China AI LLM through the lens of a DeepSeek expert

研报英文原文证据摘录

China AI LLM through the lens of a DeepSeek expert

11 May 2026

China AI EquitiesInternet Software & Services

LLM through the lens of a DeepSeek expert China

◆ We hosted DeepSeek product manager Eric Chen on 8 May Charlene Liu*

Head of Internet and Gaming Research, Asia Pacific

◆ Post-training, data, talent and chips key to boosting coding ability The Hongkong and Shanghai Banking Corporation

Limited, Singapore Branch

charlene.r.liu@hsbc.com.sg

◆ 2026 chip supply set to stay tight; domestic chips can meet +65 6658 0615

40-50% of inference demand Charlotte Wei*

Analyst, Internet Research

The Hongkong and Shanghai Banking Corporation Limited

Domestic model competitive landscape: On 1) Coding: The expert thinks Zhipu charlotte.wei@hsbc.com.hk

(Knowledge Atlas), Kimi and DeepSeek (DS) are leading domestic players. He ranks +852 2996 6539

four key drivers of coding performance: post-training strategy, data quality – such as

high-quality code on GitHub platform and files on technical discussion, talent who * Employed by a non-US affiliate of HSBC Securities (USA) Inc, and is

not registered/ qualified pursuant to FINRA regulations

understand both LLMs and programming languages, and chip resources. 2) Agentic

capability: The expert ranks b b ’s Q K and DS first on agentic ability and

tools. Between hyperscalers and frontier models, he thinks Tencent and Alibaba

benefit from their vast ecosystems consisting of a user base, high-quality proprietary

data, and ready-made products/tools for integration. Independent labs differentiate via

base-model iteration and architecture. 3) Multi-modal and video generation:

The expert calls out Minimax and Zhipu for multimodal comprehension strength and

Kuaishou and Minimax for video generation.