Wang Dong, co-founder and executive president of GPU maker Moore Threads, said the large model market is moving quickly in both China and overseas, with leading companies now rolling out a new frontier foundation model iteration about every two months on average. He said Chinese frontier foundation models show a clear cost advantage over overseas models at the same intelligence level, which in his view reflects extensive work by model developers to improve model efficiency, pricing efficiency and training costs under limited compute conditions.
On the inference side, Wang said there is no such thing as a universal chip that can serve every use case. Instead, he described the market as one built around combinations of solutions. He argued that inference applications carry a relatively low technical threshold while deployment scenarios remain highly fragmented, making it unlikely that any single company could dominate all verticals. Wang added that no single piece of hardware is absolutely perfect, and said flexible coordination between software and hardware allows each model to find a more suitable hardware mix and strike a better balance between cost and performance. He also said the market is likely to see the rise of many ISP companies serving MaaS providers and end customers with more cost-effective and flexible customized inference services.
Odaily reported that Wang Dong, co-founder and executive president of GPU maker Moore Threads, said large models are advancing quickly in both China and overseas, with leading companies completing a new iteration of frontier foundation models about every two months on average.
Wang said Chinese frontier foundation models have shown a clear cost advantage over overseas models with the same level of intelligence when it comes to calling costs, making Chinese models more cost-effective. He said that also shows model companies have done substantial work on model efficiency, pricing efficiency and training costs while operating under limited compute resources.
Inference demand is built around combinations of solutions
Wang said there is no “universal chip” in the inference market, only combinations of solutions. He described the inference market as having a relatively low technical barrier for applications while remaining highly fragmented across scenarios, which means no single company can dominate every segment.
He also said there is no absolutely perfect standalone hardware product. Through flexible coordination between software and hardware, each model can be matched with a hardware combination that suits it best, allowing for a better balance between cost and performance.
More ISP companies may emerge
Wang said the market will see a large number of ISP companies emerge to provide MaaS providers and end customers with more cost-effective and more flexible customized inference services.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
