China Desk

China News, Summarized 15 Oct 2026, 16:07 UTC 34 stories 22/23 sources

Today's brief

China's biggest apps just decided an AI agent has to be invited in

151 apps told ByteDance's phone agent no

The vote is in. ByteDance's Doubao phone assistant is an AI that reads the screen and taps through other companies' apps. On 14 October it published the result of its own consultation. 151 apps refused screen automation, 65 of them Tencent's and 23 Alibaba's. WeChat, Taobao, Meituan and DeepSeek are on the list.

Silence meant yes. The protocol, SAEP, is opt-out. Apps had a 30-day window from 14 September, and non-responders defaulted to allowed. PANews counts 155 refusals. Citing QuestMobile, it says only Douyin (ByteDance's own app) and JD.com among the ten biggest apps did not refuse.

A price, not a veto. Doubao tries structured MCP and A2A calls first, meaning interfaces apps open deliberately. Screen control is only the fallback. Refusing the click-through leaves a door: expose an interface, on your terms, revocable. My inference: the big apps are positioning to sell access, not deny it.

Refuse inbound, build outbound. Tencent's WorkBuddy now sits inside WeCom reading messages. Alibaba's Qwen Office plugs into rivals' tools. Invited connectors are fine; uninvited tapping isn't. Nor is this only China. JPMorgan chief executive Jamie Dimon said the bank won't let external agents into customer accounts. Amazon blocked Meta's Muse agent.

Phones are picking other doors. OPPO, whose group shipments fell over 20% in the third quarter per Counterpoint, is plugging Manus and ChatGPT into its assistant. StepFun's agent-first phone goes on sale tomorrow. Apple's route runs through the regulator.

Go deeper on this section: ClaudeChatGPT

Beijing waves through Apple's easy features

Cook announced it in Shanghai. Apple executive chairman Tim Cook said Apple Intelligence and Alibaba's Qwen models are approved. A beta is due late October with live translation, writing tools and photo clean-up. Siri AI is not included.

Notice who carried it. The state-backed outlet The Paper ran the interview, and Reuters relayed it. Commerce Minister Wang Wentao met Cook the same day, IT Home reports.

THE TELL: Siri AI is the part that would act across apps, the layer those 151 refusals just fenced. Yesterday I noted it had skipped Brussels and Beijing, and Beijing's half is now confirmed. My inference: content tools clear first, delegated action last.

A hotline is being drafted. SCMP's sources say US and Chinese technical teams are finalising an AI-incident framework. They hope to announce it when Xi and Donald Trump meet at APEC in Shenzhen next month.

Go deeper on this section: ClaudeChatGPT

Exports surged and Brussels hardened

Exports rose 36.3%. September's jump lifted the monthly surplus to $133.5bn and the nine-month total to $938bn. THE NUMBER: semiconductor exports are up 108.1% for January–September and autos 54.3%, AP reports.

Two framings. AP leads on mounting pressure to rebalance. Premier Li Qiang told the Canton Fair that China is as big a buyer as a seller, and that selling benefits its neighbours, per Sinocism. One is a complaint, the other a rebuttal.

Brussels read the hybrid deal as proof pressure works. SCMP reports that EU planners now plan to keep talking while tightening. My read: a hybrid cap and a rebate cut are small beside a 36% month.

Go deeper on this section: ClaudeChatGPT

Zhipu's cyber bet, and idle GPUs

Zhipu hired credibility. Yuki Chen is a five-time top researcher on Microsoft's security-response leaderboard and former head of 360's vulnerability institute. He joined this week. InfoQ China says Microsoft's latest patch round credited 23 bugs to "Yuki Chen using GLM from z.ai". One scored 9.8 of 10, though InfoQ and Zhidx name different software. Zhidx says about 100 security firms now plug GLM in.

I was half wrong yesterday. I said nobody aimed at Zhipu in Anthropic's worm paper, and that still holds. But InfoQ's recap says NIST's evaluation centre rated GLM-5.3 the strongest open-weight cyber model on 17 September, roughly four months behind US frontier. Anthropic's 29 September report had it completing 50 of 410 exploit attempts against 56 for its Mythos Preview. Evaluators keep testing Zhipu because cyber is what it sells. That is the incident class a Shenzhen hotline exists for.

Utilisation is the shortage. One lab scheduling specialist told Leiphone that industry GPU utilisation runs 25–27%, against 40% for good infrastructure teams. Top infra engineers now earn 7–8m yuan (~$1.0–1.1m), though one founder says only 5–10% capture that premium. InfoQ China adds that past a few thousand cards, communication and fault-finding decide real compute.

Ant's Ling-3.1-flash attacks the other side. It is open weights at 560B total and 25B active parameters, with 1M context. Quantising routed experts to MXFP4 cuts weights from about 534GB to 292GB and serving from eight Hopper cards to four. Artificial Analysis scores it 41, at about 210 tokens a second and roughly $0.19 per million tokens, InfoQ China reports. Jensen Huang told podcaster Lex Fridman that scarcity forces Chinese labs toward efficient architectures. Ling is the receipt.

DeepSeek's API went dark for five hours. It was down from 8:11 to 13:04 Beijing time. The notice promises credits equal to seven days' spend and to strip out "single-point dependencies", per V2EX. That lands in the week of a reported 100bn yuan (~$14bn) round.

Go deeper on this section: ClaudeChatGPT

Harbin's informer, Luckin's algorithm

Twenty students were punished for VPNs. China Digital Times reports Harbin Engineering University disciplined seven undergraduates and 13 graduate students after a two-day sweep. Unverified rumour says a classmate informed to improve her graduate-admission ranking. She reportedly admitted using a VPN herself for papers and GPT. Two of five archived commentaries were later deleted.

My read: after yesterday's Pei Yi and Kevin Tsai cases, Harbin shows the incentive underneath. Informing pays, and the rule enforced is one academic research can't obey.

Luckin's rota is an algorithm. Initium's podcast says AI schedules staff from sales, orders and hours. Managers say it keeps compressing headcount. Unpaid overtime covers what the system can't see: training, cleaning, stock counts.

Go deeper on this section: ClaudeChatGPT

Threads we are pulling

  • Chabaidao went from halt to index exit. Following yesterday's SFC freeze, Hang Seng Indexes will drop the tea chain after the close on 22 October at a nominal HK$0.0001, per 36Kr's digest. Nobody at the company has been accused.
  • The bubble warning got a price. After yesterday's STAR 50 slide and the chip-equipment chairman's caution, the ChiNext index fell 3.1% on 14 October, with AI-compute names down 5–7%. Robot maker Unitree opened at 1,100 yuan on 19 August and now trades near 420. Its market value has fallen from about 444.9bn yuan (~$63bn) to under 170bn (~$24bn), TMTPost reports.
  • Orbbec shows what the froth hid. The 3D-sensor maker trades at 278 times earnings. Insiders sold about 780m yuan (~$110m) in June and July. Its filing forecasts 2026 related-party sales to Ant Group down 68% to 80m yuan (~$11m), TMTPost reports.
  • The Claude ban thread. A V2EX developer says a year-old Max account was banned and calls the switch to Codex withdrawal. Sentiment, not data.

Go deeper on this section: ClaudeChatGPT

The river · 34

Huawei details its chip-stacking approach and next-generation roadmap

# Zhidx 3h

Huawei disclosed new details of its Kirin 9050 design and chip roadmap, presenting dense hybrid bonding and logic folding as a route to improve performance without relying solely on planar scaling.

Also: Leiphone

More
  • Huawei says the Kirin 9050 Pro uses more than 50 million hybrid bonds and more than five million effective signal connections, with a reported 1.5-micron bonding pitch.
  • The company describes the design as arranging related logic across layers to shorten signal paths; it reported a 53.5% increase in transistor density but did not provide an equivalent process-node claim.
  • The briefing also covered a 30-billion-parameter on-device MoE model, which Huawei says can run with less than 2.5 GB of additional memory through expert placement across DRAM and flash.
  • Huawei said the next Kirin generation has entered testing and results so far meet expectations; the report provides no independent test data.
  • This is a rare detailed look at how Huawei is positioning design, packaging and system optimization as alternatives to straightforward process scaling.
Open at Zhidx Ask Claude ChatGPT

VPN crackdown at Harbin university sparks debate over informant culture

# China Digital Times 10h

Chinese coverage centers on more than 20 students disciplined for unauthorized VPN use and the online backlash over alleged peer reporting, linking censorship enforcement to academic pressure and social trust.

More
  • Harbin Engineering University said seven undergraduates and 13 graduate students were disciplined for unauthorized VPN use after a two-day enforcement period.
  • Online discussion reportedly focused on a rumor that a student informed on a rival to improve her graduate-admission prospects; the university’s statement did not establish that motive.
  • Archived Chinese commentary argues that reporting systems can be weaponized for personal gain, chilling online speech and eroding trust; this is public reaction, not a verified account of the alleged informer.
  • The episode exposes a practical contradiction for researchers: access to overseas sites may be academically useful while unauthorized circumvention remains punishable.
  • The story’s value lies in the domestic argument about enforcement, competition and social trust, not simply in the school’s disciplinary notice.
Open at China Digital Times Ask Claude ChatGPT

Shanghai lab says its model helped prove a difficult mathematics conjecture

# Leiphone 8h

A human-mathematics team says Shanghai AI Lab’s Intern-S2 helped prove the three-dimensional Ghys conjecture, with a separate formalization effort checking the proof.

More
  • The team says the model produced an initial 400-page derivation and later helped generate more than a thousand pages of proof work across seven rounds of exploration.
  • Mathematicians directed the search, selected promising paths and corrected the system; the report presents the achievement as human-AI collaboration, not autonomous theorem proving.
  • A separate team used the SHEAF framework to formalize and verify the proof, an important distinction between plausible generated reasoning and machine-checkable correctness.
  • The work was posted as a preprint and its formalization code was made public, creating a basis for outside scrutiny.
  • If independently validated, the result would be notable less as a chatbot milestone than as an example of a model-harness workflow producing research candidates that mathematicians can verify.
Open at Leiphone Ask Claude ChatGPT

China and the US discuss an AI incident hotline ahead of a summit

# SCMP 6h

Chinese and US officials are discussing a basic framework for AI-incident response and a possible hotline ahead of a leaders’ summit, making risk communication part of bilateral diplomacy.

More
  • The talks reportedly cover how to handle incidents involving AI, with both sides seeking a process before the planned summit in Shenzhen.
  • A hotline could reduce uncertainty during an incident, but it would not itself establish shared safety standards or bind companies to common controls.
  • The report places the discussions alongside risks from autonomous agents, cybersecurity and other advanced capabilities.
  • The political significance is substantial: even amid competition over chips and models, both governments see some value in keeping channels open.
  • The article describes an effort under discussion, not an agreed mechanism; scope, participants and escalation rules remain unclear.
Open at SCMP Ask Claude ChatGPT

Step 5 Preview ranks second in OpenRouter usage after launch

# AI Zhidx 2h

Chinese coverage emphasizes StepFun’s rapid adoption as well as its newly opened weights, reporting about 188 trillion tokens of usage over seven days on OpenRouter.

Also: InfoQ China, IT Home

More
  • The report says Step 5 Preview ranked second on OpenRouter’s weekly and daily open-model usage lists as of October 15.
  • StepFun reports 600 billion total parameters, 27 billion active per token and a 1-million-token context; the article also cites an Artificial Analysis intelligence score of 44.
  • A comparison in the article puts a Step 5 Preview task at $0.71 versus $5.86 for Claude Opus 5, based on the cited evaluation; this is a benchmark-specific cost estimate, not a general serving-cost guarantee.
  • The article explains the model’s deeper 92-layer architecture and sparse attention as choices intended to support long-horizon agent work.
  • High platform usage is useful adoption evidence, but does not by itself establish reliable task completion or enterprise production use.
Open at Zhidx Ask Claude ChatGPT

ByteDance wins a $32 million project to build Hengdian’s film AI platform

# AI Zhidx 1d

A ByteDance-linked consortium won a 230-million-yuan (about $32 million) contract to build an AI platform for China’s Hengdian film industry, spanning data, models and production tools.

More
  • The project includes film-specific systems for script analysis, storyboards, digital characters, voice generation, costume and set design, location selection and post-production.
  • The tender requires at least nine industry datasets, including at least 50 billion pretraining tokens, 1 million instruction-tuning examples and 30,000 preference pairs.
  • It also calls for a trusted data space to manage authorization, rights, exchange and use of film assets, bringing copyright governance into the core platform design.
  • The three-year public contract is a significant test of whether a regional film hub can turn its production assets into a governed AI dataset and workflow stack.
  • The procurement specifies availability and data-processing requirements, but those requirements are targets rather than demonstrated system performance.
Open at Zhidx Ask Claude ChatGPT

Qwen Office launches an agent tailored to e-commerce operations

# AI QbitAI 14h

Alibaba’s Qwen Office introduced an e-commerce mode that connects merchant data and tools for research, content generation, store monitoring and ad optimization.

Also: Zhidx

More
  • The product is designed to connect to existing seller tools and analyze authorized business data rather than require merchants to move all their systems.
  • Its functions cover trend research, product selling points, batch creative generation, multi-platform performance dashboards and return-on-investment alerts.
  • The report says a merchant used it to research a 3D-printed shoe shop and reduce launch preparation from about 30 days to seven, spending hundreds of yuan on tokens instead of several thousand yuan on services and production.
  • That example is an early case study, not a controlled assessment of sales lift or model accuracy.
  • The strategic play is an agent layer over fragmented commerce software: the product’s value depends on reliable connectors, permissions and domain-specific recommendations.
Open at QbitAI Ask Claude ChatGPT

China’s September trade surplus reaches $133.5 billion as exports surge

# Signal Sinocism 17h

Chinese coverage foregrounds the export-and-import surge and Premier Li Qiang’s claim that China is both a major seller and buyer, while the newsletter also flags prices and credit data.

More
  • Sinocism reports exports up 36.3% year over year in dollar terms and imports up 31.0%, with a monthly surplus of $133.53 billion, the second largest on record.
  • The newsletter notes exports far exceeded the 25.5% consensus cited in its account, making the scale of the trade result more striking than the headline surplus alone.
  • Li’s Canton Fair remarks stress China’s role as a buyer as well as a seller, a framing aimed at countering foreign concern about export competition.
  • The issue also covers September prices and credit, but the supplied excerpt does not include those figures; the trade data are newsletter-reported and not independently verified here.
Open at Sinocism Ask Claude ChatGPT

Apple Intelligence will enter China testing with Alibaba’s Qwen model

# Policy Leiphone 6h

Tim Cook said Apple Intelligence features using Alibaba’s Qwen have approval to begin testing in China later this month, while Siri’s AI overhaul is not included.

More
  • Cook said the initial test includes live translation, writing tools and photo cleanup, but not an AI-enabled Siri.
  • The arrangement illustrates how foreign AI features entering China can depend on local model partners and regulatory approval.
  • The report says the features have been approved, but does not specify the review process, technical division of labor or data-handling terms.
  • For Apple, a limited release is a way to restart its AI offering in China without yet bringing its full assistant experience to the market.
Open at Leiphone Ask Claude ChatGPT

Top security researcher Yuki Chen joins Zhipu

# AI Zhidx 13h

Chinese coverage highlights Yuki Chen’s move to Zhipu after he used GLM in research credited with 23 Microsoft vulnerabilities, framing cybersecurity as a major application frontier for domestic models.

Also: InfoQ China

More
  • The report identifies Chen as a former 360 security executive and a repeat top-ranked Microsoft vulnerability researcher.
  • It says the 23 vulnerabilities received Microsoft Security Response Center acknowledgments, including a critical OpenSSH issue; that is the researcher’s and company’s attribution of the work.
  • Zhipu says about 100 security companies have integrated GLM into products or workflows, but the article provides no revenue or customer-use breakdown.
  • The hire brings human security expertise alongside model capability; it should not be read as proof that a model can independently discover and validate vulnerabilities.
  • The story reflects a wider shift in China’s AI competition toward cybersecurity, where model capabilities can support both defense and offensive research.
Open at Zhidx Ask Claude ChatGPT

151 apps reject Doubao phone assistant’s GUI automation

# Policy Leiphone 5h

Chinese coverage emphasizes the scale of domestic apps refusing ByteDance’s screen-control assistant, including WeChat, Taobao, Meituan and DeepSeek, as the debate shifts to who controls cross-app agents.

Also: 36Kr

More
  • The report says 151 apps opted out of Doubao’s proposed GUI automation, which reads screens and simulates taps and swipes; users can still open the apps manually.
  • The opt-out list spans social, commerce and service apps, limiting the assistant’s ability to complete tasks across China’s major platforms.
  • The dispute is about permissions and business control as much as security: app operators are deciding whether an outside agent may act through their interfaces.
  • The same roundup describes Tencent launching a free short-drama mini-program with a social recommendation feature, another example of Chinese platforms competing for attention inside existing ecosystems.
  • These are separate items bundled in a news roundup; the first is the consequential access-control story.
Open at Leiphone Ask Claude ChatGPT

China’s GPU bottleneck is shifting from chip supply to system efficiency

# Analysis InfoQ China 0m

As domestic accelerators become more available, Chinese infrastructure builders say cluster networking, scheduling, software adaptation and fault diagnosis now determine how much theoretical compute is usable.

More
  • The account describes a move from the earlier question of whether firms could obtain accelerators to whether thousands of mixed-vendor cards can run reliably at scale.
  • Model migration involves more than matching framework interfaces: operators, collective-communication libraries, memory management, monitoring and failure diagnosis may all need retuning.
  • A mismatch in compute and communication timing can leave accelerators idle; small delays compound across large clusters and repeated training iterations.
  • Heterogeneous deployments add integration costs because GPU and NPU vendors expose different software stacks, metrics and error conditions.
  • The strategic implication is that domestic compute competition increasingly depends on system integration and operations, not just peak chip specifications.
Open at InfoQ China Ask Claude ChatGPT

Kunlun Tech opens a single-GPU video-model game engine

# AI Leiphone 10h

Kunlun Tech released Matrix-Game 4.0, combining a conventional game engine that tracks state and rules with a video model that renders the scene in real time.

More
  • The company reports 480p at 60 frames per second on one RTX 5090, under 150 milliseconds of input-to-frame latency and an 18.5 GiB peak memory footprint.
  • The engine maintains explicit state such as object locations, health and inventory, while the video model generates the visual surface; this avoids asking a video model to remember every game rule.
  • A coding agent can create or alter game logic, while structured scene renderings constrain the video model’s output.
  • The approach is a useful architectural alternative to fully generative worlds: deterministic state and probabilistic rendering are assigned different jobs.
  • The performance figures and claim of being first are company-reported; broader gameplay consistency and hardware portability remain to be tested.
Open at Leiphone Ask Claude ChatGPT

Ant’s Ling 3.1 Flash is tested on office work and coding tasks

# AI Zhidx 10h

A Chinese newsroom reports that Ling 3.1 Flash completed document analysis, interactive dashboards and small software projects in its own tests, extending coverage of Ant’s open-weight release.

Also: InfoQ China

More
  • The newsroom says it used the model to summarize financial filings, organize work tasks and build interactive web pages and prototypes.
  • The tests are demonstrations by the outlet, not a controlled comparison against other models or evidence of reliable production use.
  • Ling 3.1 Flash is described as a 560-billion-parameter MoE with 25 billion active parameters and a 1-million-token context.
  • Its useful signal is breadth across office and coding workflows, but factual accuracy, source attribution and recovery from tool failures need independent evaluation.
Open at Zhidx Ask Claude ChatGPT

A Chinese robotics team tries to make humanoids manipulate while moving

# AI GeekPark 5h

YuanCe Future’s whole-body intelligence system pairs a motion-control model with a higher-level task model to address the gap between walking to a task and manipulating while balancing.

More
  • The company describes an S0 motion-control model and an S1 task-planning model, trained and evaluated as a coordinated system.
  • In its tests, YuanCe says S0 reduced hand-end tracking error from about 10 centimeters in its cited baseline to about 2 centimeters and added control on uneven terrain.
  • Demonstrations include lifting a fallen bicycle, pressing a rice-cooker button and moving a step stool to reach a higher object.
  • The system’s data strategy matters: better whole-body control can make human demonstrations more faithfully transferable to the robot, potentially improving the training data available for manipulation.
  • These are company demonstrations and benchmark claims; the article does not report sustained work rates or performance outside the showcased tasks.
Open at GeekPark Ask Claude ChatGPT

China opens a long-delayed bridge with North Korea to boost trade

# Policy SCMP 6h

Premier Li Qiang framed the New Yalu River Bridge opening as a way to deepen trade and exchanges with North Korea, signaling Beijing’s interest in activating a long-delayed cross-border link.

More
  • Li called the bridge a landmark in bilateral friendship and said it would expand trade and people-to-people contact, according to Xinhua.
  • The opening is a concrete infrastructure step in a relationship that has often been constrained by sanctions and political tensions.
  • The bridge’s strategic value will depend on whether traffic and commerce actually increase; the supplied report gives no operating or trade-volume data.
  • The state-media framing foregrounds friendship and connectivity, while the economic impact remains to be demonstrated.
Open at SCMP Ask Claude ChatGPT

Chinese interconnect-chip maker lays out a bid to challenge Broadcom

# Analysis Zhidx 4h

Wuxi-based Zhongxing Micro presented a portfolio spanning storage, networking, PCIe switching and AI-cluster links, with more than 800,000 units shipped according to the company.

More
  • The company says it has 12 products in volume production and has shipped into systems from Inspur, Lenovo, H3C and ZTE.
  • Its positioning reflects a systems-level bottleneck: cluster throughput depends on communication and I/O, not only accelerator peak performance.
  • The company claims its PCIe switch can substitute for Broadcom’s PEX89104 and reports a 47% write-bandwidth advantage in one 16-drive RAID5 test; these are company-reported results.
  • The announced partnerships with hardware and software vendors highlight that ecosystem adaptation is central to domestic interconnect adoption.
Open at Zhidx Ask Claude ChatGPT

AI infrastructure engineers are commanding sharply higher pay in China

# Signal Leiphone 8h

Chinese recruiters and executives say demand for AI infrastructure specialists has raised salaries, as model firms shift attention from benchmark gains to GPU utilization and serving costs.

More
  • The report cites recruiter examples of salaries rising from 300,000 yuan (about $42,000) to 1 million yuan (about $140,000), and claims some senior packages reach 7–8 million yuan (about $1.0–1.1 million).
  • The skills in demand include kernel optimization, inference scheduling, expert parallelism and improving overlap between compute and communication.
  • A source in the article estimates typical GPU utilization at 25–27% and says strong teams can reach 40%; the report does not define the workload or measurement method.
  • The labor-market shift reflects a broader change in Chinese AI economics: improving utilization can have a more direct effect on margins than another small benchmark gain.
Open at Leiphone Ask Claude ChatGPT

Chinese tech roundup flags an IPO fraud probe and a new AI safety unit

# Policy Leiphone 1d

The roundup leads with Hong Kong regulators suspending a bubble-tea chain’s shares over suspected IPO manipulation, and reports ByteDance Seed creating a dedicated model safety and alignment team.

More
  • The Hong Kong Securities and Futures Commission said suspicious groups may have created a false impression of demand during ChaPanda’s IPO and restricted dealings in up to HK$4.379 billion of assets.
  • The regulator’s account raises concern that IPO proceeds may have been routed through undisclosed arrangements; the investigation is ongoing and the company had not responded in the cited report.
  • The roundup says ByteDance Seed created a separate model-alignment and preparedness group to focus on agent safety and risks from advanced capabilities.
  • It also reports that Huawei’s next Kirin chip is in testing, but the roundup provides no new technical detail beyond the company’s claim that results meet expectations.
  • This is a multi-story digest; the regulatory action and the formal AI safety unit are distinct developments, not one event.
Open at Leiphone Ask Claude ChatGPT

Douyin e-commerce sues a JD employee over alleged platform rumors

# Policy Leiphone 7h

A Beijing court accepted Douyin e-commerce’s case against a JD employee accused of spreading false claims about its conduct during the 618 shopping festival.

More
  • The dispute centers on a social-media post alleging that a rival platform promoted negative coverage of JD; the report says the cited article had not appeared on the platform’s trending list.
  • The account describes a wider wave of similar posts with concentrated timing and wording, which it says showed signs of coordinated promotion.
  • The underlying allegations concern platform competition and claims that merchants are pressured to offer the lowest prices, bringing both reputation law and platform conduct into the dispute.
  • The court has accepted the case, but no finding on liability has been reported.
Open at Leiphone Ask Claude ChatGPT

Embryo researcher returns from the US to lead a lab in China

# Analysis SCMP 2h

Stem-cell researcher Wu Jun has returned to China, where he says many leading embryo researchers now work, reflecting the country’s growing pull in a sensitive frontier field.

More
  • Wu is based at Westlake University in Hangzhou and is known for research involving stem-cell-derived embryo models.
  • His return is a talent-flow signal, but the supplied excerpt does not give details of his lab’s funding, research program or the scale of the broader migration.
  • Embryo research raises ethical and regulatory questions that differ across countries; the report’s framing emphasizes China’s scientific capacity, not those governance limits.
Open at SCMP Ask Claude ChatGPT

SK Hynix seeks US approval for more equipment at its China plant

# Policy IT Home 1h

Reuters reports that SK Hynix is seeking permission to install more US-made tools in Dalian, while considering a US factory as part of its pitch to Washington.

More
  • The company has applied for licenses for additional equipment in an unfinished area of its Dalian NAND facility; it says installation is being discussed with the US government and is not seeking to expand capacity there.
  • The proposal illustrates how export licensing now shapes routine operations at foreign-owned semiconductor plants in China, not only new investment decisions.
  • The reported strategy links possible US manufacturing commitments to permission for equipment used at an existing China site, placing corporate investment choices inside bilateral industrial policy.
  • The account is based on unnamed sources; SK Hynix says it is not seeking to expand the Dalian facility’s capacity.
Open at IT Home Ask Claude ChatGPT

Hunan police warn rural men about cross-border marriage scams

# Culture SCMP 7h

Police in Hunan warned that paid international matchmaking can expose rural men to financial loss and legal trouble, highlighting the risks around a tightly regulated marriage-broker market.

More
  • The police post warned that some cross-border arrangements lack genuine emotional foundations and may involve illegal agencies charging high fees.
  • For-profit international matchmaking is illegal in China, according to the report.
  • The warning frames marriage as a consumer-protection and public-order issue, while the underlying concern also reflects demographic and gender imbalances that can shape demand for such services.
  • The report does not provide data on prevalence or prosecutions.
Open at SCMP Ask Claude ChatGPT

Chinese drugmaker puts its president in charge of AI adoption

# Analysis 36Kr 12h

Simcere’s president took direct control of its information department and is embedding AI staff in business units, treating organizational change as a prerequisite for AI-assisted drug discovery.

More
  • The company says AI use is mandatory for managers and is being introduced from individual workflows into target discovery, clinical development, manufacturing and commercial operations.
  • Simcere’s president says early employees wanted to use AI but did not know how, prompting the company to embed technical staff in business units.
  • The drugmaker is using AI in early molecular design and target discovery and has partnered with Schrödinger on drug discovery and development.
  • The strategy links AI investment to a shift from generics toward innovative medicines and to the goal of generating more globally differentiated drug candidates.
  • The article provides a detailed organizational account but no measured evidence yet that AI has improved discovery success, clinical outcomes or time to market.
Open at 36Kr Ask Claude ChatGPT

Three Chinese agencies open a joint probe into a car brake-pedal failure

# Policy IT Home 1h

China’s industry, cyberspace and market regulators formed a joint investigation team after reports of a broken brake-pedal support in a high-profile domestic vehicle.

More
  • The joint inquiry involves the Ministry of Industry and Information Technology, the Cyberspace Administration and the market regulator, signaling attention across product safety and online reporting.
  • The manufacturer said it would cooperate; the agencies said they would release information as the inquiry proceeds.
  • The case is a visible test of how Beijing coordinates vehicle-safety oversight as domestic automakers compete on software, electrification and brand trust.
Open at IT Home Ask Claude ChatGPT

An AI service matches strangers for dinner, then steps out of the conversation

# Culture 36Kr 10h

Chinese startup Table for Six uses AI to match six people for in-person drinks while automating the logistics behind events across 14 cities.

More
  • The company says more than 40,000 users have paid for events, and it has raised several million dollars; these figures are reported by the company and its investors.
  • AI parses open-ended answers to balance interests, age, gender and social energy, while event cards structure the in-person conversation.
  • The product keeps AI out of the social interaction itself but uses agents behind the scenes for matching, venue selection, substitutions and printed materials.
  • The model addresses a real operations problem: informal dinners are hard to standardize and scale, but the business’s ceiling and repeat-use economics remain unproven.
  • The service offers a glimpse of Chinese urban consumers paying for structured offline social connection rather than another AI companion.
Open at 36Kr Ask Claude ChatGPT

Orbbec’s robotics revenue remains hard to separate from its broader business

# Analysis TMTPost 5h

A Chinese financial investigation questions whether Orbbec’s robotics growth justifies its valuation, citing opaque segment reporting, reliance on nonrecurring gains and customer concentration.

More
  • The report says Orbbec ranked first in global robot 3D-vision sensing revenue in 2025, with a 29% share, but the company does not break out revenue by end market.
  • It cites 2026 first-half revenue of 438 million yuan (about $62 million), up 0.5%, and net profit of 41.6 million yuan (about $5.9 million), down 31.7%; more than 40% of profit reportedly came from nonrecurring sources.
  • The report also describes a close relationship with Ant Group and questions customer concentration, while noting that robotics revenue and order volumes remain difficult to verify from public filings.
  • The larger lesson is a disclosure problem common to emerging robotics suppliers: market-share claims and customer lists do not show how much revenue is recurring or profitable.
  • The analysis cites a 278-times price-to-earnings ratio after a substantial share-price rise; that figure is a snapshot, not an independent valuation judgment.
Open at TMTPost Ask Claude ChatGPT

China and Russia discuss a possible leaders’ meeting at APEC

# Policy SCMP 7h

Wang Yi and Sergey Lavrov discussed bilateral ties and international crises, with the talks potentially paving the way for a meeting with the US president at APEC in Shenzhen.

More
  • The foreign ministers met in Baku and discussed relations with the United States, Ukraine and Iran, according to the Russian account cited.
  • A possible three-leader meeting would put China’s summit diplomacy at the center of a highly visible regional event.
  • The report does not say that such a meeting has been agreed or specify what issues would be on its agenda.
Open at SCMP Ask Claude ChatGPT

Jensen Huang says China’s AI firms are building under severe compute limits

# Analysis InfoQ China 0m

In an interview excerpt, Nvidia’s Jensen Huang says Chinese AI firms face tight compute constraints but are still raising money and pushing products forward.

More
  • Huang characterizes Chinese AI companies as operating under very significant compute constraints while continuing to build and finance their businesses.
  • The interview also recounts Nvidia’s early relationship with TSMC, emphasizing Huang’s view that clear limits and candor helped establish trust between partners.
  • That contrast is relevant to China’s AI strategy: compute restrictions are a constraint, but they may also increase pressure to improve efficiency per unit of hardware.
  • The article is a selected interview transcript and offers Huang’s judgment, not independent data on Chinese firms’ compute access or model efficiency.
Open at InfoQ China Ask Claude ChatGPT

China-led team reports a possible earthquake precursor called seismo-ripples

# Analysis SCMP 4h

A China-led team says combined seismic, magnetic and ionospheric signals preceded major earthquakes in its study, a provocative claim that still needs broader validation.

More
  • The researchers describe ring-like disturbances they call seismo-ripples and report an 83% success rate in anticipating earthquakes since 2022.
  • The claim challenges the long-standing view that precise earthquake prediction is not possible, so replication and clear definitions of success and failure are essential.
  • The article excerpt does not provide the sample size, false-alarm rate or independent replication needed to judge predictive value.
Open at SCMP Ask Claude ChatGPT

Deep Robotics returns to losses as quadrupeds still drive its IPO story

# Analysis TMTPost 5h

Deep Robotics’ updated filing shows revenue growth but a first-half loss, with quadruped robots still the main business and humanoid sales at an early validation stage.

More
  • The filing cited in the report shows 2026 first-half revenue of 259 million yuan (about $36 million) and a net loss of 8.8 million yuan (about $1.2 million).
  • The company’s margin fell as lower-priced products shipped in volume; its industrial quadruped model’s average price and gross margin both declined.
  • Humanoid robots accounted for only six units sold in the first half, while quadrupeds remained the principal revenue source.
  • The report highlights customer concentration and the risk that a large buyer developing its own products could become a competitor.
  • This filing is a useful counterweight to humanoid-robot demonstrations: commercialization, margins and customer dependence remain harder tests than prototypes.
Open at TMTPost Ask Claude ChatGPT

Jensen Huang says Nvidia should cooperate with Chinese AI firms on safety

# Policy IT Home 2h

Nvidia’s CEO called for dialogue with Chinese AI companies to monitor frontier risks, against a backdrop of US-China discussions about an AI incident hotline.

More
  • Huang said cooperation should happen through dialogue and described regular exchanges with Chinese researchers.
  • The proposed focus is monitoring risks from advanced AI, not a commercial partnership or agreement on shared technical standards.
  • This matters because AI safety is increasingly part of bilateral diplomacy even as the two countries compete over chips, models and deployment.
  • The article also references a US proposal for channels covering agent failures, cybersecurity and biological risks; it does not report that a formal mechanism has been agreed.
Open at IT Home Ask Claude ChatGPT

Former EU diplomat discusses protecting industry without a trade war

# Analysis The Wire China 1d

Gunnar Wiegand discusses Europe’s industrial defenses, a tougher German posture toward Beijing and the challenge of maintaining European unity on China.

More
  • The interview is explicitly about Europe’s approach to China, making this relevant as governments debate industrial protection and trade exposure.
  • Its framing centers on the tradeoff between protecting European industry and avoiding escalation into a broader trade conflict.
  • The supplied excerpt contains no specific policy proposal or new decision, so the value is perspective rather than breaking news.
Open at The Wire China Ask Claude ChatGPT

Chinese startup aims to make six-axis camera motion accessible to small crews

# Analysis 36Kr 12h

Yueqian Innovation is developing a lighter camera robot that uses mechanical balancing to reduce motor demands and bring repeatable motion paths to smaller production teams.

More
  • The company’s Cilin X keeps six degrees of freedom but uses a balanced mechanical design so motors mainly move the camera rather than continuously support its load.
  • The company says this can reduce required joint torque from about 100 newton-meters to under 10, while limiting vibration and backlash from high-ratio gearing.
  • Its 3+3 arrangement separates camera position from orientation and is intended to reduce singularities that disrupt traditional six-axis motion.
  • The product targets independent creators and small studios priced out of large camera rigs, but the article reports no independent footage-quality or reliability tests.
  • The longer-term ambition is to use camera movement and composition data to build a visual-expression model; that remains a roadmap, not a shipped capability.
Open at 36Kr Ask Claude ChatGPT
Source ledger
ChinAI 403 0/0 HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk 200 0/20
Pekingnology 200 0/20
Ginger River Review 200 0/20
Sinocism 200 1/20
QbitAI 200 1/10
36Kr 200 7/30
Solidot 200 12/20
Leiphone 200 17/20
GeekPark 200 2/30
IT Home 200 12/60 +31 over cap
SCMP 200 10/50 +10 over cap
Initium 200 3/15
Jiemian 200 8/30 +14 over cap
China Digital Times 200 1/15
The Wire China 200 1/50
Techmeme 200 4/15
Hacker News 200 0/20
V2EX 200 8/50 +35 over cap
Zhidx 200 10/20 +2 over cap
InfoQ China 200 8/20 +12 over cap
New Wisdom 200 8/15 +2 over cap
TMTPost 200 8/19 +11 over cap