China AI & robotics developments, rewritten weekly-context-free for US readers. Published daily at 07:00 Beijing time.
Fri, Sep 11, 2026
DeepSeek extends V4 Pro API availability beyond September 2026Chinese AI startup DeepSeek announced it will continue offering its V4 Pro model via API after the originally scheduled Sept. 14, 2026 sunset date, maintaining current pricing structures to accommodate user demand.
The Hangzhou-based developer of open-weight large language models reversed a planned discontinuation following community feedback. According to IT Home, the company stated on Sept. 11 that access to V4 Pro remains available indefinitely under existing billing terms.
This decision preserves continuity for developers currently integrated with the model, which competes globally against proprietary offerings from US firms like OpenAI and Anthropic. The extension suggests sustained utility for the legacy version despite newer releases.
DeepSeek launches V4.1 Flash with new asymmetric architectureChinese AI lab DeepSeek released its V4.1 Flash model two days after initial testing, introducing a Causal-Encoder-Decoder structure that significantly reduces inference costs while supporting native multimodal vision.
The 552-billion-parameter Mixture-of-Experts model activates only 8 billion parameters for input and 16 billion for output. This asymmetric design allows for efficient processing compared to standard models of similar size. The release follows closely on the heels of a brief test phase, signaling an accelerated deployment strategy by the Hangzhou-based developer.
Enflame Tech surges on Shanghai IPO debutChinese AI chipmaker Enflame Technology saw its shares jump nearly 180% on its first day of trading on the Shanghai Stock Exchange's STAR Market, signaling strong investor appetite for domestic semiconductor alternatives.
Enflame, one of China’s leading GPU developers alongside firms like Moore Threads and Iluvatar CoreX, priced its initial public offering at 142.18 yuan (roughly $20). The stock opened at 410 yuan and peaked at 475 yuan before closing at 397 yuan.
The listing places Enflame among the "five little dragons" of Chinese GPUs, a group competing to reduce reliance on Nvidia amid US export controls. This debut follows recent listings by peers, highlighting the sector's valuation momentum in Beijing's push for self-sufficient AI infrastructure.
L'Oréal-backed fund invests in Douyin-native beauty brand DiluoweiA private equity fund linked to L'Oréal has acquired a minority stake in Shanghai Xinji Tunyu, the parent company of Chinese makeup brand Diluowei. This move highlights global beauty conglomerates' strategy to tap into China's fast-growing social commerce ecosystem.
According to business registration records from Sept. 10, Kaifu Beauty Future (Shanghai) Private Equity Investment Fund increased its stake in Shanghai Xinji Tunyu Technology Group to 4.4255%. The fund is associated with L'Oréal's investment arm, signaling continued interest in high-growth local brands.
Diluowei operates primarily through Douyin, China's version of TikTok, leveraging live-streaming and short-video content for sales. The acquisition underscores how international players are partnering with or investing in digital-native Chinese brands to maintain relevance in one of the world's largest beauty markets.
China shipbuilders capture 82% of global orders in H1 2026Chinese shipyards secured 82.3% of global new orders by deadweight tonnage in the first half of 2026, a 173% year-on-year surge that underscores their dominance over Japanese and South Korean rivals.
The data, reported by industry analysts, highlights a widening gap between Chinese yards and competitors in Japan and South Korea. This market share excludes existing backlogs, indicating sustained momentum beyond immediate bookings.
Analysts attribute this growth to China’s integration of AI in manufacturing processes, which has improved efficiency and cost competitiveness. The trend reinforces Beijing’s strategic push to lead high-tech maritime industries, challenging traditional naval powerhouses.
DeepSeek founder Liang Wenfeng shifts stance toward capital marketsLiang Wenfeng, founder of DeepSeek and quant firm High-Flyer, is reportedly reconsidering external financing or an IPO, marking a pivot from his previous insistence on independence. This potential move signals a strategic shift for one of China’s leading AI developers.
Two years ago, Liang stated DeepSeek had no short-term plans to raise funds or separate from its parent, the quantitative trading firm High-Flyer. The new report suggests this position may be evolving as competitive pressures in generative AI intensify.
High-Flyer has historically funded DeepSeek internally, allowing it to develop models like V2 without diluting ownership. A move toward public markets would provide significant capital but introduce regulatory scrutiny. Analysts view this as a response to the high costs of training frontier models.
AI shifts from software to capital-intensive infrastructureAnalysts argue that despite AI's software-like interface, the sector has fundamentally transformed into a capital-intensive industry driven by massive hardware and energy costs.
The article contrasts user-facing applications like ChatGPT, Gemini, Claude, and DeepSeek with their underlying physical requirements. While model updates appear seamless to users, they rely on expensive data center expansions. This disconnect masks the heavy financial burden required to sustain competitive large language models.
NetEase Youdao launches MFi-certified AI earbuds for iPhone usersNetEase's education subsidiary Youdao released OpenPods, the first AI-enabled earbuds with Apple MFi certification, targeting iPhone users with integrated call recording and translation features.
The device, priced at 1,499 yuan (roughly $210) after launch subsidies, is sold exclusively on JD.com. It distinguishes itself by offering native iOS integration, a gap left by many Android-centric Chinese hardware makers.
Youdao, known for its translation tools and online education platforms, positions this as an "agent" accessory rather than standard audio gear. The move highlights the trend of embedding large language model capabilities into consumer hardware to compete against established players like Apple and Huawei.
Arm launches second-gen mobile computing platform for AI agentsArm unveiled its CSS for Mobile 2 at a Shanghai event, targeting agentic AI and native graphics workloads with new C2 CPU clusters and Mali G2-Ultra NX GPUs.
The announcement occurred on Sept. 8 during Arm Everywhere China, the chip designer’s flagship regional conference. The new Compute Subsystem (CSS) integrates Arm C2 CPU clusters and Mali G2-Ultra NX GPUs, specifically engineered to support AI-native graphics and autonomous agent tasks on mobile devices.
This move underscores Arm’s strategy to capture the emerging market for on-device generative AI. By providing a pre-integrated subsystem, Arm aims to accelerate smartphone OEM adoption of specialized hardware capable of running complex local models without relying solely on cloud infrastructure.
Thu, Sep 10, 2026
Former AI researchers warn of security risks after Hugging Face hackEx-employees from OpenAI and Anthropic highlighted critical vulnerabilities in AI infrastructure following a recent breach at Hugging Face, urging stricter security protocols.
The warning comes from former staff at leading labs, emphasizing the fragility of open-source model hubs. They argue that current safeguards are insufficient for high-stakes deployments.
Hugging Face, the popular platform for sharing AI models, experienced a significant security incident recently. The breach exposed potential risks in how developers distribute and access pre-trained weights.
Industry experts suggest this event may accelerate regulatory scrutiny on AI supply chains. It underscores the need for robust authentication and monitoring in collaborative research environments.
Mech-Mind founder critiques Galaxy General amid robot IPO scrutinyShao Tianlan, founder of Beijing-based 3D vision firm Mech-Mind, publicly criticized rival Galaxy General as China's robotics sector shifts from hype to financial accountability ahead of initial public offerings.
Mech-Mind, a leading supplier of industrial robotic vision systems, faces intensifying competition in China's crowded humanoid and service robot market. Founder Shao Tianlan’s comments reflect growing investor skepticism toward valuation metrics lacking clear revenue paths.
The critique signals a broader industry pivot: Chinese robotics startups must now demonstrate profitability to satisfy regulators and public markets. This 'accounting phase' pressures firms like Galaxy General to justify their rapid growth narratives with tangible financial results.
DeepSeek V4.1 Flash cuts KV cache by 437x, slashing agent costsDeepSeek has introduced its V4.1 Flash model, featuring an architecture that reduces key-value cache memory usage by a factor of 437, potentially reshaping the economics of AI agent deployment.
The Hangzhou-based AI lab DeepSeek, known for its cost-efficient open-source models, claims this architectural shift dramatically lowers inference hardware requirements. By optimizing how attention states are stored, the company says it can serve complex agents on significantly cheaper GPUs.
This move intensifies competition among Chinese AI developers to lower operational barriers. If verified, such efficiency gains could accelerate the adoption of autonomous agents in enterprise applications where memory bandwidth is a primary bottleneck.
Leju, Huawei launch first 5G-A humanoid robot network packageChinese humanoid robotics firm Leju Robotics, alongside telecom operator Anhui Telecom and tech giant Huawei, has unveiled the nation's first dedicated 5G-Advanced networking package for embodied AI robots.
The "Embodied Wing Link" plan targets high uplink bandwidth, low latency, and multi-robot concurrency needs. It supports a pilot deployment of swarm inspection by humanoid robots in an industrial park, marking a practical application of 5G-A infrastructure in robotics.
DeepSeek V4.1 Flash model launches on China's National Supercomputing InternetDeepSeek, the Hangzhou-based AI startup known for cost-efficient large language models, has deployed its V4.1 Flash variant on China's National Supercomputing Internet, expanding access to high-performance inference APIs for domestic developers.
The launch, reported by IT Home on Sept. 10, allows users to access the model’s API through the platform’s "Model Service" portal. This integration leverages state-backed infrastructure to reduce computational barriers for enterprises seeking scalable AI solutions.
DeepSeek has previously gained global attention for its open-source R1 reasoning model and competitive pricing against US rivals like OpenAI. Placing V4.1 Flash on a national supercomputing network signals continued efforts to align private AI innovation with China’s strategic digital infrastructure goals.
SAIC Motor to pre-sell Roewe Jiayue 07 with Momenta AI driving systemSAIC Motor will begin pre-sales for its new Roewe Jiayue 07 extended-range vehicle on Sept. 21, featuring a high-level autonomous driving solution from Momenta and a claimed combined range of 1,480 kilometers.
The Roewe Jiayue 07 is an extended-range electric vehicle (EREV) from SAIC Motor, one of China’s largest state-owned automakers. It integrates Momenta’s R7 world model for advanced driver assistance systems (ADAS), marking a deeper partnership between traditional OEMs and leading Chinese AI startups.
With a total range of 1,480 km (approx. 920 miles), the vehicle targets long-distance travel needs common in the Chinese market. The launch underscores the intensifying competition in China’s EV sector, where hybrid and EREV models are gaining traction alongside pure electric vehicles.
Ninefingers launches AI-quant assistant for retail investorsNinefingers Intelligence, a Shanghai-based securities advisory firm listed in Hong Kong, unveiled its new 'Gudao Zhihang' platform on Sept. 9, aiming to integrate artificial intelligence with quantitative strategies.
The product targets the transition toward what the company calls "Securities Advisory 4.0," combining algorithmic trading signals with generative AI interfaces. Ninefingers (9636.HK) positions this as a shift from traditional human-led advice to data-driven automation.
While specific technical metrics were not detailed in the summary, the launch reflects broader industry efforts by Chinese fintech firms to leverage large language models for personalized investment guidance. The event was hosted by subsidiary Shanghai Ninefingers Cloud Intelligent Technology.
WeChat pilots AI agents within personal social networksTencent's WeChat is quietly testing the integration of artificial intelligence agents directly into its core messaging and social features, marking a significant shift in how AI interacts with established human relationships.
The Beijing-based tech giant is exploring ways to embed personal AI assistants into existing chat groups and direct messages. This move aims to leverage WeChat's massive user base by enhancing communication efficiency without disrupting the familiar social dynamic.
Unlike standalone chatbots, these agents operate within trusted circles, raising new questions about privacy and data security. Industry observers note that this strategy differentiates Tencent from competitors who focus on general-purpose AI apps, instead prioritizing deep integration into daily life tools.
Moonshot AI denies dual listing plans in Hong Kong and ShanghaiMoonshot AI, the Beijing developer of the Kimi chatbot, has dismissed reports suggesting it is preparing for a dual listing on the Hong Kong and Shanghai stock exchanges.
The denial follows earlier claims by two sources cited by the South China Morning Post that the large language model firm was exploring the move to raise capital. A person familiar with the matter told the Science and Technology Innovation Board Daily that the information was inaccurate.
This clarification comes amid heightened scrutiny of Chinese AI startups seeking public market access. Moonshot AI remains private, continuing its development of the Kimi assistant while navigating complex regulatory environments for cross-border listings.
Anthropic reports AI distillation by Chinese firmsAnthropic released a report Thursday alleging persistent model distillation attacks by Alibaba, Moonshot AI, and DeepSeek, highlighting escalating intellectual property concerns in the global AI race.
The US developer of Claude said three major Chinese entities targeted its models to replicate capabilities at lower cost. This marks a shift from isolated incidents to organized campaigns.
Moonshot AI, known for its Kimi chatbot, joined state-backed Alibaba and DeepSeek, a Hangzhou-based startup, on the list. Anthropic claims these efforts intensified recently, threatening the competitive moat built through massive compute investments.
Physix Frontier