Tencent executives claim that GPUs are sufficient to train future generations of models, and the greater demand for GPUs comes from inference

Tencent executives stated that Tencent has sufficient high-end chips to continue training models, which is sufficient for future generations of models. At present, Tencent sees a greater demand for GPUs on the inference side, with user inference demands gradually increasing. The demand for inference tokens (lexical elements) is higher than previously expected. If more advanced models are developed, the inference requirements will be higher than the current requirements, which will consume more tokens. At present, the demand for AI is very high, exceeding the computing power provided by GPU resources. (First Financial)