CLSA: Chinese AI models' share of token usage rises to about 65%

CLSA released a research report stating that in September, Chinese AI models accounted for approximately 65% of token usage and 30% of dollar spending on the OpenRouter platform, benefiting from roughly an 80% price discount versus US peers. Lower pricing drove usage-based revenue growth, consistent with the Jevons Paradox. The CLSA research report noted that OpenRouter data shows rapidly growing demand for AI models, with weekly token consumption exceeding 146 trillion by end-September and combined model annual recurring revenue (ARR) surpassing US$1.5 billion. The bank said OpenAI saw the largest increase in spending share, while among Chinese providers Z.AI (02513) led, followed by DeepSeek and Moonshot. Z.AI's progress on AWS Bedrock could be a key indicator of the monetization potential of open-weight models. The bank pointed out that in September, top models from US and Chinese large language model providers accounted for 31% and 65% of token consumption, respectively, a marked shift from 65% and 22% in January; in terms of dollar spending, US and Chinese providers accounted for 70% and 30% in September, respectively, versus 94% and 6% in January. Weekly spending on Chinese models has risen 27-fold year-to-date, driven by a 67-fold increase in token usage, while blended token prices have nearly halved. The bank estimated that OpenAI was the largest gainer in OpenRouter spending share, with its weekly spending share rising 20 percentage points year-to-date to 27%, closing in on Anthropic's 28%. Among Chinese providers, for the week ended September 27, Z.AI, DeepSeek and Moonshot led in spending share at 10.5%, 7.1% and 3.9%, respectively, while Tencent (00700) also recorded 3.9% benefiting from the Hy4 preview. Open-weight models' revenue share reached 29%, highlighting monetization opportunities for Chinese labs distributing through compute partners. Z.AI GLM 5.3's progress on AWS Bedrock remains a key signal for potential revenue and margin upgrades.