China’s AI price war converges with year end procurement, pushing workloads to domestic chips
Kai Tanaka ·
China’s largest AI vendors are entering the year end procurement window with rock bottom model pricing and a clear shift to domestic accelerators, according to company disclosures and market data tracked in Chinese industry press. The timing matters because state owned enterprises and major corporates typically finalize next year’s cloud and AI contracts in November and December, which will lock in hardware and model choices for 2025.
Baidu, Alibaba Cloud, Tencent and SenseTime cut list prices for foundation model APIs in May 2024, and base tiers have remained at or near zero for smaller models since then. The sustained discounts are pushing customers to experiment widely, then commit volume at the platform level during annual budgeting.
Chinese AI pricing Baidu’s ERNIE Speed and Lite tiers have been promoted with free or near free API access, while higher capacity variants are priced aggressively relative to U.S. rivals. Alibaba Cloud’s Qwen API cuts in 2024 lowered per token rates by double digit percentages across sizes, and ByteDance’s Doubao and Zhipu AI matched with low cost enterprise bundles.
Regulatory approvals have kept the supply of models high. By mid 2024 the Ministry of Industry and Information Technology and the Cyberspace Administration of China had cleared more than one hundred large models for public release, which intensified price competition in cloud marketplaces.
Domestic compute shift U.S. export controls on advanced GPUs, updated in late 2023 and 2024, have accelerated a move to Chinese accelerators. Huawei’s Ascend line, along with in house chips from Baidu Kunlunxin, Alibaba Hanguang and Tencent Zixiao, now feature prominently in private cloud offerings and on premises deployments described in vendor materials and provincial digital government frameworks.
That transition is reshaping China’s cloud stack. Canalys estimated Alibaba Cloud, Huawei Cloud, Tencent Cloud and Baidu AI Cloud at the top of the domestic market in 2024, and their current bids pair low cost model access with guaranteed availability on domestic hardware for inference.
For global technology companies, the near zero pricing in China and the build out around local accelerators create a parallel AI stack with different cost curves and supply chains. That will influence how multinational software vendors price inference, how component makers plan capacity for packaging and networking, and how regulators weigh data localization against access to frontier compute.