DeepSeek 擬募資 $12B+，Manus 脱離 Meta 後獲逾 $500M，Kling 籌備逾 $1B IPO

China AI Weekly Digest summarizes the week’s big news, model updates, policy, products, and research, sourced from the aggregated newsfeed of China AI Index. The digest below is mostly compiled and drafted by an AI agent sourcing from credible news outlets, then fact-checked and edited by me.

This week is a bit quiet due to China’s national day holiday: 132 stories tracked (0 editor picks, 64 English-language, 68 Chinese-language 🇨🇳) over 2026-10-03–2026-10-10.

The Big Three

DeepSeek is raising at least $12 billion, with Tencent backing the round. It would be one of the largest private rounds any AI company has raised anywhere. Reports differ on the size: CNBC’s sources say DeepSeek is weighing doubling the round to as much as $15 billion, and Chinese outlets describe it as a ¥100 billion-class raise and credit Liang Wenfeng with making the move. (Bloomberg/Reuters/CNBC/36Kr)

Manus raises more than $500 million in its first round since splitting from Meta. After its split from Meta, Manus is independent again with fresh money. It is reopening its Beijing office and hiring heavily, and it’s repositioning around personal AI assistants as OpenAI and Meta both push into “agent 2.0.” In Chinese media this was the most-covered story of the week, read as a sign the agent race is warming back up even as frontier-lab revenue forecasts get trimmed. (CNBC/Reuters/TechCrunch/Nikkei/量子位)

Kling reportedly selects banks for an IPO targeting more than $1 billion. Bloomberg reports that Kuaishou’s AI video business has selected banks for a planned listing. An IPO would bring a closer look at the economics behind generative video: how much customers pay, how often they use it and what each generation costs to serve. For now, this is a financing plan, not a completed listing. (Bloomberg)

Models

Video generation provided a concrete launch this week, while Chinese open models faced more competition abroad.

Shengshu released Vidu Q4 Preview, with its official announcement describing support for up to 15 image references and three audio references. Chinese coverage also reported the launch; the advertised improvements in expressiveness still deserve hands-on testing. (爱范儿/@ViduAI_official)
ByteDance put Doubao 2.1 Lite, a lightweight model tuned for office work, into Doubao Work alongside a new infinite-canvas feature. (雷峰网)
Reflection unveiled an open AI model positioned against Chinese competitors. The launch adds another contender to a market where Chinese labs have made open weights central to their international reach. (Bloomberg/FT)
Alibaba chairman Joe Tsai argued that open source offers Europe its best opportunity in AI. It is also a strategy that aligns with Alibaba’s efforts to distribute Qwen globally. (Bloomberg)
DeepSeek pushed a holiday update to its Harness agent tooling and is hiring heavily for an elastic-compute team. (36Kr/量子位)

Funding

Beyond the headline rounds, chipmakers are seeking more capital.

Biren is seeking approximately $510 million through a share placement, according to SCMP. The fundraising is the substantive development here—additional capital for a Chinese AI chip company competing in a demanding hardware market. (SCMP)
Chinese state funds are increasing their backing of Hua Hong’s legacy-chip push. That sits alongside the frontier AI chip race: China’s semiconductor investment extends across the manufacturing stack. (SCMP)
A $200 million AI lab started by a former ByteDance intern is taking on Fei-Fei Li’s World Labs. (Bloomberg)
Hu Han, a former core engineer on Tencent Hunyuan, is raising for his own startup at a target valuation of several hundred million dollars. (雷峰网)

Policy

Investment restrictions and agent security both moved from abstract concerns toward concrete cases.

The US issued its first outbound-investment fine involving a Chinese robotics AI deal, according to SCMP. (SCMP)
Hackers reportedly used a Chinese AI agent tool to breach South Korean banks. Separately, researchers are tracking a Chinese AI “agent fleet.” (WSJ/The Information/TechCrunch)
ChinaTalk examined how China’s AI safety work is funded, and High Capacity asked whether Beijing actually wants an AI slowdown. (ChinaTalk/High Capacity)

Products

The more revealing updates put AI into workflows people can actually use.

Doubao Work added a creation canvas and integrated Doubao 2.1 Lite. The product change brings materials and generated results into a shared workspace, extending its use for design and office tasks. (雷峰网)
Z.ai announced that GLM-5.3 is available on Amazon Bedrock. Chinese reporting also covered its expansion through overseas cloud providers—an enterprise distribution development, rather than a new model release. (财联社/@Zai_org)
StepFun announced Step 5 Preview’s availability on OpenRouter, alongside a week of free access rolling out through several coding tools. (@StepFun_ai)
Manus described its video editor as combining footage generation with research, coding and editing. (@ManusAI)
Pony.ai said it is working with Uber to bring its Gen-7 robotaxis to London. (@PonyAI_tech)
Bloomberg argues China is “trouncing” the US in consumer adoption of generative AI. (Bloomberg)
StepFun’s hardware arm will launch the STEPX Neo, which it bills as a model-native AI-agent phone, in Shanghai on October 13. (量子位)
Analysts expect Alibaba’s AI cloud revenue to grow more than 50% after its spending blitz. (SCMP)
China is racing to build AI data centers across its energy-rich interior. (FT)

Research

This week’s research reading raises questions about reliability and the evidence behind safety claims.

ByteDance researchers say they have found why DeepSeek’s performance swings between strong and weak. (量子位)
Reuters reported a study finding published safety tests for just 3.6% of Chinese model releases. (Reuters)
Synopsys wants to work with Chinese AI labs to speed up chip design. (Nikkei)

Subscribe now