At the WAIC AI compute forum, Moore Threads co-founder and president Wang Dong said leading AI firms now iterate frontier base-model versions roughly every two months. He said Chinese frontier base models show a clear cost advantage versus foreign models of comparable capability, reflecting heavy work by model firms on model efficiency, price efficiency and training cost. Wang said the inference market has no universal chip but a mix of solutions: application barriers are low and scenarios highl

At the WAIC AI compute forum, Moore Threads co-founder and president Wang Dong said leading AI firms now iterate frontier base-model versions roughly every two months. He said Chinese frontier base models show a clear cost advantage versus foreign models of comparable capability, reflecting heavy work by model firms on model efficiency, price efficiency and training cost. Wang said the inference market has no universal chip but a mix of solutions: application barriers are low and scenarios highly fragmented, so no single vendor will dominate all subsegments. He added there is no perfect single hardware; flexible software–hardware co‑design will match each model to the optimal hardware mix to balance cost and performance, and a wave of inference service providers will emerge to offer more cost‑effective, customizable inference services to MaaS providers and end clients.