The Sora Shutdown: Why High Inference Costs Killed the OpenAI Video App
In a shock announcement that underscores the brutal economics of generative AI, OpenAI has discontinued its Sora video generation platform. Despite massive hype, the project has been shuttered due to "unsustainable inference costs" that far outweighed its revenue potential.
Bottom Line: Sora was a "halo project" that failed the unit economics test. Burning $15 million daily while earning $2.1 million lifetime is a path to insolvency, even for OpenAI. The compute is being reclaimed for the GPT-6 training run.
The $15M/Day Burn
Estimates suggest a single 60-second Sora clip required 20 minutes of H100 inference time. At scale, this created a compute deficit that OpenAI could no longer justify in the face of competing demands for AGI research.
Newsletter
Get Engineering Deep-Dives in Your Inbox
Weekly breakdowns of architecture, security, and developer tooling. Join engineers who read this before standup.
![IPFS + Filecoin for AI Model Weight Storage [2026]](/assets/thumbnails/post-ipfs-filecoin-ai-model-weight-storage-2026-og.png)
![Embedding Models for Semantic Search [2026 Cheat Sheet]](/assets/thumbnails/post-embedding-models-semantic-search-2026-cheat-sheet-og.png)