China Desk

China News, Summarized 22 Oct 2026, 20:07 UTC 34 stories 21/23 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

Report: DeepSeek raises about 100 billion yuan in China’s largest AI round

# Zhidx 15h

Bloomberg, citing unnamed sources, reports that DeepSeek raised about 100 billion yuan (roughly $14B), with CATL and Tencent among the largest investors, to fund infrastructure, models and a possible IPO.

More
  • The report says DeepSeek increased its fundraising target twice, from 50 billion yuan to 80 billion and then roughly 100 billion yuan (about $14B).
  • The scale would make this the largest reported financing in China’s AI sector and a major expansion of capital available to a leading open-model lab.
  • According to Bloomberg’s sources, DeepSeek plans to invest in AI data-center infrastructure and develop additional open models for Chinese and overseas users.
  • The report links the financing to IPO preparation; a prior Reuters report cited in the article said the company had hired China International Capital Corporation to prepare for a possible Shanghai listing.
  • The claims rely on unnamed sources and the financing details are not public; the significance depends on confirmation and how much capital reaches compute, research and deployment.
Open at Zhidx Ask Claude ChatGPT

Xi invokes the Long March ahead of a key Communist Party meeting

# SCMP 9h

Days before the Party’s fifth plenum, Xi Jinping used the Long March anniversary to argue that unified Party leadership turned past defeat into victory, linking the revolutionary story to today’s geopolitical pressures.

More
  • Xi delivered the speech at a commemoration of the Long March’s 90th anniversary, according to the report.
  • He credited the Communist Party’s historical reversal of fortunes to rallying around Mao Zedong’s leadership.
  • The timing, just ahead of the fifth plenum, makes the anniversary speech part of current political messaging rather than only historical commemoration.
  • The Long March is a foundational Party narrative: endurance, unity and disciplined leadership are presented as sources of political legitimacy.
  • The speech points to an emphasis on cohesion under pressure, but the account does not identify any specific policy decision or personnel signal from the upcoming plenum.
Open at SCMP Ask Claude ChatGPT

China launches a level-four response after a 7.0 quake in Tibet

# Jiemian 4h

China activated a national level-four earthquake response after a reported magnitude-7.0 quake in sparsely populated, high-altitude Nagqu; central officials dispatched a team to guide rescue and monitoring.

More
  • The emergency ministry says the quake struck Baqing County at a depth of 10 kilometers, with the epicenter near 4,800 meters above sea level.
  • The national earthquake-relief headquarters and the emergency ministry initiated the level-four response and sent a team to the affected area.
  • The account emphasizes the government’s rapid dispatch and coordination; it gives no casualty or damage figures.
  • The sparse population may limit immediate human impact, but remoteness and altitude complicate rescue and assessment.
  • This is an initial official response, not a final damage assessment; the scale of the event depends on verified reports from the area.
Open at Jiemian Ask Claude ChatGPT

China’s humanoid-robot boom is building a market for human-hand data

# Analysis ChinaTalk 10h

A ChinaTalk essay examines the data-collection infrastructure behind China’s humanoid push: first-person recordings of people handling objects offer robot developers a cheaper, broader alternative to collecting every demonstration on robots.

More
  • The essay describes egocentric data as video captured from a person’s point of view, recording hands and object interactions that robots need to learn.
  • Its account places the work within China’s state-backed embodied-AI push and cites reports of more than 150 humanoid-robot firms and 40 state-owned robot data centers by late 2025.
  • It says collecting human-hand data is especially important because robot hardware is expensive and cannot generate diverse demonstrations as cheaply as human data collection.
  • The essay frames a new supply chain in which everyday human activity becomes training material for robotics companies, raising questions about workforce practices, consent and data ownership.
  • The excerpt ends before describing how collection is organized or compensated, so it does not establish workers’ conditions or the extent of data use in China.
Open at ChinaTalk Ask Claude ChatGPT

iFlytek says its new model was trained on domestic chips and supports a million-token context

# AI Zhidx 6h

iFlytek unveiled a cloud-and-edge model family, claiming domestic-chip training, faster long-context infrastructure and a small open model for local devices; its approach makes compute sovereignty and on-device routing central to the product.

More
  • The company says its cloud model has a one-million-token context window and its 4-billion-parameter edge model is open source.
  • iFlytek reports reducing a 2-trillion-token, 256K-context training run on its Feixing II domestic-compute platform from 4.5 months to 1.2 months.
  • For inference, the company describes a four-tier KV-cache architecture and reports raising cache-hit rate from 63% to 95.7%, with a 4.5-fold throughput improvement on its platform.
  • Its edge-cloud routing model first determines whether a request should run locally, in the cloud or across both; iFlytek says routing accuracy rose from 81.2% to 96.1%.
  • These figures are company-reported benchmarks. The consequential test is whether domestic hardware can sustain useful model quality and predictable serving cost across real deployments.
Open at Zhidx Ask Claude ChatGPT

Unitree’s Wang says humanoid robots could reach useful autonomy within two years

# AI GeekPark 8h

At Unitree’s physical-AI conference, founder Wang Xingxing argued that stronger foundation models could rapidly unlock robots, while stressing that manufacturing scale—not building a frontier model—is Unitree’s strategic bet.

More
  • Wang shortened his estimate for a humanoid-robot breakthrough to one or two years, and suggested a robot might soon handle a simple household task in an unfamiliar home.
  • He said Unitree will develop some AI itself but expects to work with leading domestic and overseas model companies rather than duplicate their frontier models.
  • The report says Wang views reliable, low-cost robot bodies and mass production as decisive; he put eventual industry demand at tens of millions of units annually.
  • Unitree’s G1 humanoid has reportedly shipped around 10,000 to 20,000 units, while the company’s strategy also includes an automated loop of paper search, code generation, simulation and real-robot testing.
  • These are the founder’s forecasts and company-reported figures, not evidence that general-purpose household autonomy is near; robust operation outside demos remains the key test.
Open at GeekPark Ask Claude ChatGPT

A Chinese startup’s robot-memory system reports new benchmark highs

# AI Leiphone 17h

China’s Beta Infinite says its BetaMem system combines persistent state, event history and retained source data, and reports top scores across four robot-memory benchmarks; the design targets robots that must operate for days, not minutes.

More
  • The paper describes memory as a maintained judgment about the world, rather than a log or a larger context window.
  • Its system combines a time-aware state graph, an append-only event log and a limited archive of selected original images and audio.
  • The company says a 9-billion-parameter open model used to generate memories achieved state-of-the-art results on M3-Bench-robot, LoCoMo, OpenEQA and EgoLifeQA.
  • The architecture aims to preserve evidence and track changes instead of overwriting an object’s old state with its latest observation.
  • The paper’s reported benchmark results need independent replication; real deployment also has to control storage costs and prevent a mistaken memory from cascading into later actions.
Open at Leiphone Ask Claude ChatGPT

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

# AI QbitAI 12h

China Telecom’s Xing4.0-29B-Intent combines intent classification with structured extraction, producing schema-bound JSON for systems such as customer-service routing and smart-home controls.

More
  • The model and 2.75 million training examples are being released with weights and an inference image, according to the company.
  • It uses a unified schema-driven parsing framework to return both an intent category and the relevant entities or parameters.
  • China Telecom reports a 58-point aggregate score across four extraction tasks, above the best general model it tested at 49; the evaluation is company-run.
  • The model is already used in the company’s customer service and smart-home products, but independent evaluation and performance on changing schemas remain open questions.
Open at QbitAI Ask Claude ChatGPT

Human first-person video becomes a robotics training-data business

# AI QbitAI 13h

A Chinese report says Silicon Valley vendor Maxinsights supplies large egocentric-video datasets to robotics labs and entered China this year, spotlighting human video as a scalable training input.

More
  • The article attributes more than two million delivered hours of data, over 1.5 million hours with hand tracking and language labels, and collection across more than 2,000 environments to Maxinsights.
  • It says Dyna-2 used one million hours of first-person human video and that prediction error fell about 12% as training data scaled from 1,000 to one million hours; the vendor's role and figures are reported claims, not independently established here.
  • First-person video offers varied examples of human hand-object interaction without requiring every example to be collected on a robot, but the report does not show how reliably that data transfers to real-world robot control.
  • The article says Maxinsights entered the Chinese market in March but gives no detail on local customers or revenue.
Open at QbitAI Ask Claude ChatGPT

EU-China trade truce frays as both sides claim the other blinked

# Policy SCMP 8h

The short-lived EU-China deal is already producing competing narratives: European officials see collective pressure winning concessions, while Beijing rejects the idea it was coerced and Brussels is considering more restrictions.

More
  • The article says Brussels and Beijing reached a deal less than two weeks before the new dispute over who conceded first.
  • European officials in Brussels and Berlin view the agreement as evidence that coordinated EU pressure can extract concessions from China.
  • The report says European officials are preparing additional trade restrictions rather than treating the deal as a settled resolution.
  • Beijing is described as bristling at the claim it was strong-armed, underscoring that the political meaning of the deal is itself contested.
  • The excerpt does not specify the deal’s terms or the proposed new restrictions, so the immediate commercial exposure cannot be assessed from this account alone.
Open at SCMP Ask Claude ChatGPT

Tencent opens its internal AI search system as a developer API

# AI GeekPark 11h

Tencent is turning Yuanbao Search from an internal service into a cloud API, positioning search as an information layer for agents that need fresh, sourced data rather than another general-purpose chatbot.

More
  • Tencent says the system supports more than 50 internal products, uses an index of hundreds of billions of items and draws on 157 authoritative sources.
  • The product is designed to return current, structured information to models and agents, rather than simply provide a list of keyword-matched web pages.
  • This reflects a broader infrastructure shift: retrieval quality, source selection and freshness can become a shared dependency across many applications, rather than a feature rebuilt by every agent team.
  • Tencent argues that self-building search requires crawling, parsing, indexing and ranking expertise that most developers cannot economically maintain.
  • The coverage is Tencent’s account of its own system; the API’s retrieval quality, source coverage and real-world reliability are not independently established here.
Open at GeekPark Ask Claude ChatGPT

AI comic studios produce more, but revenue is not keeping pace

# Culture GeekPark 11h

Chinese AI-comic makers say production costs and staffing have collapsed, while revenue has stagnated and compute now eats a larger share of budgets; the boom is shifting leverage from production capacity to story judgment and IP.

More
  • One producer says an 80-minute series now costs about 40,000 yuan (roughly $5,600), down from around 100,000 yuan (about $14,000) a year earlier.
  • The same account contrasts a former team of ten making one series a month with one person now producing about three, while monthly output surged without matching revenue growth.
  • As labor costs fall, the interviewees say AI compute has become a much larger part of production expense, with tokens replacing some of the savings from smaller crews.
  • The platform trend is toward individual creators and small studios, weakening the role of large production houses that once secured subsidies and preferred access.
  • The interviews suggest creative judgment and source material may become scarcer than production capacity, but the figures are industry anecdotes rather than audited market data.
Open at GeekPark Ask Claude ChatGPT

Qwen team details GPU optimizations for linear attention

# AI InfoQ China 2h

Alibaba’s Qwen team describes FlashQLA, an operator library for Gated Delta Network attention that targets the serial state-update bottleneck with chunked computation, warp specialization and hardware-aware algebra.

More
  • Linear attention reduces sequence-length scaling compared with standard softmax attention, but its recurrent state update creates a dependency between chunks that limits GPU parallelism.
  • The implementation uses a WY representation to move from token-by-token updates to chunks, typically of 64 tokens, enabling more work to use matrix-multiplication hardware.
  • The talk separates the computation into projections, chunk-level state recurrence and output generation; most stages can run independently, while state propagation remains sequential across chunks.
  • FlashQLA’s stated engineering focus is reducing parallelism bottlenecks, redundant memory traffic and special-function-unit load, while reusing properties of sliding windows.
  • The article excerpt does not include its final performance numbers, so the practical gain over existing implementations cannot be quantified here.
Open at InfoQ China Ask Claude ChatGPT

China and Djibouti call for open shipping lanes in the Red Sea

# Policy SCMP 8h

In their first joint statement on the fallout from the US-Iran war, China and Djibouti backed efforts to keep Red Sea routes and the Bab al-Mandeb open, signaling Beijing’s interest in maritime stability with a regional partner.

More
  • The statement followed a meeting in Beijing between Foreign Minister Wang Yi and his Djiboutian counterpart.
  • Both governments emphasized freedom of navigation and the safety of international shipping in the Red Sea and Bab al-Mandeb.
  • The wording gives China a diplomatic position on a conflict-related shipping risk without, in the excerpt, assigning blame or endorsing a specific security operation.
  • Djibouti’s location near the Bab al-Mandeb makes the partnership directly relevant to one of the world’s important maritime chokepoints.
Open at SCMP Ask Claude ChatGPT

China says DeepSeek and MiniMax completed procedures for overseas AI use

# Policy Jiemian 3h

Chinese coverage presents the proposed AI export-control catalog as boundary-setting, not a broad retreat from openness, noting that DeepSeek and MiniMax completed registration for open releases and overseas licensing.

More
  • The Ministry of Commerce and Ministry of Science and Technology sought public comment on revisions to a catalog covering restricted and prohibited technology exports.
  • The proposal includes 23 AI-related areas, including large models, agents, computing clusters and information processing.
  • The article says DeepSeek and MiniMax completed registration procedures and then carried out open-source releases and overseas licensing.
  • Chinese official messaging, as presented here, spends more time explaining what the rules do not regulate than spelling out restrictions, framing the move as delineation rather than closure.
  • The practical boundary is not yet clear from this short account; the final catalog and how authorities apply it will determine whether firms face meaningful constraints on international distribution.
Open at Jiemian Ask Claude ChatGPT

Tesla earnings show rising costs as Musk claims an FSD lead in China

# Analysis QbitAI 12h

Tesla reported sharply lower operating profit while Musk claimed its FSD technology leads Chinese automakers; the report also highlights rising output and exports from the Shanghai factory.

More
  • The article reports Shanghai factory wholesale sales of 95,366 vehicles in September, with exports up 58% year over year, tying China production to Tesla's global delivery mix.
  • Musk's claim of an FSD advantage over Chinese automakers is not accompanied in the supplied text by evidence of a China launch timetable or local deployment milestones.
Open at QbitAI Ask Claude ChatGPT

Quad summit confirmed as Manila signals possible talks among US allies

# Policy SCMP 5h

Chinese coverage places the planned Quad summit and possible informal “Squad” talks within a US-led effort to counter China’s regional influence, with the meetings expected around ASEAN gatherings in Manila.

More
  • Washington confirmed the first leaders’ summit of the Quadrilateral Security Dialogue in more than two years, to be held alongside regional talks.
  • The Philippine ambassador signaled possible meetings of the informal “Squad” defense grouping before the ASEAN gathering.
  • The report explicitly frames both groupings as part of a US strategy to balance China’s influence in the Asia-Pacific.
  • The excerpt does not say which leaders will attend the Squad talks or what policy commitments are expected.
Open at SCMP Ask Claude ChatGPT

Beijing criticizes Taipei ban on mainland travel for Sun Yat-sen events

# Policy SCMP 19h

A dispute over events marking Sun Yat-sen’s 160th birthday has reopened a cross-strait argument over who can claim the revolutionary figure, after Taipei barred officials from attending mainland commemorations.

More
  • The article says Beijing condemned Taipei’s decision to prevent Taiwanese officials from attending events in mainland China on November 12.
  • Sun Yat-sen is revered on both sides of the Strait and is associated with the end of imperial rule and the founding of the Republic of China.
  • The dispute shows how shared historical symbols remain part of present-day cross-strait political competition.
  • The excerpt does not detail the legal basis for Taipei’s restriction or Beijing’s response beyond its criticism.
Open at SCMP Ask Claude ChatGPT

Survey finds young Chinese adults saving for illness and job loss

# Culture 36Kr 7h

A 908-person survey portrays China’s post-2000 generation as prioritizing financial buffers: respondents report average annual savings of 25,400 yuan (about $3,600) and widespread use of online commercial insurance.

More
  • In the survey, illness or hospitalization was the most commonly cited financial worry, followed closely by sudden unemployment or an extended loss of income.
  • Respondents reported saving an average of 25,400 yuan (roughly $3,600) over the prior year; the survey says women saved more on average than men.
  • About 65% said they had purchased commercial insurance through online channels, while many preferred liquid savings over locking money into long-term accounts.
  • The article frames savings and insurance as a response to uncertain income, medical costs and changing employment conditions, rather than primarily as preparation for major purchases.
  • This is a self-selected survey of 908 respondents, not a representative national estimate; it offers a mood signal rather than a precise measure of all young adults.
Open at 36Kr Ask Claude ChatGPT

Keeta shifts its Brazil expansion inland as chain deals block Rio access

# Analysis TMTPost 11h

Meituan's Keeta is shifting its Brazil push inland after exclusive chain deals constrain access to Rio, trading metro density for markets with fewer locked-up restaurants.

More
  • The report says deals involving iFood and 99Food left more than half of Keeta's targeted chain outlets off-limits in Rio, despite a substantial launch budget.
  • Brazilian antitrust rules and a court action narrowed some exclusivity clauses, but legal access has not automatically translated into commercial reach.
  • Keeta is expanding into inland municipalities, where fewer outlets are tied up by single-homing deals, but longer routes and low order density complicate delivery economics.
  • The strategy reflects Meituan's search for growth beyond a mature domestic market; the report does not establish whether subsidies can produce durable repeat ordering in Brazil.
Open at TMTPost Ask Claude ChatGPT

A smart-home supplier says appliance AI must be cheap and predictable

# Analysis GeekPark 6h

BroadLink argues that household devices need small specialist models for dependable control, with a large model used mainly to review errors; its CEO claims costs can fall to about one yuan per device per year.

More
  • The company says most connected appliances already have OTA update paths, but earlier smart-home efforts failed to give consumers enough reason to use them.
  • Its proposed architecture uses small expert models for low-latency, deterministic decisions and a larger model in a supervisory role.
  • The CEO says AI operating costs can reach about one yuan per device annually, roughly $0.14, and that upgrades can often arrive through the connectivity module rather than new hardware.
  • That cost and deployment model is a company claim; adoption still depends on consumers activating connected features and manufacturers paying for the service.
Open at GeekPark Ask Claude ChatGPT

China plans a program to expand consumer use of smart products

# Policy IT Home 5h

The Commerce Ministry says it will issue a plan to accelerate smart-product adoption, linking consumer-device growth to the next five-year plan and existing incentives for phones, tablets, wearables and smart glasses.

More
  • The ministry’s proposed work plan is part of its broader effort to support consumption through product upgrades and new consumer formats.
  • The cited consumption plan targets roughly 60 trillion yuan in total retail sales by 2030 and backs smart-home interoperability and a shift from isolated smart devices to whole-home systems.
  • A 2026 replacement subsidy already covers phones, tablets, smart watches and bands, and smart glasses, with local governments encouraged to support newer devices.
  • This is an industrial-demand policy as much as a consumer measure: subsidies and interoperability goals can help domestic device makers build markets beyond individual product launches.
  • The ministry has announced a plan to come; the item does not yet give its eligibility rules, funding or implementation timetable.
Open at IT Home Ask Claude ChatGPT

Taiwan comedian reportedly hit in staged crash; why the infiltration law may not apply

# Policy Initium 14h

A Taiwan-focused report asks why the Anti-Infiltration Act may not apply after comedian Huang Yihao was reportedly struck in an alleged staged crash; the supplied text contains only the headline.

More
  • The headline frames the incident as a question about the boundary between alleged political intimidation and the reach of Taiwan's Anti-Infiltration Act.
  • The supplied text is page boilerplate rather than the article, so it does not establish who was responsible, what investigators found or why the law may not apply.
Open at Initium Ask Claude ChatGPT

Huawei executive says China’s autonomous-driving shift raises the safety bar

# Policy Leiphone 17h

Huawei-affiliated automotive executive Jin Yuzhi says China is moving from driver assistance toward higher automation, where responsibility increasingly shifts to the system and public trust becomes a competitive requirement.

More
  • He cited a Ministry of Industry and Information Technology figure that L2-or-higher driver-assistance fitment reached 70% by September.
  • He said L3 access testing is under way and a mandatory standard is scheduled to take effect in July 2027.
  • The executive says Huawei’s automotive unit has disclosed assisted-driving mileage and safety data to help build user trust, while positioning itself as a supplier rather than an automaker competitor.
  • The statements are industry advocacy; legal liability and the final standard’s scope will determine what changes for automakers and users.
Open at Leiphone Ask Claude ChatGPT

China launches a global computing-price index with a domestic unit of measure

# Signal Leiphone 17h

A new index platform aims to compare computing services across hardware and regions, using a domestic “computing unit” metric to make heterogeneous AI capacity easier to price and trade.

More
  • The platform was initiated by China Economic Information Service, with Jiuzhang Yunji participating in its measurement methodology and data collection.
  • Its proposed standard unit converts different hardware and cluster configurations into a common measure tied to model training or inference output.
  • The organizers say the index is intended to support market comparison, pricing and policy planning, but no public methodology or independently validated price series is provided here.
Open at Leiphone Ask Claude ChatGPT

US drone-command tests raise questions about a Taiwan conflict

# Analysis SCMP 9h

The report describes US tests of long-range command links for uncrewed aircraft and frames them as part of the military technology competition with China over a possible Taiwan conflict.

More
  • One test reportedly involved an F-22 pilot controlling an autonomous YFQ-44A combat drone in flight.
  • The broader capability is command and coordination across distances, not simply an autonomous aircraft operating alone.
  • The article presents the work as a potential US advantage over the PLA, but the excerpt does not provide test range, latency or operational constraints.
Open at SCMP Ask Claude ChatGPT

China newsletter highlights Long March messaging, youth unemployment and AI export controls

# Analysis Sinocism 21h

Sinocism’s China briefing flags Xi’s Long March anniversary speech, new tech-industry integration guidance, youth unemployment and proposed AI export controls as issues to watch.

More
  • The newsletter’s top-item list links Party historical messaging with the near-term political calendar.
  • It also highlights youth unemployment and AI export controls, two areas where state priorities meet social and commercial pressures.
  • The excerpt is only a topic roundup and supplies no underlying reporting or figures, so it serves as an agenda signal rather than a standalone account.
Open at Sinocism Ask Claude ChatGPT

Former US official discusses realistic goals for US-China AI-risk talks

# Analysis The Wire China 11h

A former Biden-era State Department official reflects on the first US-China talks on AI and argues that expectations for superpower dialogue should be realistic.

More
  • The interview is about what bilateral discussions on AI risk can plausibly achieve, rather than a technical model development.
  • The excerpt offers no specific proposals or account of the talks’ results, limiting its value to the framing of the dialogue.
Open at The Wire China Ask Claude ChatGPT

Beijing defends Raul Castro after Florida issues an arrest warrant

# Policy SCMP 21h

China condemned Florida’s warrant for former Cuban leader Raul Castro, portraying it as US political pressure on Havana and publicly referring to him as a respected comrade.

More
  • The warrant concerns the 1996 downing of two civilian aircraft, according to the report.
  • Foreign Ministry spokesperson Mao Ning responded at a regular briefing after a question from state media.
  • The statement fits Beijing’s support for Cuba and its criticism of US use of legal processes against foreign leaders.
Open at SCMP Ask Claude ChatGPT

IM chooses CATL batteries while rivals move toward in-house development

# Analysis Leiphone 7h

IM says its upcoming LS8 NEO will use CATL cells and packs, betting that a specialist supplier’s manufacturing and quality controls outweigh the control automakers can gain by developing batteries themselves.

More
  • The article contrasts automakers’ moves into battery development with IM’s decision to rely on CATL for both cells and packs.
  • It describes two battery options, a 103-kWh ternary pack and a 93-kWh LFP pack, and claims strong low-temperature performance for the latter.
  • CATL says its manufacturing process has more than 7,000 quality-control points and uses machine vision and AI inspection on production lines.
  • The report’s conclusion is that the strategic choice is not simply make versus buy: the determining factors are cell quality, supply assurance and system-level integration.
Open at Leiphone Ask Claude ChatGPT

Chinese apps adapt their layouts for Apple’s foldable phone

# Culture GeekPark 19h

Chinese coverage focuses on how apps such as Taobao, Amap, Feishu and WPS use the foldable screen for side-by-side workflows, turning the product story into a test of China’s app ecosystem adapting to new hardware.

More
  • The review finds the clearest advantages in shopping comparison, map navigation and document editing, where two panes reduce app switching.
  • Several proposed features, including dual-view driving navigation, are shown only in demos and await software updates.
  • The report highlights that useful cross-app actions require app-level integration; a generic operating-system split screen alone does not pass location or task context between apps.
Open at GeekPark Ask Claude ChatGPT

China’s shopping festival is becoming a conversation with AI

# Culture TMTPost 11h

Chinese coverage describes shoppers using assistants to turn plain-language needs into ranked product choices, while platforms bring promotions and merchant tools into chat interfaces.

More
  • The reported behavior is most useful for shoppers who have constraints but have not settled on a specific product; model recommendations can link into a live store.
  • Platforms are connecting assistants to order histories, product inventory and merchant operations, moving beyond keyword search and generic advice.
  • The article argues that commerce offers measurable outcomes such as conversion and returns, but an assistant without inventory, payment and returns integration remains only a recommender.
Open at TMTPost Ask Claude ChatGPT

Chinese electronics maker plans semiconductor-gas acquisition after prior goodwill write-off

# Signal TMTPost 6h

Xin Ya Process plans to acquire 67% of Qiyuan Gas to enter semiconductor specialty gases, but the target’s recent profit slowdown and the buyer’s earlier 327 million yuan goodwill write-off sharpen execution and valuation risks.

More
  • The proposed purchase would give Xin Ya a controlling stake in a supplier of bulk and specialty gases used in semiconductor manufacturing and display production.
  • The article estimates the initial cash leg could run to several hundred million yuan (roughly $30-60M), though final pricing remains undetermined.
  • Qiyuan Gas earned 30.7 million yuan in 2025 but reported 10 million yuan in the first half of 2026; sellers have promised at least 46.35 million yuan in annual profit from 2027 through 2029.
  • Xin Ya previously wrote off 327 million yuan of goodwill (about $46M) from an unrelated acquisition, and still faces litigation connected to that deal.
  • The deal’s strategic appeal is access to semiconductor materials, but the earnings commitment and funding burden make audited results and final terms decisive.
Open at TMTPost Ask Claude ChatGPT

Aito crash probe cites 139 km/h speed before fatal collision

# Culture IT Home 4h

After a fatal Aito crash in Shaanxi, the company says preliminary data show the car was being driven at 139 km/h on a road limited to 50; local traffic officials said the impact shorted and discolored a signal pole.

More
  • The vehicle maker’s preliminary statement attributes the fire to severe impact, wiring damage and fluid leakage, and says the driver was in control rather than using assisted driving.
  • Police described the incident as a single-vehicle crash with one fatality; the final investigation remains pending.
  • The unusually high reported speed is central to the safety narrative, but should not be treated as a final forensic conclusion.
Open at IT Home 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 0/20
Sinocism 200 1/20
QbitAI 200 3/10
36Kr 200 2/30
Solidot ERR 0/0 ConnectionError: HTTPSConnectionPool(host='www.solidot.org', port=443): Max retries exceeded with url: /index.rss (Caused by NewConnectionError("HTTPSConnection(host='www.solidot.org', port=443): Fail
Leiphone 200 17/20
GeekPark 200 8/30
IT Home 200 12/60 +41 over cap
SCMP 200 10/50 +4 over cap
Initium 200 5/15
Jiemian 200 8/30 +20 over cap
China Digital Times 200 0/15
The Wire China 200 1/50
Techmeme 200 0/15
Hacker News 200 0/20
V2EX 200 8/50 +36 over cap
Zhidx 200 9/20
InfoQ China 200 8/20 +5 over cap
New Wisdom 200 8/15 +2 over cap
TMTPost 200 8/17 +9 over cap