Sora shut down just six months after launch, and OpenAI will fully cease AI video generation services.

Local time on March 24, OpenAI released a statement, officially announcing that it would shut down its AI video generation platform Sora, which had been launched for only six months. OpenAI said, “We’re going to say goodbye to Sora.” Thanks to every user who created with Sora, shared it, and helped build the related community. We will soon publish more details, including the shutdown schedule for the app and API, as well as a plan for saving users’ works.

据 it is reported, OpenAI CEO Sam Altman informed all employees of this series of changes this Tuesday. In addition to the standalone Sora application for ordinary users, the Sora API service for developers and the video generation feature planned to be embedded in ChatGPT will also be discontinued.

In February 2024, OpenAI released a technical preview of Sora 1.0. With its innovative Diffusion Transformer architecture, the model can generate video clips up to 60 seconds long with fine-grained details. After more than a year of iteration, Sora 2 was officially launched in September 2025 as a standalone social app, adding advanced features such as synchronized audio and video and the insertion of live-action characters. It quickly surged to the top of the free chart on the Apple App Store, and downloads exceeded one million within five days.

Last December, OpenAI reached a three-year cooperation agreement with entertainment giant Disney. Under the agreement, Disney would invest $1 billion in OpenAI and grant authorization for its more than 200 classic IP characters from Disney, Marvel, Pixar, and Star Wars to be used for users to generate video content on the Sora platform.

However, only a few months later, Sora reached its end, and the massive deal with Disney also fell through. A Disney spokesperson said they respected OpenAI’s decision to exit the video generation business and shift its priorities elsewhere.

According to a previously leaked internal memo from OpenAI, the drawbacks of the company’s “spray-and-pray” product strategy were starting to show. In the memo, OpenAI’s application business CEO Fidji Simo said, “We realized that we spread our effort across too many applications and technology stacks, and we must simplify. This kind of fragmentation has been slowing us down and making it difficult for us to meet the quality standards we want.”

Over the past year, OpenAI has rolled out multiple products one after another, including Sora, the AI programming platform Codex, and its own browser Atlas. Each product set up its own development team, which dispersed a large amount of resources, resulting in no single product being able to concentrate enough strength to reach the extreme.

In addition, Sora’s high compute costs and uncertain business prospects are also important reasons for OpenAI’s decision to shut it down. The GPU computing resources consumed to generate a high-quality AI video are far higher than those needed to generate text or static images. Data show that since its launch, Sora’s cumulative consumer total revenue on iOS and Android has been only about $1.4 million, and its best single-month revenue has been no more than $540k—completely disproportionate to the enormous compute consumption behind it.

Industry data show that OpenAI’s revenue in 2025 was $13.1 billion, with a loss of $8 billion. In 2026, the expected loss is projected to skyrocket to $25 billion, and its cash burn rate is as high as 83.3%. Against this backdrop, shutting down Sora, which has high costs and low returns, and focusing on core businesses that can achieve commercial monetization, has become a必然 choice.

In the two years since Sora’s technology debut, the global AI video segment has entered an “age of warring states.” Currently, there are many players in the AI video generation industry. In addition to OpenAI’s Sora, there are overseas companies such as Runway and Pika Labs. In China, there are products such as ByteDance’s Jyming AI, Alibaba’s Tongyi Wansxiang, and Kuaishou’s KeLing AI; meanwhile, domestic products have achieved technological breakthroughs, supporting multimodal input and long video generation, with cost control that is more advantageous. Among them, Seedance 2.0, which was in internal testing at ByteDance, drew global attention based on its “director-level” ability of “text-to-generate multi-camera, movie-grade videos.”

Insiders say that OpenAI’s strategic adjustments will have a profound impact on the entire AI video generation industry. It could trigger a “elimination round” across the industry, leading more companies to abandon unrealistic plans for the consumer end, and instead focus on professional areas or core technology research and development, pushing the industry to leap from “toys” to “tools,” and ultimately to “infrastructure.”

A wealth of information and precise analysis—right on the Sina Finance app

责任编辑：江钰涵