OpenAI Shutters Sora, Shifts Business Strategy Ahead of IPO
Back in 2022, OpenAI set off a chain reaction in the tech world when it released ChatGPT, its popular generative AI chatbot. Since then, the company has released several consumer-facing applications like its video generation platform Sora, rolled out to paid users in December 2024. But now something has changed. This week, the company announced it is shutting down Sora.
“We’re saying goodbye to the Sora app. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing,” the company posted on X on Tuesday, adding that it would give more details at a later date about the timeline of the shutdown and how users can save prior work on the platform.
The biggest consequence of the shutdown so far is the loss of OpenAI’s billion-dollar deal with Disney. Announced in December, the three-year deal came with a $1 billion investment in OpenAI and permission to use around 200 Disney characters when generating video with Sora. According to a Reuters report, Disney was “blindsided” by the decision, hearing about it just 30 minutes after a Sora-related meeting between the two companies. The media giant subsequently canceled the deal.
(Novikov Aleksey/Shutterstock)
“As the nascent AI field advances rapidly, we respect OpenAI’s decision to exit the video generation business and to shift its priorities elsewhere,” Disney said in a statement. “We appreciate the constructive collaboration between our teams and what we learned from it, and we will continue to engage with AI platforms to find new ways to meet fans where they are while responsibly embracing new technologies that respect IP and the rights of creators.”
So what happened? Some are speculating that OpenAI is looking to protect its reputation. The loss of the Disney deal suggests a pullback from a category of applications that are technically demanding and legally complex. Media generation systems operate in an environment shaped by copyright concerns, licensing negotiations, and brand sensitivities. By stepping away from a flagship video product, OpenAI may be reducing its exposure to those challenges while redirecting attention to areas where it can move more quickly and with greater control.
Other hypotheses point to the price tag of running Sora. A Forbes report from November estimated the application’s inference costs to be a staggering $15 million per day, or $5.4 billion a year, though the outlet admits its estimate relies on “moving targets” like GPU prices, inference efficiency, user count, and the number of videos created each day. In October, OpenAI Sora chief Bill Peebles said in an X thread that the video platform’s “economics are completely unsustainable.”
Some have theorized that OpenAI’s IPO ambitions may also have something to do with the decision and could explain why the costs are no longer sustainable. The company hired former Instacart CEO Fidji Simo last May to lead its applications business, and according to CNBC, she has been pushing for “product focus and discipline” as the company prepares for its highly anticipated initial IPO that is alleged to be happening sometime this year. The company recently had an all-hands meeting aimed at clarifying its priorities with staff. Simo said OpenAI is “orienting aggressively toward high-productivity use cases,” the CNBC report states.
A still from a promotional Sora video, featuring a tribe of woolly mammoths trekking through the snow (Credit: OpenAI)
When a company prepares for an IPO in the United States, it must file a registration statement with the SEC that includes audited financials and detailed disclosures about its cost structure, risks, and revenue. This process, typically through an S-1 filing, forces a level of transparency that many private companies have not previously faced, particularly around the economics of specific product lines. In OpenAI’s case, moving toward an IPO would likely have brought greater scrutiny to the cost profile of compute-intensive systems like Sora. That kind of exposure can influence internal decisions about which products are sustainable at scale and which may be harder to justify to public market investors.
If Sora is now off the table, what are OpenAI’s new priorities? CNBC says that during the all-hands meeting, Simo told employees the company is focusing on its enterprise business and making ChatGPT users more productive. ChatGPT now supports more than 900 million weekly active users, and OpenAI is seeking to streamline the user experience by combining its Atlas web browser, ChatGPT app, and Codex coding app into a singular desktop “super app.” Simo and OpenAI President Greg Brockman will be directing the effort.
“Our opportunity now is to take those 900 million users and turn them into high-compute users,” Simo said, according to a partial transcript of the all-hands meeting reviewed by CNBC. “We’ll do that by transforming ChatGPT into a productivity tool.”
In addition to the super app, OpenAI is working on an “AI researcher” described as a multi-agent system that can autonomously carry out a full research workflow to solve complex problems. According to MIT Technology Review, the company’s new “North Star” is to build an “autonomous research intern” that can tackle smaller, more specific research problems by September as a “precursor to a fully automated multi-agent research system that the company plans to debut in 2028.” The outlet reports those tasks include math and physics problems like creating new proofs or conjectures, along with life sciences, business, and policy use cases.
(sdecoret/Shutterstock)
Interviewed by MIT Technology Review for the piece, OpenAI Chief Scientist Jakub Pachocki framed the “AI researcher” project as an effort to extend current models into systems that can reason, plan, and iterate over long time horizons, rather than simply generate responses. He described the goal as building agents that can operate across multiple steps of a problem, using tools like code execution and information retrieval while continuously refining their approach. Pachocki said this goal-directed behavior is a necessary step toward more general intelligence, though he also acknowledged that reliability and evaluation of large language models remain an open challenge.
OpenAI’s recent shift toward agentic systems and enterprise-facing tools mirrors a strategy that has worked well for Anthropic, which has focused on productivity use cases and tightly integrated applications rather than consumer experiments. Anthropic’s emphasis on reliability, coding, and workflow support has helped establish its models as practical infrastructure for everyday knowledge work. Additionally, the push toward “AI researchers” is not unique to OpenAI. It reflects the larger industry trend toward reasoning systems that can carry out multi-step scientific and technical tasks with limited supervision. Organizations like Ai2 and FutureHouse have been early builders in this direction, advancing open research with agent-based scientific frameworks aimed at enabling AI systems to understand more about the physical world and support real-world problem solving.
For now, OpenAI’s recent decisions indicate a serious narrowing of focus. The company is stepping back from one of its most visible consumer-facing applications and from a major media partnership, while accelerating its work on agentic systems and enterprise platforms. The result could be a clearer alignment between its research agenda and its commercial strategy. Whether this shift reflects a temporary adjustment or a more lasting change in strategy remains to be seen.
Related
OpenAI Shutters Sora, Shifts Business Strategy Ahead of IPO
Back in 2022, OpenAI set off a chain reaction in the tech world when it…
LANL Develops Diffusion AI Model for Electroplating Process Optimization
March 26, 2026 — Electrochemical deposition, or electroplating, is a common industrial technique that coats materials…
PNNL: AI Effort Moves from Novelty to Front Lines of National Lab’s Cyber Protection
RICHLAND, Wash., March 26, 2026 — A research effort to explore how artificial intelligence can offer…
ORNL Introduces ‘Photon’ Framework for Accelerating AI Vulnerability Discovery on Frontier
March 25, 2026 — Oak Ridge National Laboratory’s Center for Artificial Intelligence Security Research (CAISER) is…
Gartner Forecasts 90% Drop in LLM Inference Costs by 2030
STAMFORD, Conn., March 25, 2026 — By 2030, performing inference on a large language model (LLM)…
Dario Gil at GTC: DOE’s Genesis Mission Moving from Vision to Execution
At Nvidia’s GTC conference last week, a fireside chat between Nvidia’s Ian Buck and U.S. Department of Energy Under…
