with inference costs — even Meta's recent Arm chip partnership explicitly targets "2x performance-per-rack" improvements, signaling that current hardware economics remain unsustainable for many AI applications.
While the original reporting focuses on the Disney deal collapse, the deeper story is OpenAI's strategic retreat from consumer-facing products that don't generate immediate returns. The company is clearly prioritizing enterprise customers who can pay premium prices over flashy demos that drain resources. This shift aligns with their reported IPO timeline, where consistent revenue streams matter more than technological showcases.
For developers building video AI applications, Sora's demise is a stark reminder that impressive capabilities don't guarantee viable products. The inference cost math needs to work from day one — $15M daily burn rates are sustainable for exactly zero startups, and even OpenAI couldn't make it pencil out.