OpenAI killed Sora because it was burning through $1 million every day and nobody was using it. Then it announced a cinema camera.
The sequence, laid out by the Wall Street Journal and confirmed by TechCrunch, looks like this: Sora peaked at roughly a million users after launch, then collapsed to under 500,000. Meanwhile the compute cost was non-negotiable — video generation is expensive in a way that text generation is not, and the math never worked. Estimates for the daily burn rate diverge: Forbes cited $15 million per day; TechCrunch reported roughly $1 million per day at peak. The gap likely reflects different measurement windows or what counts as attributable costs, but the direction is the same either way: the economics did not work. Sam Altman made the call: shut it down, free the chips, refocus. The WSJ reported that Disney, which had been in discussions to invest as much as $1 billion, found out less than an hour before the public announcement. Nothing had been formalized and no money had changed hands. The deal died with Sora.
That is the business story. The cultural story is Engine Cinema.
At a closed keynote in Cupertino, Altman unveiled what OpenAI is calling Engine Cinema: a large-format cinema camera with a 36mm square sensor, positioned not as a generative AI product but as a physical imaging tool for working cinematographers. The framing from sources present at the event, as reported by Y.M.Cinema, was that generative video had destabilized the very ecosystem it aimed to augment, and that moving toward real-world cinematography was a step back toward craft. Sora did not fail. It forced a reconsideration.
The specs sound impressive on a spec sheet. They sound less impressive when you note that there is no footage yet — no output to evaluate, no production history to point to, no evidence the camera does what the keynote claimed it does. OpenAI has a history of announcing things before they exist. Engine Cinema fits that pattern. The company also said it would release the API in September, which means the current timeline runs through a summer of demonstrations and controlled footage before anyone outside a demonstration environment can evaluate whether the camera performs.
The Sora shutdown was staged. The app closes in April; the API follows in September. That gap is not accidental. It gives OpenAI time to position Engine Cinema as the constructive alternative — the thing that replaces disruption with contribution. The filmmakers who were frightened of Sora are being offered a camera instead. Whether that offer is genuine or tactical is not yet answerable.
The pivot-from-hardware framing is available and tempting. It is also probably wrong as a total account. OpenAI did not abandon generative AI. It freed up compute from a product that was losing money and users simultaneously, and pointed it at something with clearer professional credibility and a community that had already declared war on AI video. Engine Cinema is not OpenAI becoming a camera company. It is OpenAI buying goodwill in an industry it unsettled, on terms that also happen to free up GPU capacity for the products that are actually working.
Whether Engine Cinema produces a real product or another announcement is a question that cannot be answered with a keynote.
Sources: TechCrunch | Y.M.Cinema | The Decoder
