The Chinese model API price war, explained
Why Chinese foundation model API pricing fell by over 90% in 2024–2025, and how commodity token economics differ between Chinese cloud giants and Western AI providers.
The collapse in token prices
Starting in mid-2024, Chinese model providers entered an aggressive price war. ByteDance's Volcano Engine cut Doubao API prices by over 99%, followed immediately by Alibaba Cloud (Qwen), Tencent (Hunyuan), Baidu (ERNIE), and DeepSeek.
Inference token prices for frontier-class models dropped to mere cents per million tokens, far below Western API counterparts like GPT-4o or Claude 3.5 Sonnet.
Cloud ecosystems vs API standalone business
The underlying economic driver is ecosystem lock-in: cloud conglomerates treat API tokens as a top-of-funnel customer acquisition channel for their broader cloud storage, compute, and database services.
At the same time, rapid optimization in FP8 quantized inference, continuous batching, and custom serving engines enabled providers to handle millions of queries at sustainable operating margins despite nominal price drops.
Frequently asked questions
Why did Chinese model API prices fall so far in 2024-2025?
Starting in mid-2024, Chinese model providers entered an aggressive price war. ByteDance's Volcano Engine cut Doubao API prices by over 99%, followed by Alibaba Cloud, Tencent, Baidu and DeepSeek, dropping inference token prices for frontier-class models to mere cents per million tokens.
How can Chinese providers sustain such low token prices?
Cloud conglomerates treat API tokens as a top-of-funnel customer acquisition channel for their broader cloud services, while FP8 quantized inference, continuous batching and custom serving engines let them handle millions of queries at sustainable operating margins.