Why OpenAI shut down Sora

A prosumer app in a consumer world.

We’ve just experienced our first AI product shutdown and it’s quite a big one. OpenAI just closed up Sora, their TikTok-like app designed to allow for quick and easy video generation for the consumer and prosumer markets.

The app, OpenAI’s first standalone app after ChatGPT, shot to the top of the iPhone App Store soon after its September launch and saw about three million users. That rise brought scrutiny just as fast. Copyright holders questioned how their work was being used. Others pointed to the use of people’s likenesses without clear guardrails. Critics also argued the app was adding to a growing wave of low quality, synthetic content that muddies the line between real and fake. Disney pledged a billion dollars to the service to provide characters and likenesses that users could remix into their own work. That deal, according The Hollywood Reporter, is now off.

What’s OpenAI’s official position?

“As we focus and compute demand grows, the Sora research team continues to focus on world simulation research to advance robotics that will help people solve real-world, physical tasks,” said an OpenAI spokesperson said in a statement. Basically they found out that video was too hard and fraught with potential copyright landmines.

An Artist’s View

Ali Powell runs Animeta AI Media, a studio focused on AI driven video production and experimentation. He has tested most of the major tools in the space, and his view of OpenAI’s Sora is blunt. In his experience, the product has not kept pace with how creators actually work.

“OpenAI has always been a janky company that starts the train while the track is still being built, so it fits their MO,” he said. “I really think it has more to do with their DoD deal. Sora’s main strength is being able to take 30 seconds of video and make clean clones of people you can reuse over and over, not to mention generating people from scratch.”

He also points to a shift in priorities inside OpenAI itself.

“I think it’s because they lost a lot of their users and it was fading as a social platform,” he said.

Powell’s broader argument is that the center of gravity may be moving away from consumer tools and toward institutional buyers. With governments putting real money into AI systems, he suspects companies like OpenAI are starting to optimize for those relationships rather than for everyday creators.

“Given how much governments are investing in AI right now, it wouldn’t surprise me if they’re selling this tech directly at that level,” he said. “If that’s the case, the need to polish a consumer product drops. That’s my conspiracy theory.”

At the same time, he is dismissive of the idea that problems with Sora signal a wider collapse in AI video.

“What’s funny about the Sora situation is you’ve got people cheering like the whole AI video space is falling apart,” he said. “But nobody serious was using Sora to make films. It was mostly people making memes or testing ideas.”

For Powell, the demand side has not changed. If anything, it is getting stronger.

“This isn’t going anywhere,” he said. “Most younger audiences don’t care about the ethics debate. They care about whether something is interesting enough to watch. If the tools can deliver that, people will use them.”

His view reflects a split that is becoming more visible in the space. The tools are uneven and often frustrating, but the direction is clear. Creators are not waiting for perfect systems and instead they are working with what exists, stitching together workflows, and pushing the outputs into real projects. In that context, Sora’s limitations matter less than the broader shift toward faster, cheaper, and more flexible video production.

The Tooling Issue

To get a behind the scenes look I spoke to (add)ventures Executive Vice President Scott Maiocchi. Scott’s role has been to implement AI video into his company’s preproduction systems.

“What I’m trying to do is develop the stack of intelligence workflow and governance,” he said. He’s followed the Sora story with interest.

When Sora first appeared, he said, it landed at the right moment. Image generation had already gone mainstream through tools like Midjourney, and video felt like the next step. Early reactions reflected that sense of anticipation.

“I think when Sora first came out, it was really one of the first ones out there,” he said. “In video content creation, you’re always excited about a new tool because imagery had been out there for quite a while.”

Then there was the familiar Silicon Valley pattern: big claims, unclear mechanics, and a sense that something powerful was happening inside a system that few people fully understood.

“Companies claim it can do a bunch of different things and it’s kind of like this black box of magic,” he said. “So there was a lot of potential, I think. Trying it, I thought it was pretty good. It was kind of interesting, especially if you’re going to do a one-off.”

The problem is that video is not a collection of isolated clips. It is a system of constraints. Lighting, composition, lens choice, continuity, all of it has to line up. That is where tools like Sora still struggle.

“With video especially, it’s very hard to control all the variables as a director,” he said.

“If I’m doing a real piece, I want control over lighting, composition, shot selection, lens selection. Getting a one-off shot was always good with Sora,” he said. “But if you’re trying to make a cohesive piece or build a story, it would probably take more time to get it right than just shooting it.”

That tradeoff pushes the tool into a narrow role. Useful, but not central, which made Sora a novelty and not a real tool. Further, OpenAI couldn’t keep up with competitors.

“There’s new services out there, like Martini, that are more workflow based, which are better,” he said.

What surprised Maiocchi was not the technical limitation, but the apparent pullback. For him, the explanation is less about failure and more about priorities.

“I just think it’s probably a shift in business priorities,” he said. “If they’re moving toward an enterprise model, like Anthropic, there’s a lot more money there.”

Video, by contrast, is expensive. It consumes compute, and it does not yet offer the same return.

“There’s fighting for compute,” he said. “It’s probably bogging down what they could do, and there’s not a lot of return.”

More broadly, he sees an industry that is still unstable at the top.

“With any AI company, things change on a dime,” he said. “What a model did great yesterday, it doesn’t do great now, and someone else does something better.”

That dynamic is shaping how companies position themselves. Some are chasing enterprise contracts. Some are pushing consumer tools. Some are trying to own the entire stack.

Sora wasn’t part of OpenAI’s strategy.

What’s Next for AI Video?

Sora’s shutdown is less an ending than a signal about where the real constraints and incentives in this space sit. Video generation remains expensive, difficult to control, and exposed to legal risk, while the strongest revenue opportunities are increasingly tied to enterprise and government use rather than consumer facing tools.

None of that, however, changes the broader trajectory. Creators are still building with these systems, often stitching together multiple tools to get usable results, and the pace of iteration continues to accelerate even as individual products come and go. What Sora represented was not a finished solution but an early attempt to collapse production timelines and widen access to video creation, and that underlying goal has not gone away.

If anything, its disappearance clarifies what the next phase requires. Tools will need to fit into real production workflows, offer directors meaningful control over variables that define a scene, and connect directly to outcomes that matter for businesses rather than just outputs that look impressive in isolation. Generating a clip is easy to demonstrate, but building something cohesive and watchable remains the harder problem.

In that sense, Sora’s role was transitional. It proved demand, exposed technical and economic limits, and pushed the concept forward. According to the experts I spoke to, the next generation of tools will be judged less by novelty and more by whether they can hold together under the pressure of real use.