Disney Pledged $1 Billion for Sora. They Learned It Was Being Shut Down Less Than an Hour Before Everyone Else.

The number that ended Sora is not what you think.

Every headline cited either "$1 million per day" (WSJ) or "$15 million per day at peak" (Cantor Fitzgerald via Forbes). Both figures were analyst estimates. Neither was confirmed by OpenAI. The actual daily operating cost of Sora has never been publicly disclosed. Coverage chose a number and treated it as fact; there is a fifteen-fold gap between the two estimates, and that gap is not a rounding error — it's the difference between a costly failure and a catastrophic one.

What's confirmed: Sora generated $2.1 million in total lifetime revenue, measured by Appfigures tracking in-app purchase receipts. The app launched in November 2025, peaked at 3.3 million downloads, and had fewer than 500,000 active users by early 2026. OpenAI announced the shutdown on March 24 — app closing April 26, API retiring September 24. At the conservative WSJ estimate of $1 million per day across roughly 16 months of operation, total operating costs would be approximately $480 million against $2.1 million in in-app revenue. That's an extraordinary ratio even before the estimate uncertainty.

Now to the part that received almost no coverage.

Disney had pledged a $1 billion equity investment in OpenAI. The investment came with a companion 3-year licensing agreement covering more than 200 characters from Disney, Marvel, Pixar, and Star Wars. It was the first major entertainment IP licensing deal in OpenAI's history — the signal that the industry was willing to bet on AI video as a commercial platform. The agreement was signed in December 2025.

No money ever changed hands.

On March 24, according to sources familiar with the matter confirmed to CNN and Variety, Disney was notified of the shutdown less than an hour before the public announcement. A $1 billion strategic partner. Less than an hour's notice.

There are two explanations for this, and both are worth sitting with. The first: the Sora shutdown was an emergency decision made in the final hours before announcement — something broke (financially, technically, reputationally) that forced the decision quickly enough that standard partner notification protocols weren't followed. If true, what broke? OpenAI has not said. The second: the notification was deliberately minimal to prevent Disney from attempting to block or delay the announcement. If true, OpenAI made a calculated decision to surprise a billion-dollar partner rather than risk negotiation. The Ropes & Gray legal analysis published after the shutdown is direct: "a cautionary tale about deal structure when strategic partnerships are dependent on specific product roadmaps" — which is lawyer language for "Disney had no closing conditions protecting them from exactly this outcome."

The unit economics failure is also structurally worth examining. The core assumption behind Sora was that ChatGPT's success would transfer to video — that users who had built habits around text generation would build similar habits around video generation. They didn't. Downloads peaked at launch, active users declined sharply, and the revenue curve never developed. The product-market fit problem is not mysterious in retrospect: generating a good video takes more effort than generating a good text response, has fewer everyday use cases for most people, and the output is harder to iterate on. Creative professionals — the natural power users — have high expectations and specific workflows that Sora's consumer-facing product didn't accommodate. Casual users tried it and didn't return.

The AI video + entertainment IP combination is now 0-for-2 in 2026. ByteDance shelved Seedance 2.0's international rollout "indefinitely following copyright complaints from Hollywood studios." The structural problem is IP licensing at the enterprise level: studios won't license characters for AI training, user-generated content with IP creates liability, and the volume of professional video use cases hasn't materialized at the scale needed to cover the compute costs.

OpenAI's stated plan for the Sora research team is "world models, with the long-term goal of automating the physical economy." The compute reallocation is more immediate: enterprise coding tools. The "automating the physical economy" framing appears in OpenAI's official communications; the robotics interpretation circulating in coverage is an extrapolation. OpenAI has given no timeline, no team size, and no product milestones for this direction. It may become something real. At the moment of the shutdown announcement, it was a face-saving description of a research team that no longer has a product.

Kling 3.0 Turbo, Kuaishou's video model, was leading the video AI ELO leaderboard at 1243 when Sora shut down. Runway, which had been positioning its enterprise video tools against Sora's consumer focus, now gains the most prominent US-lab competitor's exit. Sora API customers have until September 24 to migrate.

The IPO consequences are not straightforward. The Wall Street consensus, per CBC analysis, is that shutting down a money-losing product before an IPO signals financial discipline — the kind of decision public market investors want to see. The counterargument: Sora was supposed to be evidence that OpenAI could build consumer products beyond ChatGPT. Its failure removes the main data point for that thesis from the S-1 narrative, and the Disney collapse will require a risk factor disclosure.

I don't know what happened in the hours before March 24. I don't know whether the Sora shutdown was a planned wind-down that accelerated or a genuine emergency. OpenAI hasn't said, Disney hasn't said, and the coverage hasn't asked with much persistence. What I know is that the operating cost numbers everyone cited were estimates, neither confirmed, with a fifteen-fold gap between them — and that a $1 billion strategic partner received less than an hour's notice of a decision that eliminated the product they'd bet on.

Those two facts are the story the coverage missed.