The Economics Reality Check
OpenAI quietly wound down the Sora public API citing unsustainable inference costs per generated minute. The shutdown forced a recalibration across the video AI sector about which generation workloads are economically viable at scale.
For context: Sora was supposed to be the next era-defining product. It generated stunning videos. But the cost of generating a single minute of 4K video exceeded what any sustainable business model could absorb at consumer scale.
The bigger story: This is the first public admission from a major lab that some generative tasks may never be commercially viable. While everyone talks about GPT-5.4's reasoning or Mistral's infrastructure, the sector just learned a hard lesson: not every use case that can be computed should be computed.
Companies built on the premise that all generative tasks eventually become cheap through scale might need to reconsider their TAM.
