How OpenAI ceded video to Google and China

Within hours of MiniMax claiming the top of the independent leaderboards for video editing—at roughly half the price of anything comparable—ByteDance shipped an upgrade of its own, unwilling to let a rival own the moment alone. The two firms, a Shanghai artificial-intelligence startup and the Chinese owner of TikTok, were fighting over a market the company that created it had abandoned. OpenAI, whose Sora model made photorealistic text-to-video a mainstream idea in 2024, was nowhere in it. It had wound down its video program in the spring.

OpenAI's exit was deliberate and, by the standard of its own balance sheet, defensible. This spring it wound down the Sora consumer app, told staff it was ending the products built on its video models, and let its Disney partnership lapse; the developer interface is scheduled to go dark in September. Sam Altman, OpenAI's chief executive, was preparing the company for a public listing, and a loss-making video app pitched against TikTok did nothing for the story he needed to tell investors. Video sat far from the enterprise and agentic products where the company's revenue actually lives. Cutting it looked like discipline.

Focus pull

The trouble is what discipline overlooks. The video-generation market is still small in absolute terms—a few billion dollars a year at most—but it is compounding at more than 20% a year, and it is one of the few AI applications that reliably turns into paying customers. Advertising teams, e-commerce sellers and games studios do not buy tokens for their own sake; they buy finished clips, and they buy them again next quarter. That is why MiniMax and ByteDance are fighting over the editing workflow rather than raw generation. Editing—the supers, the brand grading, the timing, the unglamorous packaging layer—is where commercial video actually gets made, and where the spending recurs. OpenAI left the one modality its rivals had learned to charge for.