The great national supercomputing effort, it turns out, is mostly busy making very short television. A report from ChineseAll says that about 55% of the computing resources used by Chinese AI services goes to producing short drama series and AI-generated video clips.
The numbers
The split, as reported, is lopsided:
| Use of compute | Share |
|---|---|
| Short drama series and AI video | about 55% |
| Advertising content for e-commerce | 24% |
| Fundamental research and technology development | 15% |
Add the first two rows and you get roughly 79% of capacity pointed at content meant for a mass audience or at selling things to it. Research gets the remainder of the headline figures, which is 15%.
What the arithmetic leaves out
The three listed categories sum to about 94%. The summary of the report does not say where the other 6% goes, and neither does it say how the figures were measured. It does not say which services were counted, or over what period. Those gaps matter, because a share of compute is a slippery number. It depends on what you count as an AI service in the first place.
So treat the percentages as the report's claim, not as an audited ledger of every GPU in the country.
Why it started an argument
The figures set off a discussion in the technology industry about how well computing capacity is divided between commercial work and the development of new technology. That is a polite way of asking whether the machines are being used for the thing people say they are being built for.
The case for the current split is not hard to imagine. Video generation and e-commerce advertising are tasks that customers pay for today. Fundamental research is a task that customers pay for, with luck, eventually. The report itself does not make either argument, and the sources I have do not include anyone defending or attacking the allocation by name.
The plain reading
What the report does show, if its figures hold, is that demand for AI compute in China is being driven mostly by the cheapest, fastest route from a model to revenue. Short dramas and product ads are the output. Research is the overhead.
That is not a scandal so much as a market doing what markets do. The awkward part is only that the same capacity is described, in other settings, as strategic infrastructure for the future of the technology. A few minutes of melodrama per episode is a fairly modest use of strategy.
What to watch
Two things would make this more than a headline. First, the methodology, which would tell us whether 55% is a measured share or a rough estimate. Second, whether the 15% moves over time. If it does not, the argument about allocation will keep returning, and the short dramas will keep being very well rendered.