From the spotlight to a hasty end: Why did Sora become OpenAI's "discarded pawn"?

OpenAI once positioned its video generation tool, Sora, as the most important consumer-grade product after ChatGPT. Now, less than six months after opening it to the public, the company has abruptly halted it. This outcome may suggest that in the AI race where compute is scarce and competition is intensifying, products that keep burning money without being able to monetize have no room to survive.

The immediate trigger for Sora’s shutdown is a grim set of business numbers—global active users fell sharply from a peak of about one million to fewer than 500,000, with daily operating losses of roughly one million dollars. At the same time, OpenAI is scrambling to allocate compute for a new model codenamed “Spud” in order to support the development of programming and enterprise service products, while Sora is consuming too much of this most scarce resource.

This decision dealt a direct blow to OpenAI’s core partners. Disney previously signed a multi-year licensing agreement with OpenAI, committing to invest $1 billion and to make more than 200 IP characters from its roster available for use in Sora. Many Disney executives learned of the shutdown decision less than an hour before the news was made public; the $1 billion investment has yet to be realized, and the relationship between the two sides has effectively stalled.

OpenAI said it will concentrate the freed-up compute resources on productivity tools to catch up with its competition—Anthropic—which is gaining the lead in the enterprise market. This shift indicates that at the crucial stage leading up to the IPO, monetization capability has become the top criterion for how the company allocates resources.

A flash in the pan: growth challenges behind the hype

Sora was born from the academic ideals of two PhD graduates from the University of California, Berkeley. Tim Brooks and Bill Peebles joined OpenAI in early 2023, aiming to build an AI model that could generate high-quality videos from text and thereby simulate the physical world. In February 2024, they named the system after the Japanese word for “sky,” and it made its first public appearance. Sora shocked the industry with simulated videos such as a mammoth crossing a snowy plain and a fashionable woman strolling through the neon streets of Tokyo. Sam Altman then invited users on the X platform to submit text prompts, jointly showcasing its generation capabilities.

That December, OpenAI officially rolled out Sora’s consumer application to the general public. Within a week of launch, the app quickly climbed to the top of the App Store. Users only needed to enter a prompt to get a ten-second video within minutes. With support for uploading their own faces, users could star in all kinds of bizarre scenarios, and Altman himself also contributed his likeness, sparking a wave of users eager to try it.

However, the hype faded just as quickly. After global user numbers hit a peak of around one million, they continued to decline. Over the following months, the figure shrank to fewer than 500,000. According to Similarweb, a data analytics firm, usage had already begun to stagnate before the end of the year. Altman compared Sora’s launch to the historic moment of ChatGPT’s first release, but the app ultimately failed to realize creators’ ambitions. The Wall Street Journal described its actual performance as “more like AI sludge than AI magic.”

Compute black hole: the price of a million dollars a day

Sora’s high operating costs stem from the technical characteristics of the video generation model itself. Unlike language models that learn from text, video models need to understand and reconstruct complete dynamic scenes, making training and inference costs far higher than those of text-based products. Every user who embeds their own face into World War II newsreel scenes or Hollywood car-chase scenarios is consuming a limited share of AI compute. According to media reports citing people familiar with the matter, Sora’s daily operating loss is roughly one million dollars.

Inside OpenAI, there are dashboards that track how compute is allocated across teams. Some employees were surprised by the share of compute that the Sora team received—since the video generation tool neither brings in significant revenue nor helps improve language model capabilities. This resource allocation has triggered internal questions.

External competitive pressure has made the issue even more urgent. On the consumer side, Google’s Gemini won broad user adoption; Anthropic’s code tool Claude Code, thanks to its highly autonomous programming capabilities, quickly captured the mindshare of Silicon Valley software engineers, leaving OpenAI off guard. OpenAI rushed out its own programming product, Codex, in a new version, but still struggled to close the gap.

The new model codenamed “Spud” urgently needs more compute. The company also planned to train a separate new model specifically for video generation features within ChatGPT. After cost estimates, OpenAI ultimately decided to cancel the training plan and shut down Sora entirely.

Codename “Spud” new model urgently needs more compute, and the company also plans to train a new model separately for video generation features within ChatGPT. After estimating costs, OpenAI ultimately decided to cancel that training plan and shut down Sora completely.

Meta recruits talent and creates internal isolation

Sora was once at risk of being ended early in a talent battle. In the spring of 2025, Meta CEO Mark Zuckerberg launched a large-scale effort to poach talent from OpenAI, directly contacting dozens of top researchers to lure them into his own AI lab with high compensation. Sora co-founder Bill Peebles was also on the list, and he seriously considered the offer. According to media reports citing people familiar with the matter, OpenAI ultimately retained Peebles by raising his pay, and afterward expanded his responsibilities on the Sora project—where he led the training of the next generation of video generation models and the development of a consumer application.

However, within the company, Sora has always operated in a highly closed manner. The project sits under the world simulation team led by Aditya Ramesh, and it is independent from the core research team responsible for the ChatGPT language model. Its progress is kept highly confidential from other departments. Some former employees describe Sora as a “company within a company.” To some extent, this isolated operating model has made it harder for Sora’s strategic value to gain broader internal recognition, and it has left the project relatively vulnerable in the company’s competition for resources.

Disney’s dreams shattered: $1 billion investment turns to nothing

Sora’s most representative partnership ended in the same sudden way.

In December 2024, OpenAI and Disney announced that they had reached a multi-year licensing agreement covering more than 200 well-known IP characters, including Marvel and Pixar; Disney also promised to invest $1 billion in OpenAI and become its primary enterprise customer. In an interview with CNBC, Disney CEO Bob Iger said the partnership would allow Disney to ride the rapid growth wave of AI and help shape new forms of media entertainment. Notably, the day before Disney announced the deal with OpenAI, it had just sent a cease-and-desist letter alleging copyright infringement to Google—an action widely seen as an important signal that the company was endorsing the licensing-based business model for AI.

In February of this year, during an earnings call, Iger revealed that short videos made using Sora would soon appear in Disney+’s vertical video information feed. The two sides were also in talks about rolling out ChatGPT across Disney throughout the company.

However, before OpenAI announced its shutdown of Sora, many Disney executives learned about the news with less than an hour to spare. The $1 billion investment has yet to be delivered, and the collaboration between the two sides has effectively come to a standstill. Disney later issued a statement saying: “As AI evolves rapidly, we respect OpenAI’s decision to exit the video generation business and refocus in other directions. We thank both teams for the productive collaboration between them.” Under the leadership of the newly appointed CEO Josh D’Amaro, Disney is currently actively discussing plans with more than ten partners to introduce other AI tools.

Betting on a “super app”: practical logic fully takes the lead

Sora’s shutdown is not simply a product failure; it is also a snapshot of OpenAI’s strategy narrowing even faster ahead of the IPO.

In an internal letter to employees, Altman characterized the decision as “a difficult but necessary sacrifice for the company’s overall goals,” and said he is glad that employees were willing to make “hard trade-offs.” An OpenAI female spokesperson said the company is making a “ruthless prioritization” of compute resources based on principles of maximizing long-term economic value, adding: “This focus will allow us to grow and innovate faster, and serve enterprises and developers more efficiently.”

The company’s current focus is shifting to a planned “super app” that will integrate so-called “agents” AI tools, enabling it to autonomously execute tasks for users such as writing code, analyzing data, and booking itineraries. Such productivity-oriented products are spreading quickly among enterprises and developer communities, but OpenAI has lagged behind Anthropic in this market so far. Altman said the original Sora team’s focus will shift to directions with more long-term potential, such as robotics technology.

For OpenAI, Sora’s journey was an expensive strategic misjudgment. A product that Altman once hoped would reshape popular culture and open up a new source of revenue ultimately floundered due to a missing business model and improper resource allocation. At the critical stage when the company is focusing on profitability, it became a burden that had to be cut.

Risk notice and disclaimer

        The market involves risks; investing requires caution. This article does not constitute personal investment advice, and it does not consider any individual user’s specific investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article align with their specific circumstances. Invest at your own risk.