Global AI Trend Tracker: Moonshot’s Kimi K3 – China’s DeepSeek 2.0 moment? And what are the implications for the global AI supply chain?

研报英文原文证据摘录

Global AI Trend Tracker: Moonshot’s Kimi K3 – China’s DeepSeek 2.0 moment? And what are the implications for the global AI supply chain?

M to high

China Internet & New Mediacost-performance LLM.

Jialong Shi - NIHK

How does Kimi K3 change the LLM competitive landscape? Will China’s LLM Jialong.shi@nomura.com

players disrupt global market? +852 2252 1409

K3’s key breakthrough is that the model is engineered for long-horizon agentic work, not Rachel Guo - NIHK

just chat. On Moonshot AI’s official website, there are use cases on complex agentic task, rachel.guo@nomura.com

including an unattended chip-design process in 48-hours, end-to-end knowledge work, in- +852 2252 1400

depth industry research with interactive visualisation. For example, Kimi K3 has now

joined a Chinese elite LLM club in the global market, together with Zhipu (2513 HK, Not

rated), DeepSeek (unlisted), etc, and their target markets now span the full price stack

(premium/economy/low-end), in our view. On cost-per-task, K3 (USD0.94) sits at parity

with GPT-5.6 Sol (USD1.04), well below Opus 4.8 (USD1.80) and Fable 5 (~USD2.75),

but far above GLM-5.2 (USD0.32–0.47) and DeepSeek V4 Pro (USD0.04). According to

Open Router’s statistics, Chinese models already exceed 45% of the global

developers’ token volume (<2% a year ago, link). We think China’s LLM players

(including both close-source and open-source) will continue to gain traction in the global

market, providing highly cost-efficient models which could fulfil the demand for a wide

range of IT workloads. Meanwhile, top tier global LLMs could also accelerate their

innovation, targeting the most complex workloads (i.e. science) and maintaining their

technology & pricing premium, in our view. Moreover, due to geopolitical risks and the