Will there be a Frontier AI lab in China before 2026?

As of April 2023, the largest model currently known is GPT-4 and while information about the compute needed to train the model is not public, GPT-4 is estimated by Epoch to have been trained with 2.1E+25 FLOPs.

The largest Chinese-made model is GLM-130B, estimated at 4.60E+22 FLOPs, about 1/500 the amount of GPT-4. It was made by Tsinghua KEG, a university.

As of April 28, 2023, the Frontier Labs (defined in resolution criteria) are OpenAI and Google Deepmind (source) (and possibly Anthropic, as the cost to train Claude has not been publicly estimated). Labs that would join this with a ~10x growth in training compute include Microsoft, Meta, Yandex, Amazon, and Hugging Face.

Potential Chinese companies that could become a Frontier AI Lab are SenseTime, Baidu, Alibaba, Tencent, and ByteDance.