Sora Shutdown: $15M Daily Inference Cost vs $2.1M Lifetime Revenue Tells the Full Story
OpenAI Killed Sora: The Math Was Never Going to Work
OpenAI shut down Sora on March 24, 2026, and while the tech community mourned a promising product, the numbers tell a very different story: Sora was economically unviable from day one.
The Staggering Numbers
- $15 million/day: Estimated daily inference cost at peak (Forbes)
- $2.1 million: Total lifetime in-app purchase revenue (Appfigures)
- -66%: Download decline from November 2025 to February 2026
- $1 billion: Disney deal, now dead
The Economics Problem
Video generation is categorically more compute-intensive than text. Every second of video requires rendering hundreds of frames, each needing spatial reasoning about motion, lighting, physics, and temporal consistency. At peak, OpenAI was spending approximately $15M daily on inference — an annualized $5.4 billion just to keep servers running.
Against that: $2.1M total lifetime revenue. Bill Peebles, OpenAI's own head of Sora, admitted the economics were "completely unsustainable."
The User Collapse
The hype around Sora's App Store launch obscured a rapid user drop-off. Downloads fell 66% in just three months, making the business case impossible. The product never reached the critical mass needed to justify its costs.
The Failed Disney Deal
A reported $1 billion deal with Disney was "always fragile" and ultimately fell through, removing any hope of a revenue lifeline that could have justified the infrastructure investment.
Legal and Reputational Costs
Copyright lawsuits over training data were growing faster than revenue, adding another dimension of cost that had no clear resolution.
What This Means
Sora's failure is not just about one product. It highlights a fundamental challenge in the AI industry: some AI applications, no matter how technically impressive, have cost structures that make consumer pricing impossible at current compute prices. Compare Sora to ChatGPT's 900M weekly active users on a subscription model — the economics of text AI and video AI are in completely different leagues.
The shutdown was not premature. The real question is why it lasted as long as it did.