China Desk

China News, Summarized 23 Oct 2026, 08:33 UTC 34 stories 22/23 sources

Today's brief

China is closing ranks: Xi around one core, industry around DeepSeek

Xi turns the Long March into a loyalty lesson

Xi wants one core. Marking the Long March's 90th anniversary on Thursday, Xi said the Red Army survived by rallying around Mao Zedong and that China now faces a "new Long March". It came four days before the fifth plenum, the Central Committee's closed-door meeting, set for 26 to 29 October.

The line aimed inward. Seoul's Kyunghyang Shinmun notes that Xi blamed "leftist dogmatism" for nearly sinking the Party before Mao took command in 1935. That passage was absent from the 80th-anniversary speech a decade ago. The paper reads it as aimed at the Party itself. My read: divided leadership nearly ended it, and one core saved it.

Huang Qifan is the context. On 13 October the Central Commission for Discipline Inspection, the Party's anti-graft body, opened a probe into Huang, 74, Chongqing's mayor from 2010 to 2016. He had outlasted the falls of his superiors Bo Xilai and Sun Zhengcai.

The count is the point. SCMP's tally, relayed by ua.news, is that 35 of the Central Committee's 205 full members have been investigated or lost their legislative mandates. On 17 October I argued that surviving one purge protects nobody from the next. A retired mayor fits that.

THE TELL: state broadcaster CGTN's English summary stresses "loyalty, truth, justice and unity" and the road to a modern socialist China. In the version I read, it never mentions leftism or a political core. SCMP and the Korean paper lead on both. No mainland outlet in today's feed covered the speech.

The AI treaty reaches the legislature. The National People's Congress Standing Committee meets 2 to 6 November and will consider ratifying the agreement creating the World AI Cooperation Organization. Twenty-nine countries signed it in Shanghai on 16 July, Russia, Brazil and Cuba among them. It takes effect once a third state deposits ratification. Beijing is building its own AI institution while Washington sanctions its AI investors.

Go deeper on this section: ClaudeChatGPT

DeepSeek's investors are industrial China

Nobody is on the record. DeepSeek has raised about 100 billion yuan (~$14 billion), Bloomberg reported, which prints $15 billion. The target rose twice, from 50 billion yuan (~$7 billion) to 80 billion (~$11 billion), then settled higher. Reuters says the round is still closing and figures could change. No report carries a comment from DeepSeek.

Suppliers and customers, not funds. CATL, the battery giant, and Tencent are among the biggest investors. Quartz, citing the Wall Street Journal, adds carmaker Geely and Lenovo-backed Legend Capital as newcomers, with a valuation near $74 billion going in. Reuters says many investors are state-backed or corporate. My read: this is a national-champion round, not a venture bet.

CATL is paying while bleeding. On 20 October I flagged that CATL's profit lagged its sales. TMTPost reports the shares then fell over 12% on 21 October, the biggest one-day drop since listing, to 227.9 yuan (~$32). Third-quarter gross margin slid to 20.69%, and nine-month operating cash flow fell 23.6%. It has also put over 10 billion yuan (~$1.4 billion) into AI-energy plays.

THE NUMBER: $1 billion. That is DeepSeek's annualised revenue by 24 September, per China Daily's account of The Information's report, more than double a few months earlier. Liang Wenfeng, the former hedge-fund manager who founded DeepSeek, told investors price rises hadn't driven customers away. On my arithmetic, the valuation is about 74 times that. Pricing power is what the money is buying.

Chinese coverage is a rewrite. Zhidx, the tech site, relays Bloomberg as China's largest AI financing and ends by saying products and an IPO must prove it. It omits the state-backed investors Reuters flagged. It does add that Moonshot AI is reportedly eyeing a Hong Kong listing in early 2027, while DeepSeek has hired CITIC Securities for a Shanghai listing.

Go deeper on this section: ClaudeChatGPT

Narrow models, hard kernels, robot faith

Two decision models in one day. On Thursday Ant Group's enterprise-technology unit released LingDT-Decision-1.0 at 4 billion and 27 billion parameters. China Telecom's AI lab open-sourced Xing4.0-29B-Intent. Both pick from a fixed list of options instead of writing an answer, targeting high-volume routing that general models handle expensively.

Self-reported, but specific. Ant says token use falls over 90%, and customer-service risk detection rises from 76% to 89% versus Qwen-3.7-flash. China Telecom scores 58 against 49 for the best of 18 general models on 6,000 public-dataset questions, at under 200 milliseconds. Ant adds ordered decisions expressed as a dependency graph.

WHY IT MATTERS: generation is the wrong tool for routing. A refund, a renewal cancellation and an invoice request arrive in one sentence, and the answer set is closed. Ant concedes that spotting dependencies is not executing them: money and permissions still need business rules.

Qwen and Kimi share one attention recipe. At AICon Shenzhen a Qwen engineer laid out FlashQLA, a GPU kernel library for linear attention. Qwen runs three linear-attention layers to each softmax layer. Kimi K3 uses the same three-to-one split with its own variant.

The catch is occupancy. The recurrence allows one thread block per head, so a 32-head model leaves most of a GPU's 100-plus multiprocessors idle. The fix is sequence parallelism with a correction matrix, written as hand-built kernels because Triton can't schedule it.

Unitree's founder says he hasn't believed enough. Wang Xingxing spent 36 minutes at Unitree's Hangzhou conference urging the room to trust AI. Since August he has cut his estimate for robotics' ChatGPT moment from two-to-ten years to one-to-three. He showed GPT-6 driving a robot untrained. GeekPark's reading: if rivals supply the brains, the winner builds bodies cheapest. He wants 10 million units a year, not tens of thousands.

Belief carries a multiple. On 20 October Unitree itself warned of a trailing price-to-earnings ratio of 388. A chairman asking for more faith is also defending that number. That is my inference.

Go deeper on this section: ClaudeChatGPT

China's helium price crashed behind a wall

Helium fell 80% at home. Domestic high-purity helium peaked near 497.5 yuan (~$70) per cubic metre in April, after the Middle East conflict hit Qatar's output. It now sells for 61 to 105 yuan (~$9 to $15), while overseas prices still climb. Beijing banned helium exports on 10 July. Jiemian gives that equal billing with record Russian imports and new domestic plants.

Who it protects. China imported about 84% of its helium last year. My read: the wall insulates buyers who can't stockpile, since liquid helium boils away within roughly 45 days.

Go deeper on this section: ClaudeChatGPT

Threads we are pulling

  • The export draft is a ledger. On 20 October I read Beijing's draft as a registry more than a gate. ChinaTalk's three-day reading agrees. It counts 40 additions across AI, solar and biomedicine, 14 banned and 26 licensed. Interim registration applies even to exports already under way, which it reads as Beijing building a dataset before deciding what to license.
  • AI short drama got cheaper, not richer. On 20 October I said margin moves to rights and the shelf. GeekPark's interview with Huang Haonan, founder of studio Soy Sauce Culture, adds numbers. Cost per minute fell from 1,000 yuan (~$141) to 400 (~$56). Monthly output rose from 40–50 titles to 200–300, yet revenue stayed near 20 million yuan (~$2.8 million). Taste is a third owner: capacity scales, judgement doesn't.
  • Brussels is not softening. SCMP reports that Brussels and Berlin feel emboldened and are drafting fresh restrictions, while Beijing bristles at the idea it was strong-armed. On 15 October I said planners would keep talking while tightening. The tightening is holding; the talking is the fragile part.
  • Memory keeps climbing, on one poster's word. A V2EX user says a pair of 32GB sticks costs nearly 6,000 yuan (~$845), and a friend paid 7,100 (~$1,000) days earlier. That is one anecdote, not data, but it fits the shipment falls I have tracked since 14 October.

Go deeper on this section: ClaudeChatGPT

The river · 34

Chinese commentary challenges sensational coverage of young Japanese skipping meals

# China Digital Times 5h

A translated Chinese essay argues that domestic outlets exaggerated a Japanese woman's meal-skipping into a story of national decline, exposing how negative foreign news can serve a rosy home narrative.

More
  • The essay contrasts a highly personal account with the underlying survey, which reportedly found lack of time and hunger were more common reasons for skipping meals than saving money.
  • It situates the coverage alongside other Chinese media narratives about hardship abroad, arguing that selective framing can inflate traffic and reinforce national self-confidence.
  • The piece itself is a critique of Chinese media standards, a useful counterpoint to the more triumphalist foreign-country coverage it describes.
  • The essay acknowledges that food prices are a real pressure for vulnerable people in Japan; its criticism is of how the story was generalized and politicized.
Open at China Digital Times Ask Claude ChatGPT

Unitree founder says robot makers must believe in rapidly advancing AI

# GeekPark 20h

At Unitree's tenth-anniversary event, founder Wang Xingxing argued that better foundation models will accelerate physical AI and make scalable robot manufacturing the decisive advantage.

More
  • Wang shortened his estimate for a household-robot breakthrough to roughly one to two years and said a robot might soon handle a simple home task from a spoken instruction; these are forecasts, not demonstrated capabilities.
  • He described a strategy that relies on outside multimodal models as well as internal development, shifting the company's differentiation toward producing reliable robot bodies at much larger volumes.
  • Unitree's proposed model-improvement loop combines literature search, control-code generation, simulation, real-robot testing and feedback, but the article offers no independent results showing this loop works at scale.
  • The gap between his ambitious timelines and today's showcased stunts is the central question: software progress must translate into robust perception, manipulation and safe operation in unfamiliar environments.
Open at GeekPark Ask Claude ChatGPT

Unitree founder sees a physical AI breakthrough within three years

# 36Kr 5h

At Unitree’s first ecosystem conference, Wang Xingxing argued that stronger general models could soon unlock robot skills, while acknowledging that real-world manipulation remains unreliable.

More
  • Wang said planning and navigation are advancing faster than manipulation, where robots still struggle to adapt to object positions, forces and failed actions.
  • Unitree is developing a self-evolving system that writes control code, trains in simulation, tests on real robots and feeds results back into evaluation.
  • The pitch marks a strategic expansion beyond Unitree’s established strengths in robot bodies, actuators and motion control toward models and training systems.
  • Wang described demonstrations using a general model for painting and Go as requiring no task-specific training; these are demonstrations, not evidence of robust household capability.
  • Whether the approach matters depends on measured transfer to unfamiliar environments, repeatable task success and the economics of manufacturing at much larger scale.
Open at 36Kr Ask Claude ChatGPT

Xi orders urgent rescue after a 7.0-magnitude Tibet earthquake kills 16

# SCMP 6h

Xi Jinping ordered search and rescue after a powerful earthquake in Tibet killed 16 and injured 65, with high altitude, cold weather and secondary hazards complicating response.

More
  • The report says Xi called for rescue and medical care to be prioritized and for operations to adapt to seismic activity and weather.
  • The focus on a presidential instruction reflects how disaster response is presented as a direct central-government responsibility, especially in remote regions.
  • The supplied account does not detail damage, displacement, local response capacity or the time needed to reach affected communities.
Open at SCMP Ask Claude ChatGPT

Chinese drone-show company wins a German customer for 7,000-drone fleets

# 36Kr 7h

Shenzhen firm DAMODA is pitching an automated drone-show system that combines fleet control with rapid deployment and return-to-dock, and says a German customer ordered about 7,000 aircraft.

More
  • The reported system spans flight control, synchronized routing, positioning and ground operations; its vehicle-based launch setup is said to prepare a show in about a minute.
  • The German customer reportedly ordered 800 drones, then 1,000 and later 5,000, while DAMODA says exports reached more than 50 countries and regions.
  • The strategic shift is from selling a spectacle to selling a repeatable operating system; customer economics depend on reliability, local service and reducing the labor needed to stage large shows.
Open at 36Kr Ask Claude ChatGPT

China tightens CCC certification for e-bikes, toys and children's ride-ons

# Jiemian 13m

China's market regulator is strengthening post-certification checks and traceability, responding to e-bike makers that swap components or leave ways to defeat speed limits.

Also: IT Home

More
  • The e-bike rules require each vehicle, battery and charger to be traceable as a matched set, with required components supplied together and vehicle documentation aligned with registration.
  • Regulators have found production-control failures and designs that leave room for speed-limit tampering; certification bodies have already suspended or withdrawn some certificates.
  • The changes also set boundaries for toy certification and require clearer age and hazard warnings, showing how China uses compulsory certification to move safety requirements into factory controls and retail packaging.
Open at Jiemian Ask Claude ChatGPT

iFlytek says its new models were trained on domestic chips

# AI Zhidx 18h

iFlytek unveiled a cloud-and-edge model family, a speech model and embodied-interaction work, highlighting domestic compute, long-context inference and a small-model-first deployment strategy.

Also: InfoQ China

More
  • The release includes a cloud model with a one-million-token context window and an open 4-billion-parameter edge model intended for devices such as PCs, cars and robots.
  • iFlytek says training a long-context workload on its domestic 10,000-chip platform fell from 4.5 months to 1.2 months after optimization; these are company-reported figures.
  • It also describes a four-layer KV-cache design, with reported cache hit rate rising from 63% to 95.7%, aimed at easing memory and latency bottlenecks in long-context inference.
  • The company’s proposed routing strategy assigns simple or sensitive work to the device, more complex work to the cloud, or splits tasks between them; it reports routing accuracy of 96.1%.
  • The release is strategic evidence of efforts to make an end-to-end domestic stack work, but the supplied benchmarks are vendor-reported and do not independently establish comparable quality or cost.
Open at Zhidx Ask Claude ChatGPT

China’s humanoid-robot boom is creating a market for human action data

# Analysis ChinaTalk 22h

A ChinaTalk essay describes the push to collect first-person recordings of human hand use for robot training, a labor-intensive data source now treated as strategic infrastructure.

More
  • The essay places the collection effort in a national push for embodied intelligence, citing reported robot companies and state-owned data centers as part of the growing ecosystem.
  • First-person video captures people’s hand movements and interactions with objects, offering varied demonstrations without requiring robots to generate every training example themselves.
  • Unlike online clicks, this data requires people to perform and record physical tasks, raising questions about collection labor, consent and how workers are compensated.
  • The essay says Chinese robotics firms are increasingly focused on hands because dexterity and coordinated manipulation remain difficult to reproduce.
  • The account is an essay-contest report and the excerpt is incomplete; its scale and industry figures should be treated as reported claims, not an audited census.
Open at ChinaTalk Ask Claude ChatGPT

Ant's LingDT model targets faster, cheaper enterprise decisions

# AI QbitAI 23h

The company says its 4B and 27B models outperform several decision-model benchmarks while supporting multi-select and ordered decisions for business workflows.

Also: Leiphone

More
  • LingDT is designed to choose among defined outcomes rather than generate prose, potentially reducing token use and downstream parsing for routing, risk and customer-service tasks.
  • The model's ordered-decision feature represents dependencies between actions, a useful fit for requests that require several operations in a particular sequence.
  • This is a second account of the same launch; its performance claims are based on Ant's reported evaluations, not independent testing.
Open at QbitAI Ask Claude ChatGPT

Xi uses Long March anniversary to stress the Party’s central leadership

# Policy Sinocism 11h

At a major anniversary ceremony, Xi Jinping made the Party’s correct leadership and the need for a central core the speech’s leading explanation of the Long March’s victory.

More
  • Sinocism notes that Xi’s speech was shorter than the 2016 anniversary address but gave more prominence to Party leadership as the source of historical success.
  • The framing turns a military-history commemoration into an argument for centralized political authority and discipline.
  • The newsletter contrasts the new speech’s emphasis with the 2016 version, which foregrounded ideals and faith.
  • The event included the top leadership and about 3,000 attendees, according to the newsletter.
  • For readers tracking political messaging, the notable signal is not a new policy but the way a canonical historical episode is being used to reinforce present-day governance claims.
Open at Sinocism Ask Claude ChatGPT

AI makes Chinese animated mini-dramas cheaper, but not more profitable

# Culture GeekPark 1d

Chinese studios say AI has slashed production costs and multiplied output, while revenue has barely moved; the scarce advantage is shifting from production labor to story judgment and audience insight.

More
  • One producer cited in the article says an 80-minute production fell from roughly 100,000 yuan (about $14,000) to 40,000 yuan (about $5,600), while a smaller team can now make several titles in a month.
  • More output has not translated into proportionate revenue: the producer reports monthly production rising several-fold while domestic revenue stayed around 20 million yuan (about $2.8 million).
  • The cost mix has flipped: compute and tokens now take a much larger share, while labor has fallen, showing how generative AI can move rather than eliminate production costs.
  • Platforms are courting individual creators and small studios, weakening the old role of large production houses as the main supply channel.
  • The account is based on industry interviews and figures, not a comprehensive market dataset; the durability of the creator shift and the economics of hit-making remain uncertain.
Open at GeekPark Ask Claude ChatGPT

EU weighs sanctions relief for Chinese firm accused of aiding Russia

# Policy SCMP 12h

The EU is considering lifting sanctions on Chinese textile maker Yusha Group in exchange for Beijing’s pledge to curb shipments to Russia and remove a German firm from its own export controls list.

More
  • The report describes an unusual reciprocal bargain: Brussels would ease restrictions on Yusha while China would constrain exports to Russian military-linked buyers.
  • The arrangement would make enforcement a subject of direct negotiation, rather than relying only on sanctions and unilateral controls.
  • It also links the case to Beijing’s own export-control dispute with Germany’s Sindlhauser Materials.
  • The account is based on sources familiar with the talks; the text does not establish that either side has finalized the deal.
Open at SCMP Ask Claude ChatGPT

China Telecom open-sources a model for intent parsing and extraction

# AI QbitAI 1d

China Telecom’s Xing4.0-29B-Intent model combines intent classification with parameter extraction, returning structured fields for downstream business systems rather than unconstrained prose.

More
  • The open release includes model weights, an inference image and 2.75 million training examples, according to the announcement.
  • Its schema-driven output is intended to identify both the action and its parameters, such as a room, device and requested setting in a voice-control command.
  • The model is already used in China Telecom customer service and smart-home products, the company says.
  • The supplied evaluation compares it with general-purpose models across selected public datasets; results do not by themselves demonstrate performance on every production workload.
Open at QbitAI Ask Claude ChatGPT

Ant launches a decision model for multi-step enterprise workflows

# AI Zhidx 19h

Ant Digital released LingDT-Decision-1.0 in 4B and 27B versions, designed to return structured decisions directly rather than generate text that downstream systems must parse.

More
  • The model targets high-volume tasks such as customer-service routing, risk screening and tool selection, where a constrained decision can be cheaper and more predictable than a long answer.
  • Ant says its native multi-select and dependency-aware ordering features can capture several requests and sequence them, rather than choosing only one label.
  • In company tests, it reports up to 97% lower token use and roughly 10-times faster processing for some data-labeling tasks; test conditions are not fully detailed in the supplied report.
  • The model is available through Ant’s platform, but the claimed benchmark leadership is company-reported and should be validated against production workloads.
  • Specialized decision models may become a useful lower-cost layer in agent systems, particularly where outputs must be machine-readable and errors have workflow consequences.
Open at Zhidx Ask Claude ChatGPT

China issues its first industry standard for window-cleaning robots

# Policy Leiphone 5h

A new Chinese standard sets common safety and cleaning tests for window-cleaning robots, a fast-growing product category where a malfunction can mean a fall rather than a missed patch of floor.

More
  • The standard defines test conditions for several glass types and sets minimum thresholds for overall cleaning, edge cleaning and durability.
  • The article cites a minimum overall cleaning score of 70%, edge score of 65% and 200 hours of operating durability.
  • The standard gives buyers a shared vocabulary and comparison method in a category where vendors have used inconsistent claims about suction and cleaning performance.
  • Its initial thresholds are a baseline, not a guarantee of strong performance on unusual windows or in real homes; the article says even leading manufacturers use broader internal test scenarios.
  • As production scales, the harder challenge is translating a lab pass into reliable performance across components, manufacturing lots and varied home layouts.
Open at Leiphone Ask Claude ChatGPT

Tencent opens its search infrastructure to AI developers

# AI GeekPark 1d

Tencent has opened Yuanbao Search as an API, positioning an internally used search system as infrastructure for agents that need current, sourced information rather than model memory alone.

More
  • The report says the system supports more than 50 Tencent products, indexes data at a scale of hundreds of billions of items and draws on 157 designated sources.
  • Its intended output is structured, current information for models and agents, rather than a conventional ranked list of web pages for people.
  • This places search between foundation models and applications: a retrieval layer can ground answers, but its source selection and ranking also shape what agents can know.
  • Tencent says it will emphasize trusted sources to limit low-quality generated material; that is a governance choice as much as a retrieval-engine problem.
  • The article provides company-supplied usage figures but no independent retrieval-quality benchmarks or comparison with rival search APIs.
Open at GeekPark Ask Claude ChatGPT

Legal AI company Legora raises $550 million as enterprise demand grows

# Analysis Zhidx 21h

Swedish legal-AI company Legora raised $550 million at an $8.8 billion valuation, reporting more than $200 million in annual recurring revenue and expanding beyond law firms into corporate legal teams.

More
  • Legora’s platform combines client documents, legal tools and workflow agents for tasks including contract review, drafting and evidence organization.
  • The company says its annual recurring revenue doubled from $100 million to more than $200 million in under six months, and that it serves over 140,000 legal professionals.
  • Its rival Harvey also recently raised $550 million, illustrating intense competition for large law firms and in-house legal departments.
  • The report notes Harvey has worked with Kimi for a legal model, but Legora’s fundraising and growth are the main subject; the Chinese-model connection is not central to the story.
  • The commercial test is whether document-grounded outputs are traceable and useful enough for lawyers to rely on, not merely whether a platform can chain multiple steps.
Open at Zhidx Ask Claude ChatGPT

Tmall wants AI to surface needs shoppers have not searched for

# AI GeekPark 2h

Alibaba’s Singles’ Day strategy casts AI as a way to infer shopping intent from context, shifting competition from matching keywords to deciding which products answer a scenario.

More
  • Tmall executives describe using life-stage and behavior signals to infer demand a shopper has not explicitly searched for, such as linking travel plans with sports interests.
  • Taobao is applying large models to text, voice, image and video queries, then combining them with product and shopping knowledge to generate scenario-based results.
  • The article reports 50.4% growth in broad knowledge queries and an 11.3% click rate for knowledge units, figures supplied by the company.
  • If conversational shopping works, sellers may compete to be understood as the answer to a situation rather than to rank for a fixed keyword.
  • The article does not establish whether AI recommendations create new purchases or redirect demand from existing search and recommendation surfaces.
Open at GeekPark Ask Claude ChatGPT

Chinese internet users read France’s protests through domestic concerns

# Culture SCMP 4h

Chinese coverage finds a mix of geopolitical cynicism, national pride and concern for damaged landmarks in online reactions to French protests over living costs and public services.

More
  • The protests began with school grievances in Paris before widening into demonstrations over living costs and deteriorating public services, according to the report.
  • The article describes Chinese online reactions as combining criticism of France with pride in China and sadness about damage to French landmarks.
  • That mixture is a useful snapshot of how foreign unrest is discussed in China: commentary can serve as both geopolitical comparison and a lens on domestic confidence.
  • The excerpt gives a broad characterization rather than examples, survey data or a breakdown of how common each reaction is.
Open at SCMP Ask Claude ChatGPT

A Chinese cancer drug reaches US shelves after patients traveled to China

# Analysis SCMP 10h

AstraZeneca says it will begin selling a Chinese-developed cancer drug in the US after Americans traveled to Shanghai to obtain it, highlighting delays between approval and commercial access.

More
  • The report says the drug had US Food and Drug Administration approval but was not yet available through ordinary US sales channels.
  • Patients’ trips to China exposed a gap between regulatory clearance and actual market availability, prompting criticism of the delay.
  • The case signals that Chinese-developed medicines can become relevant to patients abroad, not only to domestic clinical and commercial markets.
  • The excerpt does not identify the drug or explain the regulatory and distribution steps behind the delay, so the broader implications depend on details beyond the supplied text.
Open at SCMP Ask Claude ChatGPT

BroadLink says specialized models can bring appliance AI costs to about $1 a year

# AI GeekPark 19h

Chinese smart-home supplier BroadLink argues that appliances need low-cost, low-latency decisions more than open-ended generation, using small specialist models with a larger model as a reviewer.

More
  • The company's proposed architecture puts compact expert models in the control path and uses a larger model to inspect errors and support iteration, aiming to avoid paying for a general-purpose model on every interaction.
  • Its CEO says some appliance functions can be added through over-the-air updates to existing connected modules, avoiding hardware redesign and recertification.
  • The reported annual cost of about one yuan per device is a company claim; the interview provides no workload breakdown, accuracy results or independent cost comparison.
Open at GeekPark Ask Claude ChatGPT

A former US official weighs realistic goals for AI-risk talks with China

# Policy The Wire China 23h

A former Biden-era State Department official discusses leading the first US-China AI-risk talks and argues that dialogue should pursue practical, limited objectives.

More
  • The interview concerns bilateral discussion of AI risks, a subject where technical competition and strategic distrust make broad agreements difficult.
  • Its value is the former official’s account of what direct engagement with Chinese counterparts can realistically accomplish.
  • The supplied description does not give the interview’s specific proposals, so it cannot support claims about agreed safeguards or outcomes.
Open at The Wire China Ask Claude ChatGPT

Chinese astronomers find a stable three-star system with FAST

# Signal SCMP 7h

Researchers using China’s FAST radio telescope identified a stable system of two dead stars and a living star, a real three-body configuration unlike the chaotic fictional one.

More
  • The system is about 7,000 light-years away and includes a pulsar spinning more than 300 times per second, according to the report.
  • FAST, located in southwest China, enabled the radio observations behind the discovery.
  • The finding matters scientifically because three-body systems can be dynamically complex; this one appears to maintain a stable orbital arrangement.
  • The article’s headline invokes the Chinese science-fiction novel The Three-Body Problem, but the research result itself is an astronomy discovery, not a validation of its plot.
Open at SCMP Ask Claude ChatGPT

China’s Singles’ Day loses its mass-shopping excitement

# Culture TMTPost 40m

Chinese coverage portrays the 18th Singles’ Day as a less compelling consumer event: platforms are simplifying discounts and shifting toward members, service and AI-driven shopping.

More
  • The article says promotions have stretched across much of the year, making a month-long shopping festival less distinctive to consumers and harder for merchants to justify.
  • Platforms are moving away from complicated deposits and cross-store discounts toward direct reductions, targeted coupons and member benefits; discounting itself has not disappeared.
  • The article links the shift partly to new rules against coercing merchants into low prices or mandatory automatic repricing.
  • Alibaba, JD.com and ByteDance are testing AI-assisted product discovery, but the article says none has disclosed how much genuinely incremental sales it generates.
  • The strategic question is whether conversational shopping creates new demand or simply redirects transactions that would otherwise have come through search, livestreams or recommendations.
Open at TMTPost Ask Claude ChatGPT

Robot insurers are trying to price risks before the data exists

# Analysis TMTPost 57m

Chinese insurers are beginning to price embodied-robot risk using hardware, operating context and system logs, but fast-changing models and sparse claims data make reliable underwriting difficult.

More
  • Insurers’ reported pricing factors include robot value, repair cost, operating environment, runtime, safety history and technical maturity.
  • One insurer’s framework divides a robot into algorithmic decision-making, motion control and physical hardware, suggesting software updates could affect insurance risk differently from fixed components.
  • The article cites an early robot-damage claim of 5,977 yuan (about $840), while a separate injury dispute remains unresolved amid missing operating records and training documentation.
  • A model upgrade can change the insured system during a policy term, while today’s small number of deployments offers little actuarial history.
  • If insurers require auditable logs and consistent incident reporting, premiums could become a practical incentive for safer design and more transparent robot operations.
Open at TMTPost Ask Claude ChatGPT

Tesla's profit falls as it touts FSD growth and China ambitions

# Analysis QbitAI 1d

Tesla's quarterly operating profit fell sharply despite stronger vehicle gross margins; Elon Musk also claimed its driver-assistance technology would lead Chinese automakers, without specifying a China launch timeline.

More
  • The report says operating profit fell 56% year over year as operating expenses rose, while the company cited research spending and other costs.
  • Tesla reported 1.74 million global paid FSD users, with monthly subscriptions exceeding half of the total for the first time; those figures do not indicate how many users are in China.
  • The China angle is the gap between Tesla's competitive claims and the absence of a concrete FSD availability date in the market where local automakers are rapidly building their own assisted-driving systems.
Open at QbitAI Ask Claude ChatGPT

ByteDance and Tencent take opposite routes in China’s mini-drama market

# Culture TMTPost 57m

ByteDance plans to phase out Red Fruit’s coin-reward acquisition program as Tencent boosts creator payouts on WeChat, revealing different bets on ad-funded viewing and paid content.

More
  • Red Fruit’s reward coins attracted viewers through task incentives; the article argues that weaker ad returns make this low-intent traffic less valuable.
  • Tencent’s reported 100% revenue incentive is aimed at producers and promotion teams, not at paying viewers to watch.
  • The contrast reflects different distribution economics: Red Fruit is a standalone app, while WeChat can draw on existing social, payment and publishing infrastructure.
  • The article suggests the shift may favor paid, higher-quality drama over ad-supported volume, but this remains a strategic interpretation rather than demonstrated outcome.
  • Any creator migration will depend on conversion and acquisition costs; generous platform shares can lose their advantage if competition drives up traffic prices.
Open at TMTPost Ask Claude ChatGPT

How an independent wrestling promotion built an audience in China

# Culture Ginger River Review 31m

A small promotion founded by an American in China has spent years teaching audiences the rituals of pro wrestling while adapting shows to local culture.

More
  • Middle Kingdom Wrestling has staged events in China since 2015, bringing Chinese and international wrestlers together; the reported Tianjin event featured five matches.
  • The organizer describes newcomers learning when to cheer and boo, while local touches include traditional elements and a ring announcer using Tianjin dialect.
  • The story offers a rare view of how a niche imported entertainment form builds an audience through repeated live events rather than a large platform or national sports system.
Open at Ginger River Review Ask Claude ChatGPT

Chinese startups are turning tennis-ball machines into interactive coaches

# Analysis GeekPark 2h

Several China-linked AI tennis products have drawn strong crowdfunding, but the category is still defining whether it sells a training tool, a moving opponent or a software-led service.

More
  • AceiiLab’s device moves along court lines and predicts where a player’s shot will land, then launches a new ball to create a rally-like rhythm rather than literally returning the shot.
  • The product uses cameras, programmed ball patterns and an app; the report says users may need time and skill to adapt to its pace.
  • The article describes a shift from one-time hardware sales toward software, training data and ongoing engagement as ways to reduce the risk that a ball machine gathers dust.
  • Crowdfunding results and founder claims show demand signals, but do not establish retention, training gains or a sustainable service business.
Open at GeekPark Ask Claude ChatGPT

An AI agent completes a Chinese home sale without charging commission

# Analysis 36Kr 7h

An AI intermediary reportedly handled the online matching, verification and negotiation for a 1.95 million yuan (about $275,000) resale home in Zhuhai, with humans retaining in-person and legal work.

More
  • The report says the AI handled listing, buyer matching, listing verification, comparable-price information and online negotiation; people handled viewing, contract witnessing and transfer.
  • The seller waived a reported 40,000 yuan (about $5,600) traditional commission in favor of the buyer.
  • This model separates information and matchmaking from physical fulfillment, challenging the traditional bundling of those tasks into a broker commission.
  • One completed transaction demonstrates a possible workflow, not yet a scalable alternative; the article says publicly documented end-to-end examples remain rare.
  • Scaling depends on local fulfillment networks, sellers’ willingness to list and how buyers assign responsibility when an AI-led transaction goes wrong.
Open at 36Kr Ask Claude ChatGPT

A Chinese founder’s tax bill raises questions about corporate accountability

# Policy TMTPost 47m

Qudian says it will pay founder Luo Min’s roughly 790 million yuan (about $110 million) personal tax bill tied to historical equity restructuring, provoking scrutiny because the founder controls the company.

More
  • The company’s stated rationale is that the equity was held on its behalf and the restructuring followed company instructions; the tax authority reportedly treated it as a personal transfer of property.
  • The article says the payment is large relative to Qudian’s market value and cash, while its operating business has sharply contracted.
  • It contrasts the company’s decision with other Chinese founders reportedly raising funds personally to cover tax liabilities, making the use of listed-company assets the central governance issue.
  • The source presents the arrangement as a potential transfer of a founder’s historical cost to public shareholders; the company’s explanation and the tax ruling are not independently adjudicated in the excerpt.
  • The episode illustrates how tax enforcement around older ownership structures can collide with founder control and minority-shareholder protections.
Open at TMTPost Ask Claude ChatGPT

Micron workers in Taiwan strike over profit-sharing terms

# Culture Solidot 5h

Workers at Micron's Taoyuan site staged a one-day strike over a proposed long-term profit-sharing system and the transparency of annual bonuses.

More
  • The union says it wants a formula tied to company operating profit and quarterly rather than annual payouts; it reports about 2,400 members among roughly 3,000 site workers.
  • A separate union at Micron's Taichung site signed a labor agreement, but its terms were not disclosed in the report.
  • The dispute highlights how competition for semiconductor talent intersects with local expectations over how company performance is shared with employees.
Open at Solidot Ask Claude ChatGPT

Chinese car brands in France are being sorted by showroom economics

# Analysis TMTPost 20m

A China-focused analysis says Chery’s Jaecoo is gaining more traction than Omoda in France because price positioning, leasing and dealer execution matter more than announcing two brands at once.

More
  • The article says the two Chery brands share a parent and sales network but entered at different price points, and early registration patterns diverged.
  • Jaecoo’s positioning in a more affordable hybrid-SUV segment appears better aligned with ordinary household budgets than Omoda’s higher-priced launch range.
  • The article argues that dealer economics, vehicle availability, lease payments and service coverage determine whether Chinese manufacturing strength becomes repeat retail demand.
  • It notes that French sales mixes for Chinese brands include hybrids and plug-in hybrids, complicating the assumption that their European expansion is primarily electric.
  • The early gap may reflect launch timing as much as durable brand preference; later registration and repeat-purchase data will matter.
Open at TMTPost Ask Claude ChatGPT

China's AI performers draw attention, but durable business models remain unproven

# Culture TMTPost 1h

Chinese platforms and creators are building AI performers into ongoing talent brands, but a viral short-drama character's ad controversy shows how hard it is to turn attention into trusted commerce.

More
  • The article distinguishes characters spun out of successful short dramas from original AI performers launched before they have a hit work; the former appear more likely to gain audiences.
  • A prominent AI character's reported advertising quote was not a sale price, and an ad claiming personal experience with contact lenses drew criticism because the character is virtual.
  • The debate raises questions about endorsement truthfulness, responsibility for synthetic personalities and whether virtual performers can maintain attention after their source drama fades.
Open at TMTPost Ask Claude ChatGPT
Source ledger
ChinAI 403 0/0 HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk 200 1/20
Pekingnology 200 0/20
Ginger River Review 200 1/20
Sinocism 200 1/20
QbitAI 200 5/10
36Kr 200 10/30
Solidot 200 13/20
Leiphone 200 17/20
GeekPark 200 10/30
IT Home 200 12/60 +37 over cap
SCMP 200 10/50 +7 over cap
Initium 200 3/15
Jiemian 200 8/30 +8 over cap
China Digital Times 200 1/15
The Wire China 200 1/50
Techmeme 200 2/15
Hacker News 200 0/20
V2EX 200 8/50 +38 over cap
Zhidx 200 8/20
InfoQ China 200 8/20 +12 over cap
New Wisdom 200 8/15 +3 over cap
TMTPost 200 8/17 +9 over cap