By 2030, if any Chinese company is selling or renting chips to the Chinese or international market which can perform to within 20% of the flops/$ cost of training an AI at the frontier of capabilities?

This includes GPUs, TPUs, or any other processing unit used to train such models.

If there are no frontier models being trained using Chinese chips, then if first principles calculations, or judgements by an AI engineer training such models, or something equivalent convince the author that such models could be trained to SOTA with Chinese chips to within 20% of the cost of non-Chinese produces chips, then this question will resolve yes.