摩尔线程联合创始人王东：推理市场没有“万能芯片”，而是解决方案的组合

Odaily reports, Wang Dong, co-founder and executive president of GPU manufacturer Moore Threads, stated, "The development of large models is progressing rapidly both domestically and internationally. Top-tier companies now complete a new iteration of their frontier foundation models every two months on average. In terms of model inference costs, we have observed that China's frontier foundation models hold a significant cost advantage over models with comparable intelligence levels abroad, offering better cost-effectiveness. This also demonstrates that model companies, despite limited computational power, have done extensive work on improving model efficiency, pricing efficiency, and training costs."

Wang Dong pointed out that the inference market does not have a "universal chip" but rather a combination of "solutions." "The technical application threshold for the inference market is relatively low, and the scenarios are highly fragmented. No single company can monopolize all niche application scenarios. There is no single perfect hardware. Through flexible software-hardware co-optimization, every model can find its most suitable hardware combination to achieve the optimal balance between cost and performance. The market will see the emergence of numerous ISP companies, providing MaaS providers or end customers with more cost-effective and flexible customized inference services."