OpenAI pulls the plug on Sora

OpenAI just announced that the Sora app and API will be shutting down, a move that, even with some warning signs, still lands abruptly. Basically, Sora was OpenAI's text-to-video model: type a prompt and get a short (and eerily) realistic video clip back. When it debuted in early 2024, it was genuinely unlike anything available publicly at the time, and some said back then that it made Hollywood nervous enough to pay attention. Still, the app went from cool demo to standalone social network to dead product in roughly two years. The official statement was vague: "As we focus and compute demand grows, the Sora research team continues to focus on world simulation research to advance robotics."

The compute angle is probably the honest part. Video generation is notoriously expensive - every rendered second burns through GPU time in ways a text response simply doesn't. Late last year, Bill Peebles - Head of Sora - already announced limits on video generation due to chip shortages, and by shifting away from Sora, OpenAI can reallocate those chips to more (lucrative) coding and reasoning tasks.

But the funniest part is the Disney deal collapse. Just three months ago, Disney had signed a three-year licensing agreement for over 200 characters across Disney, Marvel, Pixar, and Star Wars, with a planned $1 billion investment in OpenAI [Variety]. The deal is now off, and no money ever changed hands, not precisely a quiet wind-down. Honestly, the "world simulation and robotics" framing sounds like a face-saving exit from "we ran out of reasons to keep the lights on" - which is fine really, not every product needs to survive. But if OpenAI couldn't make generative video work at this scale, with their funding and their brand recognition and everything else, maybe that says more about the space's economics than what we think.