When Sam Altman arrived in Los Angeles to attend Vanity Fair's Oscar afterparty earlier this month, his company was just weeks away from being able to license its Sora video-generation tools to Hollywood studios.
After the smash success of ChatGPT, Sora was hyped as AI's next consumer-friendly frontier -- a simple app that allowed users to put themselves and their friends in whatever video settings they choose, from dribbling like the Harlem Globetrotters to clashing lightsabers with Darth Vader.
Disney's Bob Iger signed on to the vision, agreeing to have his company invest $1 billion in OpenAI and allowing the studio's Marvel, Pixar and other characters to appear in Sora videos. Just as importantly, he put Disney's valuable imprimatur behind the nascent technology amid widespread fears about protecting the industry's creative work from AI.
Then OpenAI abruptly decided to shut Sora down.
Disney executives, many of whom learned about the decision less than an hour before it was announced, were shocked. What they didn't know was that Sora had quietly turned into a liability for OpenAI in the months after its release, particularly as the startup tightened its focus ahead of a looming IPO.
OpenAI was weeks away from finishing work on a new AI model, code-named Spud, and needed to free up more computing resources to power the coding and enterprise products that would run on it. AI chips are the most precious commodity at any leading research lab, and at OpenAI, Sora was eating up far too many of them.
The product wasn't profitable, and every user who spliced themselves into a World War II newsreel or Hollywood chase scene drew down a finite resource.
Sora now looks like an expensive strategic miscalculation, one that was led by key employees who were at the center of the AI talent war raging across Silicon Valley.
Altman represented the move as a difficult but necessary sacrifice toward the company's larger goals, writing in a note to employees that he was encouraged by how they were willing to make "difficult trade-offs" for the good of the company.
The decision to kill Sora marked a stunning end to a project that Altman once dreamed would turn OpenAI into the creative pioneer of the AI era and could have been a lucrative new source of revenue.
The company first previewed Sora to the world two years ago, showing dreamlike landscapes conjured up by the technology that invoked the fantastical worlds of Hayao Miyazaki, or perhaps the surrealism of Salvador Dalí. When OpenAI launched a stand-alone Sora app for consumers last September, Altman likened it to the moment when the company first released ChatGPT.
But the app never took off in the ways its creators imagined -- it was more AI slop than AI magic. Usage flatlined by the end of the year. With OpenAI's purse strings tightening ahead of its IPO, company executives began taking a more critical look at Sora -- and didn't like what they saw.
The research team there was about to begin a training run for a new model meant to power video-generation in ChatGPT. Unlike language models, which learn from text, video models have to make sense of entire moving worlds, making them far more expensive to create. After running the numbers on how much it would cost, OpenAI decided to cancel it.
OpenAI expects to redirect its focus toward a new "superapp" the company is building that incorporates so-called agentic AI tools that can autonomously execute tasks for users, such as writing software, analyzing data and booking travel. Such productivity-focused products are becoming widely adopted in the workforce, and OpenAI has so far lagged behind its startup rival Anthropic in winning this market, imperiling its lead in the AI race.
Altman told staff the Sora team will now focus on longer-term bets such as robotics.
An OpenAI spokeswoman said the company is ruthlessly giving priority to its computing resources based on where it drives the most long-term economic value. "This disciplined focus on where we apply that compute allows us to grow, innovate faster, and deliver more efficiently to enterprises and developers," she said.
Meta's talent raid
Sora was the brainchild of Tim Brooks and Bill Peebles, two researchers who became close friends while completing doctorate degrees at the University of California, Berkeley. The pair joined OpenAI in early 2023, and set their sights on building models that could simulate the physical world by creating high-quality video from text.
They previewed their research to the public in February 2024, naming the system they developed after the Japanese word for sky. Sora transfixed the tech world by creating seemingly realistic videos of everything from woolly mammoths trekking across a snowy field to a stylish woman walking down a Tokyo street filled with glowing neon signs. Altman asked users on X to submit text descriptions for Sora before sharing their creations.
That December, OpenAI released Sora to the broader public.
The company housed Sora under its world simulation team, led by Aditya Ramesh. The division worked separately from OpenAI's core research team, which built the language models that powered ChatGPT.
Last spring, Meta CEO Mark Zuckerberg launched a full-scale talent raid of OpenAI, personally reaching out to dozens of the company's top researchers and luring them to join his new AI lab with giant pay packages. One of his targets was Peebles, who received an offer and briefly considered joining the company.
OpenAI managed to keep Peebles after giving him a raise, according to people familiar with the matter. Soon after, his responsibilities with Sora also expanded. Peebles oversaw training of a new video-generation model, and the creation of the Sora consumer app.
OpenAI's researchers are able to track how AI chips are allocated between different groups through an internal dashboard. Some of them were surprised by the amount of computing resources the company gave to the Sora team, given that video-generation tools didn't make much money, nor improve the capabilities of its language models.
Sora's work was closely guarded from the rest of the company, leading some former employees to describe the project as a startup within a startup.
As the year progressed, there were growing signs that OpenAI was falling behind competitors in key parts of the AI race. Google's Gemini grew popular with consumers. And Anthropic's coding tool, Claude Code, was winning over software engineers across Silicon Valley thanks to its ability to write software programs without much oversight. OpenAI scrambled to release a new version of its own coding product, Codex.
But Altman wanted OpenAI to also be the AI company that used the technology to reshape popular culture and entertainment . In early 2025, he asked former Twitter CEO Parag Agrawal to informally consult a team inside OpenAI that was working on a separate social-media project similar to X, people familiar with the matter said. Altman also was working with then-Disney CEO Iger on a deal to enable the entertainment giant's fans to bring their favorite characters to life with AI through Sora.
OpenAI previewed a version of the new Sora app to its employees in late September before releasing it to the public at the end of that month. The product garnered mixed reviews internally. Some employees felt launching a social-media app built around engagement would hurt the company's brand. Others had concerns about the safety implications of allowing users to create AI-generated videos, even with guardrails.
Sora shot up to the top of the App Store in the week after its launch, despite an invitation-only user base. Users who made it found it to be a marvel: type in anything they wanted -- Homer Simpson doing Riverdance -- and a 10-second video of it would appear in a few minutes. And since the app allowed people to upload their own faces, they were suddenly making short, crazy films starring themselves. Altman himself volunteered his likeness, leading to absurd, sometimes violent or disturbing short films, that he didn't seem to mind.
Since the copyright guardrails were fairly loose, videos soon appeared in the feed that pushed the boundaries of taste. A big featured player was Martin Luther King Jr., who could be instructed to share his dream of anything, from a new season of Fortnite to changing Sora's content-violation policy. King's estate complained about the images, leading the platform to announce that it had removed his likeness.
The worldwide user count peaked at roughly a million soon after the app's launch, but never reached that level again. In the subsequent months, it dwindled to less than 500,000, according to data from Similarweb.
Sora was losing roughly a million dollars a day, according to a person familiar with the matter.
Disney's AI dream
Still, OpenAI tried to find a way to make Sora work. In December, it announced a multiyear deal with Disney to license more than 200 characters from the entertainment giant's cinematic universe. As part of the deal, Disney agreed to become a major customer of OpenAI and invest $1 billion into the startup.
In a CNBC interview, Iger said the deal gave Disney the opportunity to play a part in the fast growth of AI and new forms of media and entertainment. Altman said he hoped the partnership would give users a new way to be creative with AI.
For Disney, the deal represented proof that there was a business model for licensing its intellectual property for the use of AI. The day before it announced its deal with OpenAI, Disney sent a cease-and-desist letter to Google accusing the tech giant of "infringing Disney's copyrights on a massive scale."
In February, Iger said on an earnings call that short-form videos created with Sora would soon appear on the streaming service Disney+, which was preparing to launch a vertical video feed. Disney was also in talks with OpenAI to use ChatGPT across the company, according to people familiar with the matter.
Comments