OpenAI Killed Sora Over Poor Unit Economics

OpenAI killed Sora on Tuesday, and the autopsy reads like a textbook case of where the money wasn't. Six months after its September 2025 launch — The Guardian reported a stand-alone app debut — the video generation tool had accumulated $1.4 million in global net in-app revenue over its entire lifetime, according to the BBC — roughly what one training run costs, if that. ChatGPT made $1.9 billion in the same period. The comparison writes itself, and OpenAI appears to have been the one writing it.

"We cannot miss this moment because we are distracted by side quests," Sam Altman told staff, in a quote reported by The Verge. "We really have to nail productivity in general and particularly productivity on the business front. That means backing off projects like Sora." The Sora research team will continue in world simulation work for robotics — useful, inoffensive, and a long way from a consumer product that couldn't keep users.

The download curve tells the story without needing the financial data. Sora started at 4.8 million worldwide installs in October 2025, hit 6.1 million in November, then fell through the floor: 3.2 million in December, 2.1 million in January, 1.4 million in February, and 1.1 million in March — an 81 percent collapse from peak, according to The Verge. The hardware survived. The unit economics did not.

Behind the headline numbers, the dysfunction was specific. Bill Peebles, who led the Sora project, told staff in November that the GPUs were melting under compute load — not an abstract complaint when your flagship demo keeps crashing during live demonstrations. The safety blog post for Sora went live on March 23, according to OpenAI — one day before the shutdown announcement. That timing is either a remarkable coincidence or the world's most expensive legal cover-your-assets exercise.

The Disney relationship crystallizes the chaos. Walt Disney Co was negotiating what Reuters described as a $1 billion equity investment in OpenAI, with Disney licensing more than 200 characters for AI-generated video — plus an option on another $1 billion. The deal never closed. No money changed hands. And on Monday evening, Disney and OpenAI teams were in the same room working on a Sora-related project. Thirty minutes after the meeting ended, Disney was blindsided with the shutdown news. In Hollywood, that's called being fired.

What makes this worth noting beyond the obvious hubris angle is what OpenAI is pivoting toward. Codex — the company's coding agent — crossed $1 billion in annualized revenue in January, according to Wired. That is the product that got the resources. Codex generates revenue. Sora generated compute bills. In a company restructuring its priorities around productivity and business-facing products, the choice was not a difficult one.

The internal signals were visible earlier than the shutdown suggests. Jerry Tworek, a vice president of research at OpenAI, left the company in January after struggling to secure resources for his next big bet — a detail that reads differently in retrospect, when "next big bet" apparently wasn't going to be video. Fidji Simo, previously CEO of Applications, had her title changed to CEO of AGI Deployment — a pivot from running consumer products to overseeing what the company now calls its core mission. And Altman removed direct reporting lines for security and safety teams, moving them out of his chain of command. These are not the actions of a company preparing to ship another consumer demo.

The competitive picture adds texture. Sora launched into a market that had quietly shifted underneath it. Google released a video product. A budget Chinese app — widely understood to be MiniMax or a comparable low-cost entrant — captured the casual user market at a price point Sora couldn't match. OpenAI's $120 billion valuation, underpinned by a recent $10 billion raise, means the company has resources most startups would kill for — and still couldn't make video work at consumer scale. That is a useful data point for anyone who believes compute advantages automatically translate into product dominance.

What the Sora shutdown actually signals is not that generative video is dead. It is that OpenAI — for now, at least — has decided it is not its problem to solve. The team moves to robotics simulation. The capital rotates toward Codex. The consumer ambition that defined the Sora launch in September 2025 has been replaced by something more prosaic and more honest: a company trying to make the numbers work before its IPO, killing products that don't pull their weight, and hoping the story about AGI deployment carries the rest.