OpenAI Killed Sora After It Burned $1M a Day and Lost Half Its Users

OpenAI is pulling the plug on Sora, its consumer-facing AI video-generation product, after an internal performance review concluded the app was financially unsustainable and failing to retain users at the rate required to justify its infrastructure cost.

According to reporting from The Decoder, Sora was burning through approximately $1 million per day in compute — a staggering run rate for a product that had attracted significant user interest at launch but saw its active base drop by more than 50% within its first few months of availability. The decision to shut it down represents one of the most notable strategic reversals in OpenAI's history.

The Economics Were Never Viable

Video generation is computationally among the most expensive tasks in the generative AI stack. Producing a short clip at Sora's quality level requires orders of magnitude more GPU time than generating a text response or even a high-resolution image. At launch in late 2024, OpenAI priced Sora as part of its ChatGPT Plus and Pro subscription tiers — which meant the company was effectively subsidising every video generated.

The subsidy model might have been defensible if Sora had achieved the user stickiness and growth trajectory that would eventually allow OpenAI to price video generation on a standalone basis. But retention data reportedly told a different story: users who tried Sora once or twice did not become habitual users. The gap between novelty and utility proved wider than anticipated.

Resources Redirected to Higher-Priority Bets

OpenAI is redirecting the freed compute capacity toward three categories it has identified as more commercially promising: coding, enterprise, and agentic AI. This allocation shift is revealing. All three represent areas where AI generates measurable, repeatable business value — where users return daily, where enterprises pay predictable subscription fees, and where OpenAI can credibly compete against Anthropic, Google, and the growing field of specialised coding models.

The coding market in particular is experiencing explosive enterprise demand. Tools like GitHub Copilot, Cursor, and Anthropic's Claude Code have demonstrated that software developers will pay a significant premium for AI assistance that saves hours per day. The enterprise segment, meanwhile, offers the kind of reliable, high-margin contract revenue that consumer products with unpredictable retention rates cannot match.

A Reality Check for AI Video

Sora's shutdown is likely to prompt a broader reassessment of the AI video market. Several well-funded startups — including Runway, Pika, and Kling — have built their entire businesses around the premise that AI video generation will become a mass-market product. OpenAI's retreat signals that the path from technical capability to commercial sustainability in this category is considerably harder than the hype cycle suggested.

The question now is whether the difficulty is fundamental (video generation is simply too expensive to offer at consumer-friendly price points) or temporal (costs will decline as hardware improves and inference optimisation advances). OpenAI's decision suggests the company believes the timeline to viability is long enough to be a poor use of capital today.

For the AI video ecosystem, Sora's closure is not a death sentence — but it is a serious signal that the sector needs to find paths to unit economics that don't depend on the world's best-capitalised AI lab continuing to subsidise the market.