Rise of Chinese AI Models Triggers Reassessment of Global AI Investment Risks
nashnova research
Chinese open-source models ramped token throughput from 4.37 trillion to 76.12 trillion in months, dwarfing the US total of 14.2 trillion; the market's focus is shifting from 'who builds fastest' to 'who can service the debt.'
How fast did China's AI compute share grow?
OpenRouter data for the week ending September 27 shows Chinese models processed 76.12 trillion tokens — versus just 14.2 trillion for US models.
In late April, China's figure was only 4.37 trillion tokens. In under five months it rose more than 17-fold.
This means → Chinese low-cost, open-source models are grabbing global compute share far faster than anyone projected. The narrative of US dominance no longer holds.
How are banks reassessing AI investment risk?
DBS CEO Susan Chen said the market now demands higher returns and better financing structures for AI deals.
She noted that weaker companies may face credit-rating downgrades, but called this a healthy tightening of market discipline.
In plain terms = Banks haven't stopped funding AI — they've started picking winners. Companies that can't show profits will find borrowing far more expensive.
What pressure are data-center developers under?
Jefferies global equity strategist Chris Wood warned that the AI boom is increasingly debt-dependent, raising risk for data-center developers.
The concern: if large-language-model providers can't build viable businesses, the hyperscalers expanding capacity for OpenAI and Anthropic will bear the fallout.
This means → The entire data-center expansion thesis rests on one premise — that AI companies will make money. If that premise fails, expansion turns into pure liability.
Why do Chinese open-source models make this more urgent?
Wood added that lower-cost open-source models mean more capital could flow to China.
In plain terms = If cheap, capable Chinese models keep eating market share, the data centers US firms are building at enormous cost may end up oversupplied — built but underused.
This reflects a deeper contradiction: current AI-infrastructure debt is sized for a market the US dominates alone. China's rise is undermining that assumption.
How exposed is DBS itself?
DBS handled roughly S$25 billion in AI-related transactions in 2026, mostly in semiconductor, storage, and data-center hardware financing.
The bank itself retained only a small portion of those positions.
This means → DBS acted largely as an intermediary — arranging deals for clients while passing most of the risk on to other investors in the market.
市场有风险，内容仅供研究参考，不构成投资建议。