China AI impact on global AI theme: A guide to the right questions and their possible answers

研报英文原文证据摘录

China AI impact on global AI theme: A guide to the right questions and their possible answers

Powered by Lab Global15 JulyResearch2026ab Evidence UBS YES

EM & APAC Equity Strategy Equity Strategy

GlobalChina AI impact on global AI theme: A guide to

the right questions and their possible answers Sunil Tirumalai

Strategist

sunil.tirumalai@ubs.com

+91-22-6155 6080

China-grown LLMs are gaining traction in the West, and are cost effective Wei Xiong

Chinese LLMs are gaining traction with Western AI customers - both enterprise (Figure 1UBS Analyst

EvidenceLabsurvey:WhichifanyofthefolowingLMs(largelanguagemodels)areyourcompanycurentlyusing?(selectalthataply)) and consumer. They are increasingly seen as 'good enough to do most tasks' at a S1460518100005

wei.xiong@ubs.comfraction of the cost (usually 15-20%) of leading US models. Their cost efficiencies

+86-21-3866 8883

emerge from design choices and ecosystem advantages - not all of which can be

replicated easily by US based models (see the detailed work done by our China internet Randy Abrams

team). We sharpen the key questions that arise from this trend on investment into EM Analyst

randy.abrams@ubs.comand Asian AI themes.

+886-2-8722 7338

Impact on US LLMs ≠ Impact on US hyperscalers Karen Hizon

The pressure to optimise token costs is understandable on the US frontier model

karen.hizon@ubs.com

providers (our US software teams work is here: "The Ramp in Token Optimization"). +852-2971 6741

However - it is important to note that the most popular Chinese models in the West are

Kruti Shah, CFAopen-weight models (like Deepseek, Alibaba's Qwen, Zhipu's GLM) - where the

inference computing mostly happens on some data centre closer to (or within the same

kruti.shah@ubs.com

country as) the end user.