1. Not Enough Revenue & High “Compute” Costs
Generating high-quality video is incredibly expensive and requires massive amounts of processing power (GPUs). As against that there was nowhere close to the kind of revenue being generated from the generative video business. The revenue vs resources ratio for generative video is the worst among all the AI verticals. It made sense for OpenAI to decide to use those resources for “reasoning” models (like future versions of ChatGPT) rather than generative video.

2. A Shift to Robotics and “World Sim”
Instead of a video, the Sora team is moving back into pure research. They want to focus on “world simulation” – essentially teaching AI how the physical world works so it can be used to power physical robots that can navigate and perform tasks in real life.

3. Safety and Moderation Struggles
Problems over deepfakes, misinformation, and copyright issues. Despite new safety updates released just days before the shutdown, the “moderation headache” of a public video-sharing platform likely became more trouble than it was worth.

After being on the peak of inflated expectations in 2023, Generative Video is on its way down and signs like these are early indications of Gen video being close to the trough of disillusionment.

In all my classes / presentations, I’ve been putting out the Gartner Hype Cycle specifically for this reason. To help people understand how trends are forming. And this one was visible from a kilometer away.

———————————————-