China’s AI Models Compete on a New Frontier: Cost-Per-Token Advantage
According to the source citing UBS estimates, leading Chinese AI models may cost about one-tenth as much to train as comparable overseas systems and are priced at roughly 10%–20% of foreign API alternatives. The document suggests this cost curve—supported by MoE architectures, higher GPU utilization, and lower infrastructure costs—could become a durable enterprise advantage if compute capacity keeps pace with demand.