The AI arms race has taken a new turn, with Chinese AI platforms offering models that perform competitively with US frontier systems at prices that make Silicon Valley's offerings look like luxury goods. Companies like Coinbase and DoorDash are shifting workloads to Chinese models that cost a fraction of their American counterparts, forcing US providers to rethink their pricing strategies.
Key Findings
- Chinese AI models offer API costs as low as $0.14 per million input tokens, compared to $5 for US alternatives like Claude Opus.
- Coinbase cut its AI spending by 50% after adopting Chinese models GLM-5.2 and Kimi K3, with actual token consumption increasing during the same period.
- DoorDash's CTO has pointed to better quality and lower costs with Moonshot's Kimi AI for specific tasks.
- Airbnb and Siemens have also begun shifting workloads to Chinese-developed models.
Implications
- The conventional wisdom that Chinese AI models lag US counterparts by 6-12 months on the capability frontier may no longer apply for most enterprise workloads.
- Efficiency and pricing are becoming key differentiators in a sector that historically competed almost exclusively on raw capability.
- Security concerns remain, particularly for enterprises in regulated industries, government contractors, and companies handling sensitive consumer data.
Market Shift
- By March 2026, 41% of downloads on Hugging Face were for Chinese open-source models, up from a relatively small share in the past.
- Open-weight models let companies self-host, customize, and maintain control over their data without sending it through a third-party API.
- The pressure is already reshaping how American AI companies think about their business models.