Svmuu News: Wang Dong, co-founder and CEO of GPU manufacturer Moore Threads, said, “ Large-scale models are developing very rapidly both domestically and internationally. Currently, leading companies are releasing new iterations of cutting-edge foundational models on average every two months. In terms of model invocation costs, we have found that China’s cutting-edge foundational models offer a significant cost advantage over foreign models of comparable intelligence levels; Chinese models provide better value for money, which precisely demonstrates that, given limited computing power, model companies have put in a great deal of effort to improve model efficiency, price-performance, and reduce training costs.”
Wang Dong pointed out that there is no “one-size-fits-all chip” in the inference market, but rather a combination of “solutions.” “The technical barriers to entry in the inference market are relatively low, and use cases are highly fragmented; no single company can monopolize all application scenarios. There is no single, absolutely perfect piece of hardware; through flexible hardware-software coordination, each model can find the hardware combination best suited to it, achieving the optimal balance between cost and performance. The market will see a surge in ISP companies, providing MaaS providers or end customers with more cost-effective and flexible customized inference services.”