OpenAI reportedly pulls back Sora and other high‑risk consumer features

OpenAI exits consumer video generation, it has been reported

OpenAI has reportedly begun winding down its consumer-facing video efforts, announcing the shutdown of the standalone Sora app, its API, and the ChatGPT‑embedded video feature — a move that signals a broader retreat from high‑risk consumer functionality, including adult‑content tolerant modes. It has been reported that Sora, which quickly became a viral consumer product after launching in September 2025, saw more than a million downloads in its first five days but then fell into a steep engagement decline.

Costs, retention and regulatory risk

Why the reversal? It has been reported that falling user activity and weak retention could not offset massive operating expenses. Chinese media凤凰网 (ifeng) and state broadcaster CCTV Finance (央视财经) cited analysts who calculate extraordinarily high daily costs for running large video models — figures reportedly reaching $15 million per day and roughly $5.4 billion annually — driven by GPU rental, electricity and inference overhead. Sora’s architecture reportedly required rendering some 30 images per generated second, producing a huge compute multiplier and what analysts call a severe revenue‑to‑cost inversion.

Broader implications for AI products and policy

This episode raises wider questions for the industry: can consumer video generation be sustained at scale without new business models or cheaper specialized hardware? It has been reported that companies are also tightening features that invite legal or reputational risk — including adult content — as regulators and platforms push for safer defaults. Geopolitics matters too: export controls and trade policy that constrain access to top‑tier GPUs have amplified costs for cloud‑native models globally, complicating the economics of consumer AI. Chinese outlets additionally noted a platform notice from Dafeng Hao (大风号) underscoring that some content was user‑uploaded, illustrating the moderation and liability challenges platforms now face.