OpenAI pulled the plug on Sora this week, quietly shutting down the video generation app and its API. At the same time, the much-publicised licensing deal with Disney, which had been held up as proof that AI was going to reshape creative industries, is also being wound down. For a product that was front and centre in OpenAI’s public-facing ambitions just months ago, it’s a pretty swift exit. On the surface it looks like a straightforward business decision.
Video generation is eye-wateringly expensive to run, the consumer use case never really materialised, and OpenAI is heading towards an IPO with investors who want to see revenue, not just impressive demos. Fair enough. But I think there’s something more significant going on here, and it’s worth paying attention to.
The economics never made sense
Generating video with AI is not cheap. Reports at the time of Sora’s launch suggested it could cost OpenAI somewhere in the region of tens of millions of dollars per day to operate at scale. That is an extraordinary burn rate for a product where, honestly, most people I know tried it once and never went back. The Disney deal added a veneer of legitimacy, but a licensing arrangement with a single studio was never going to cover that kind of overhead.
The business model problem was always obvious. You can charge users per token for text generation and it feels roughly proportionate to what they’re getting. Charging the same way for video, where a 10-second clip costs a significant multiple of a written paragraph, is a much harder sell. And the B2B angle, selling video generation to businesses, puts you in a crowded market against competitors who have been at it just as long.
When the Disney deal goes at the same time as the product itself, that tells you something. It was not a quiet pivot. It was a retreat.
Money is starting to matter
Sora is not the only signal here. Perplexity recently removed the five euro API credit it offered to users, a small change but a telling one. The era of AI companies spending freely to build user bases, generate buzz, and worry about the numbers later is giving way to something more familiar: actual commercial pressure.
This was always going to happen. The early days of any technology wave involve a certain amount of consequence-free experimentation, where the promise of future dominance justifies present losses. But investors get impatient, and IPO timelines focus minds quickly. OpenAI shifting its attention to business productivity tools, apparently codename Spud internally, is the company acknowledging that it needs to build things people will consistently pay for rather than things that generate headlines.
AI slop fatigue is real
There is another thread running through this that I think gets overlooked. Sora struggled not just because of costs, but because the appetite for AI-generated video among ordinary users turned out to be fairly limited once the novelty wore off. People tried it, found it impressive in a technical sense, and then largely went back to watching content made by humans.
That is not a coincidence. There is a growing weariness with AI-generated content across the board. The volume of low-effort, algorithmically produced text, images, and video that has flooded the internet over the past couple of years has made people more discerning, not less. AI slop is a real phenomenon, and users are getting better at spotting it and switching off.
Sora was, in many ways, a product built for a version of the market that did not quite exist. The assumption was that people would want to generate their own video content at scale. In practice, most people just wanted to watch things that felt worth watching.
Is the bubble starting to show cracks?
I want to be careful here, because “AI bubble” has been called prematurely more than once, and the underlying technology is genuinely transformative in plenty of areas. But what Sora’s shutdown represents, to me, is the first clear evidence that AI companies are not operating outside the normal rules of business. Products can fail. Deals can fall apart. Investors can lose patience.
That is not a disaster. It is actually healthy. An industry that has to make its economics work, rather than simply impressing people with what is technically possible, will produce more durable and useful products in the long run. The question worth asking is how many other flagship AI products are quietly in the same position, sustained more by narrative than by genuine demand.
Sora being dead is not the end of AI video generation. But it is a sign that the free-spending, consequence-free phase of this wave is winding down, and the harder work of building things that actually earn their keep is beginning.
