Why OpenAI Shut Down Sora: The Costly Rise and Fall of AI Video Generation

Peak Collapse

In February 2024, OpenAI released the technical preview of Sora, an AI model that generated high‑quality short videos from text prompts, instantly captivating the visual‑AI community. By December 2024, Sora launched publicly and was hailed as a breakthrough that turned the notion of "one‑sentence video creation" into reality. However, on March 24, 2026, OpenAI announced the permanent shutdown of the Sora application, citing a shift in research focus toward world‑simulation technologies.

Money‑Eating Beast

For each 10‑second video, the platform consumed roughly $1.30 of GPU compute. At peak traffic the service handled about 11.3 million video requests per day, driving daily hardware‑operating costs to approximately $15 million and an annual cash burn near $5.4 billion. Pricing was set at $0.10 per second for standard‑definition video and $0.30 per second for premium versions, leaving the business model deeply unprofitable. The app’s first week saw massive downloads (up to 10 million), but user retention fell to about 1 % after 30 days. Meanwhile, OpenAI was preparing for an IPO, promoting a narrative of $200 billion revenue and 800 million weekly active users, while the video‑generation line threatened to become a financial liability.

Forbes‑style analysis also highlighted a rumored $10 billion Disney partnership that would have granted Sora access to over 200 Disney‑related IPs, yet the deal never progressed beyond a signed letter of intent. The collaboration collapsed as OpenAI prepared to retire the product.

Shifting Battlefield

Intense competition from Google’s Gemini, Anthropic’s Claude, and other emerging video models forced OpenAI to abandon its consumer‑grade video business. Anthropic’s B2B focus and strong revenue growth (projected $190 billion annual revenue by early 2026) illustrated a more sustainable path. OpenAI’s own xAI platform, leveraging the massive native social network of X, offered faster, cheaper video generation (an 8‑second 720p clip with audio in about 45 seconds, roughly 30 % faster than rivals) and attracted a niche audience interested in edgy or NSFW content.

Overall, the loss of a first‑mover advantage, the inability to adapt the model to Hollywood‑scale production pipelines, and the unsustainable cost structure in the consumer entertainment market led OpenAI to reallocate resources toward enterprise tools such as office assistants and code‑generation services.