US media reveals inside story of Sora's shutdown: daily losses of millions, user base halved, facing competitor poaching

Earlier this month, after OpenAI CEO Sam Altman arrived in Los Angeles to attend an after-party following the Academy Awards ceremony hosted by Vanity Fair, his company had only a few weeks left before licensing its Sora video generation tool to major Hollywood studios.

After ChatGPT’s runaway success, Sora was being hyped as the next consumer-facing AI frontier technology. The app made it easy for users to put themselves and their friends into all kinds of video scenarios—whether they were dribbling like players on a Harlem-style basketball team or dueling Darth Vader from Star Wars with a lightsaber.

Disney’s then CEO, Bob Iger, also saw potential in it, agreed for the company to invest $1 billion in OpenAI, and allowed characters from the studio’s Marvel, Pixar Animation, and other properties to appear in videos generated by Sora. Just as important, at a time when the industry broadly worried that AI would threaten creative work, he stamped this emerging technology with Disney’s valuable seal of authority.

Compute burden

However, OpenAI suddenly decided to shut down Sora.

Disney executives were shocked by this. Many of them only learned about it less than an hour before the decision was set to be announced. What they didn’t know was that, in the months after launch, Sora had quietly become a major drain on OpenAI—especially as the startup scrambled to adjust its business focus ahead of its initial public offering (IPO).

At the time, OpenAI was only a few weeks away from completing a new AI model codenamed “Spud,” and needed to free up more computing resources to support the coding tools and enterprise products running on that model. AI chips were the most valuable resource for any top research lab, and at OpenAI, Sora was consuming too many of them.

And Sora wasn’t profitable yet—every user who “stitched” themselves into World War II newsreel footage or Hollywood chase scenes was consuming limited compute resources.

Strategic miscalculation

Today, Sora looks like an expensive strategic misstep, and the key employees driving the project are at the center of the AI talent battle sweeping across Silicon Valley.

Altman described the decision to shut down Sora as a difficult but necessary sacrifice the company made to achieve bigger goals. In a memo to employees, he wrote that seeing everyone willing to make “hard trade-offs” for the company’s interests deeply encouraged him.

Altman

The decision marked a stunning end to a project Altman had high hopes for. He had dreamed of using Sora to turn OpenAI into a creative vanguard for the AI era and bring in a substantial new revenue stream for the company.

Two years ago, when OpenAI first showcased Sora to the world, it presented dreamlike scenes generated by the technology—like the fantastical world of Hayao Miyazaki, or the surrealistic paintings of the renowned artist Salvador Dalí. In September of last year, when OpenAI launched a standalone Sora app for consumers, Altman compared it to the moment the company had first released ChatGPT.

But the app never became as popular as its developers had envisioned. It was more like low-quality content churned out by AI than awe-inspiring creative work. By the end of last year, its usage had leveled off.

Shortly after the app launched, the number of global users briefly peaked at around 1 million, but it never returned to that level. According to data from AI research firm Similarweb, in the following months, the user count fell to below 500,000.

A person familiar with the matter said that Sora lost about $1 million per day. As OpenAI tightened its cash flow ahead of its IPO, executives began scrutinizing Sora with even stricter eyes, and the results did not satisfy them.

At the time, OpenAI’s research team was about to begin training a new model designed to support ChatGPT’s video generation feature. Unlike language models that learn from text, a video model needs to understand the entire dynamic world, so its creation costs much more. After carefully accounting for the required costs, OpenAI decided to cancel Sora.

OpenAI planned to shift its focus to a new “super app” it was building—one that would integrate so-called AI agent tools that could carry out tasks autonomously for users, such as writing software, analyzing data, and booking travel. These productivity-focused products are increasingly being adopted across the labor market, and OpenAI currently lags behind competitor Anthropic in this race, threatening its leading position in the AI competition.

Altman told employees that the Sora team would next focus on longer-term strategic efforts such as robotics.

An OpenAI spokesperson said the company is assigning strict priorities to computing resources based on areas that can deliver the greatest long-term economic value. She said, “This prudent focus on where to put compute enables us to achieve growth, speed up innovation, and serve enterprises and developers more efficiently.”

Zuckerberg goes after talent

Sora is the result of the creativity of Tim Brooks and Bill Peebles. While pursuing their PhDs at the University of California, Berkeley, the two researchers became close friends. In early 2023, they joined OpenAI with the goal of building models that could simulate the physical world by generating high-quality video from text.

Sora falls under OpenAI’s world simulation team, led by Aditya Ramesh. This division operates independently of OpenAI’s core research team, which is responsible for building the large language models that power ChatGPT.

Last spring, Meta CEO Mark Zuckerberg launched a comprehensive talent battle against OpenAI. He personally contacted dozens of OpenAI’s top researchers, luring them to join his new AI lab with big pay packages. One of his targets was Peebles, who received an invitation and briefly considered joining Meta.

According to a person familiar with the matter, OpenAI managed to keep Peebles on board by giving him a raise. Shortly thereafter, his responsibilities on the Sora project expanded as well. Peebles was tasked with overseeing the training of the updated video generation model and the development of Sora’s consumer-facing application.

Disney’s AI dream

Despite Sora’s $1 million daily losses, OpenAI tried to find a way to make Sora work. In December last year, it announced a multi-year deal with Disney to license more than 200 characters from the entertainment giant’s film library. As part of the agreement, Disney agreed to become OpenAI’s primary customer and invested $1 billion in it.

Iger, in an interview with CNBC, said the deal gave Disney a chance to take part in the rapid development of AI and new forms of media entertainment. Altman said he hoped the collaboration would offer users a new way to express creativity with the help of AI.

Disney’s then CEO, Iger

For Disney, the deal demonstrated that there was a viable business model for AI licensing built around its intellectual property. The day before it announced its agreement with OpenAI, Disney had just sent a cease-and-desist notice to Google, accusing the tech giant of “massively infringing Disney’s copyrights.”

Cloud service solutions

In February this year, during a quarterly earnings call, Iger said short videos generated with Sora would soon appear on the streaming platform Disney+, which was preparing to launch a vertical video information feed at the time. According to a person familiar with the matter, Disney was also in talks with OpenAI, planning to roll out ChatGPT across the company.

In recent weeks, OpenAI began piloting an enterprise version of Sora, allowing companies like Disney to use the tool with safety and security safeguards. Disney had expected to launch the tool as early as this spring. It would enable some Disney executives to use Sora across various stages—from designing marketing campaigns to visual-effects production—without OpenAI accessing their work content.

But OpenAI was already considering how to back out of the Sora project. The gap between the company and rival Anthropic widened further, as Anthropic’s recent technical advances once again reignited concerns that AI might replace traditional software and services, briefly triggering a sharp drop in software stocks. OpenAI realized it needed to put more resources into building these so-called productivity tools and began downgrading priorities in certain areas.

After initially planning to continue offering video generation features through ChatGPT, the company ultimately decided to fully shut down Sora.

Disney’s $1 billion investment in OpenAI ultimately didn’t pan out, and the relationship between the two parties has effectively stalled.

According to a person familiar with the matter, under the leadership of the newly appointed CEO Josh D’Amaro, Disney is actively discussing with more than a dozen partners how to apply other AI tools.

Real-time information push

In a statement, Disney said, “As the emerging AI space rapidly develops, we respect OpenAI’s decision to exit the video generation business and readjust its priorities. We appreciate the constructive collaboration between both teams and the experience we gained from it.”

On Sora’s X account, its team posted something like a digital obituary: “To everyone who created with Sora, shared with Sora, and built a community around it: Thank you.”