China AI Insight

Enterprise AI Launches

• 🔥 Tencent Dayuan in WeCom: Tencent will launch AI assistant Dayuan inside WeCom built on DeepSeek models, with...

• GLM-5.2 以四分之一 token 成本接近 Anthropic 企业级表现，形成直接市场挤压
• 中国模型在 OpenRouter 周处理 token 量从 4.37 万亿跃升至 21.37 万亿，抢占份额
• DeepSeek 获 74 亿美元融资，资本恐慌加剧 Anthropic 上市前压力

DeepSeek's $7.4B raise and workforce doubling plans highlight aggressive supply-side scaling, backed by Tencent as an investor.

• Demand signal:...

Anthropic is urging US regulators to close legal loopholes that let Chinese firms like Alibaba distill frontier models such as Qwen, citing...

China's embrace of AI reflects pragmatic optimism shaped by poverty memories and the Century of Humiliation, driving a resolve to master technology or...

Lenovo's AI Host Mini marks a hardware vendor's bet on AI-native devices, swapping conventional software installs for a proprietary marketplace of 8,000+ downloadable AI skills that connect directly to Feishu, WeChat and QQ.

Qwen Code's five releases this week target exactly the pain points slowing developers down.

• Voice input replaces tedious typing with /voice — the...

The Economist's Drum Tower newsletter interviews the Zhipu AI team on their technical progress, business model, and future vision, framing the company at the center of China's latest AI moment where a new model has narrowed the gap with America.

Shanghai startup Fysics AI unveiled Fysiverse, a physics-based world model that embeds real-world physical laws directly into its simulator instead of...

China's equity market is undergoing a historic pivot as AI and semiconductor firms rapidly displace state-owned banks in market cap rankings. Zhongji...

Tencent and Shunwei's $300 million investment in LiblibAI marks a clear pivot: capital is now chasing AI-native consumer apps with strong creator...

ByteDance launches paid subscriptions for Doubao, the AI chatbot with over 300 million MAU, while rivals like DeepSeek remain free. This tests paid tiers in China's consumer AI market.

Open-Weight Model Breakthroughs

• 🔥 GLM 5.2 Release: Z.ai (formerly Zhipu AI) released GLM 5.2, an open-weight model that outperforms ChatGPT,...

Huawei AI Hardware Shifts

• 🔥 Huawei AI Systems: Huawei AI systems deliver 3x Nvidia memory and double the compute via CloudMatrix and Atlas...

DeepSeek closed its first external round at over $50 billion with one non-negotiable term: investors must promise not to poach staff or push them to...

Baidu ERNIE-4.5-VL introduces a multimodal vision-language model family with thinking variants, visual grounding, reward model, GRPO, and DAPO-style sampling.

• 字节跳动年消费超10亿美元，通过微软Azure购买OpenAI模型
• OpenAI/Anthropic拒直接服务中国，微软却凭借特殊协议继续销售
• 蚂蚁、美团、腾讯等多家中国企业同样通过Azure采购AI模型
• 微软中国AI收入增速全球最快，但模型不托管于境内数据中心

US export controls on advanced chips prompted Beijing to block purchases from American firms, ending reliance on US suppliers and spurring domestic...

China is accelerating a global AI cooperation organization and offering free, downloadable models like DeepSeek and Qwen to the Global South, directly...

Investors are using pair trades to bet on AI outperformers like Alibaba while shorting laggards such as Meituan, delivering around 130% returns YTD.
-...