OpenAI has quietly shut down Sora, its short-form AI video app that promised to let anyone create viral videos from text prompts, after just six months, the company announced in a social media post. The move closes one of the company’s most visible consumer experiments and forces a sharper question than the one floating around social media. Why did OpenAI really walk away from the app?
Many want to point to OpenAI walking away due to the risks involved in generating video with deepfakes, celebrity misuse and the broader challenges of synthetic media. That theory is not groundless. OpenAI’s own safety materials show the company took misuse seriously from the start. However, the most likely reasons for Sora being discontinued point somewhere else. Reuters reported that OpenAI is discontinuing Sora as part of a broader shift toward coding tools, enterprise customers, robotics and work tied to artificial general intelligence. Reuters also reported that Sora consumed substantial computing resources, leaving less capacity for other teams.
A Short Life for a High Profile App
Sora never had much time to settle into the market. Access to the Sora app began with invite-only iOS users on September 30, 2025, then expanded from there. But only six months later, on March 24, 2026, OpenAI was saying goodbye to the app. In product terms, that is a very short birth-to-death cycle, especially for a tool that arrived with the weight of enormous public expectations.
That short lifespan suggests OpenAI did not merely let the app fade. It made an active decision to stop supporting a product that had recently been presented as a meaningful part of its broader creative AI push. The speed of that reversal makes more sense when viewed through the company’s widening list of priorities. OpenAI is no longer just a chatbot company, if it ever was. It is racing in coding, selling to businesses, investing in infrastructure, talking openly about robotics and still presenting AGI as its long-range mission. In that context, a compute-hungry social video app begins to look less like a pillar and more like a luxury.
The resources that were devoted to Sora came from increasingly strained internal resources. In its Sora 1 sunset FAQ, the company said Sora 1 relied on older models and infrastructure. OpenAI said moving to a single Sora experience would reduce complexity and allow continued improvements in Sora 2 across web and mobile. It added that, in the United States, Sora 1 was removed on March 13, 2026, with Sora 2 becoming the default experience. OpenAI was already simplifying the product stack and decided that a dedicated Sora app ranked below other bets for capital, engineering time, compute and executive focus.
Safety Concerns Are Still Real
None of this wipes away the safety side of the story. OpenAI’s public documentation shows it knew Sora lived in a high-risk category. At launch, the company said all Sora videos would carry visible watermarks and C2PA metadata. It said it maintained tools that could trace videos back to Sora with high accuracy. It said it blocks depictions of public figures and applies extra protections around content involving people’s likenesses. In a separate March 23, 2026 safety post, OpenAI described stricter guardrails for videos involving characters and added protections for sensitive categories of content
Safety considerations were built into the product and into the company’s messaging. OpenAI did not say moderation costs forced the decision, but increasing requirements for human or machine-augmented moderation no doubt would increase operating costs. What is documented is that Sora was expensive to run and sat outside the set of priorities OpenAI now seems most determined to fund.
AI video is expensive in ways text products are not. The compute burden is heavier. Storage costs are high. Moderation is hard and the potential for abuse is much higher than text or code generation outputs. Every one of those layers adds cost, friction, or both. A dazzling demo can win attention in a week, but the realities of sustaining infrastructure and cost-heavy technology is a different matter.