Why is OpenAI shutting down Sora?

OpenAI, the company behind ChatGPT, just announced they’re sunsetting Sora, their generative AI video service that was once lauded as one of the best tools for AI-generated video. It gained recognition after signing — and then retracting — a $1 billion deal to use Disney character likeness in generative AI.

The shutdown is worth paying attention to, not because Sora was particularly successful, but because of what it signals about where the AI industry is headed. Here’s what OpenAI pulling the plug actually means … and are we headed into a bubble?

1. Tech companies don’t want to be media companies

Remember when Facebook was criticized for claiming it wasn’t a media company? Before the pandemic, Facebook was caught in a bind: it was reaping the benefits of media distribution without owning a single media asset. As traditional and online news networks relied heavily on Facebook for reach, the platform faced questions about whether it was pushing certain agendas through algorithms and user behavior manipulation.

OpenAI was walking into the same trap. Moving into generative video means dealing with copyright law, deepfakes, content moderation, and liability. Those are media company problems, not infrastructure company problems. OpenAI’s core business is selling AI infrastructure to enterprises — API access, custom models, compute. Sora pulled focus away from that.

The decision to shut it down suggests OpenAI learned something from Facebook’s experience: if you’re not prepared to handle the regulatory, legal, and ethical baggage that comes with being a media platform, don’t become one.

2. The hyperscaler hype cycle is ending, enterprise focus is winning

A friend once commented that he hates how NVIDIA hoarded everyone’s RAM only to release something like DLSS5. Generative AI has been the controversial elephant in the room, and at this point, we still don’t know if commercial-forward AI companies are actually making money or just burning through investor capital on experiments.

Sora was a consumer-facing experiment. It was flashy, it generated headlines, it looked impressive in demos. But it didn’t fit OpenAI’s revenue model. The company makes money selling API access to businesses, not subscriptions to hobbyists making AI videos. Consumer AI tools require constant moderation, customer support, and infrastructure at scale — all of which eat into margins.

The hyperscalers — OpenAI, Google, Microsoft — are pivoting hard toward enterprise. They’re selling AI infrastructure to Fortune 500 companies, not building consumer apps. Sora’s shutdown is part of that shift. The hype phase is over. The boring, lucrative phase — selling AI as enterprise tooling — is what’s left.

3. AI models as we know them are heavily subsidized

Do you remember when Grab and Uber were extremely cheap? Or when Airbnb was so affordable people said it would kill the hotel industry? Or when Facebook was the darling of the internet, before algorithms and ads made their intrusions?

It’s the same pattern with AI companies, especially those using generative AI for public-facing products. They subsidize usage to gain market share, hoping to lock users in before raising prices or monetizing through other means. But that model only works if you can eventually flip the switch and start charging. Sora never reached that point.

The cost of running generative video models at scale is enormous. Training the model costs millions. Inference — actually generating videos for users — requires massive compute. OpenAI was likely subsidizing every video generated through Sora, hoping usage would justify the investment. It didn’t. And unlike text models, where inference costs have dropped significantly, video generation remains expensive.

This is the reality check for consumer-facing generative AI: if you can’t make the unit economics work, the product doesn’t survive. The tap is turning off. Companies are realizing that subsidizing free or cheap AI tools indefinitely isn’t sustainable, and if there’s no clear path to profitability, the experiment ends.

Welcome to the bubble. It’s going to burst.

Sora’s shutdown won’t be the last. Expect more consumer-facing AI experiments to quietly disappear as companies refocus on enterprise markets where customers actually pay for the service. The AI industry is maturing, and that means fewer flashy demos, fewer moonshot projects, and more focus on boring, profitable infrastructure.

If you’re an AI startup trying to offer a specific service, either try to find ways to make compute cheaper, or get acquired.