China Desk

China News, Summarized 27 Aug 2026, 09:27 UTC 34 stories 22/23 sources

Today's brief

Beijing is writing the carmaker onto the ticket, and Tesla will not be there to collect

The manufacturer now answers for the violation

Autonomous driving entered Chinese law on Tuesday. The Road Traffic Safety Law revision draft went to the National People's Congress Standing Committee for first reading with a dedicated chapter on autonomous vehicles — the first time the category exists in statute. Nine chapters, 170 articles. The existing law dates to 2004.

The mechanism is the news. When an autonomous vehicle commits a traffic violation with the autonomous function activated, the manufacturer or importer is processed for it — not the person in the seat. Vehicles with the function off, and cars carrying only assisted-driving features, are policed exactly as ordinary cars.

Read that as a wall, not a gift. Chinese carmakers have spent two years selling "smart driving" to buyers who heard self-driving. Every IT Home car story now carries boilerplate insisting assisted driving is not autonomous driving, plus the full L0-to-L5 ladder — a trade press disclaiming its own subject matter.

The draft turns that disclaimer into liability. Legal standing attaches only to certified autonomy, and the ambiguity stops paying. Same week, the Party and State Council told cadres they answer for serious environmental damage on their watch for life, regardless of transfer or promotion. Permanent, named liability is this week's instrument.

Meanwhile the wheel is coming out. A test mule of SAIC's IM LS6 — the large electric SUV on sale from 197,900 yuan (~$27,900) — is circulating with the driver's-side steering wheel deleted entirely. Commenters noted the production car's half-wheel was already halfway there.

And Tesla is reportedly gone. Industry accounts relayed in Leiphone's morning digest say the Shanghai data centre built for FSD data-localisation compliance has been emptied and the liaison team withdrawn; China is absent from Tesla's latest twelve-region FSD availability list after being listed as pending in May. Owners paid 64,000 yuan (~$9,000) for the option.

Go deeper on this section: ClaudeChatGPT

Hundreds missing at the Gyirong crossing

A mudslide on the Nepalese side hit the border port at about 10:30am. Xinhua reported major casualties and missing persons at Gyirong Port in Gyirong county, Shigatse, Tibet; Xi Jinping ordered an all-out search and warned about secondary disasters. SCMP put the missing in the hundreds, many of them foreign tourists. No confirmed toll yet.

The location matters beyond the rescue. Gyirong is Tibet's principal foreign-trade port and handles the bulk of China–Nepal trade, including the corridor for Chinese electric cars going into Nepal. On the Nepali side the Bhote Koshi burst its banks through Rasuwa.

Go deeper on this section: ClaudeChatGPT

Washington stops interviewing, Huawei calls Cairo

Immigrant visa interviews were cancelled worldwide, reason given as "training." An immigration agent posting on RedNote said the Guangzhou consulate scrapped all of Tuesday's slots, per SCMP. Third narrowing in a week, after the proposed $103,265 H-1B petition fee and the consular refusals that hollowed out Chinese attendance at IJCAI in Bremen.

Huawei went shopping elsewhere. Bloomberg reports it has proposed exporting high-end Ascend 950-series chips to build AI data centres for the Egyptian government. Egypt is on nobody's list of chip-diplomacy battlegrounds, which is the point — the pitch tests whether American pressure reaches places Washington is not watching.

Go deeper on this section: ClaudeChatGPT

Alibaba shipped the Qwen4 blueprint early

Qwen3.8-Flash went live at 11pm Beijing time. A multimodal mixture-of-experts model: 125B in the main model plus a 51B N-gram embedding, 6B active per token, 262,144-token native context stretching to 1M with YaRN. API at 1 yuan (~$0.14) per million input tokens, 3 yuan (~$0.42) out.

The architecture is the actual release. Attention is Gated DeltaNet plus Qwen Sparse Attention, a light indexer selecting context at micro-block granularity — 7.6x prefill and 4.9x decode speedup on the 1M kernel. Residuals fan into four gated branches held in FP8; the N-gram table offloads to host memory with async prefetch. Muon optimiser, scaling law refit. Training cost about one ninth of Qwen3.7-Plus.

What got opened is not the product. It is Qwen3.8-Flash-Next — the new architecture's weights, released deliberately ahead of the full Qwen4 family so the community can break the structure first. Best-in-class on eight of fourteen benchmarks at 6B active. Nobody else is publishing next-generation structure before shipping on it.

The wires read the same day as a leaderboard. Bloomberg's take on Qwen3.8-Flash was a cheaper platform that rivals Opus 4.6 and V4-Flash. On MiniMax, the Shanghai lab, the framing was that its M3 model trails rivals like Z.ai's GLM-5.2.

Jiemian led with the money instead. Founder and chief executive Yan Junjie told the interim results call that August annualised recurring revenue passed $800m (5.38bn yuan), with July token consumption twenty times January's. H1 revenue about $120m, up 283%; net loss $358m, narrowed 11%. One story is a lab falling behind. The other is a business compounding. Both are the same company.

And Moonshot wants a percentage from the hyperscalers. Reuters says the Beijing lab is in early talks with Microsoft, Amazon and Google to host Kimi K3, seeking up to 30% of the resulting cloud revenue. Yesterday's story was Western firms building on free Chinese weights; today the weights start asking for rent. Separately, Z.ai confirmed the free model that swept OpenRouter — "Ox Alpha" — is a new GLM, weights tonight.

Go deeper on this section: ClaudeChatGPT

Podcasts stopped asking what AI can do

The Chinese AI conversation moved to what is left for people. Xiaoyuzhou, the podcast app, published a 2026 trends report: AI creators up 187% and shows up 239% in 2025, searches for "collaborative AI" up 227% this year. Discussion of human value alone has produced 335,000 comments and 12.86 million characters.

The audience is younger than the industry assumes. Tech-podcast listeners are 39.72% aged 19 to 25; the youngest guest on the platform is thirteen. Searches for AI founders and researchers are up 470%. Long audio has become where this industry talks to itself — Unitree founder Wang Xingxing recorded his year-in-review there on New Year's Eve.

A state-produced legal drama is doing something unexpected. "The Pillar," made by China Media Group with the film centres of the Supreme People's Procuratorate and Supreme People's Court, dramatises eleven real cases from 1979 to 1997 — the catch-all crime of hooliganism, and a 1984 case in which defence lawyers were arrested for arguing the evidence was insufficient.

Then a very famous professor used it. Luo Xiang, criminal-law professor at China University of Political Science and Law with over 32 million Bilibili followers, argued in Phoenix TV's opinion column that how a society treats lawyers defending unpopular defendants tests whether it accepts that criminal justice must restrain state power. Pekingnology translated it.

Go deeper on this section: ClaudeChatGPT

Threads we are pulling

  • The memory squeeze reached retail. Apple's new Mac mini raised each Chinese tier by 2,500 yuan (~$352) — the entry machine from 4,499 to 6,999 yuan (~$986) — after first-quarter general DRAM contract prices rose 93–98% in a single quarter, per GeekPark's account. Gartner expects the sub-$500 PC to vanish by 2028. Same day, DapuStor booked an enterprise PCIe SSD order worth $514.6m, or 152.7% of its entire 2025 main-business revenue, delivering in 18 weeks. I under-weighted DRAM on 14 August and it keeps compounding.
  • Jack Ma joined the buying, and I was low on the size. Yesterday I counted about HK$120m from chairman Joe Tsai and chief executive Eddie Wu; the figure now reported is HK$200m for the two of them plus over HK$600m (~$77m) from founder Jack Ma, after the HK$80bn placement drew near three times cover.
  • Chang'e-7's slip is now a NASA question. No official reason for Sunday's postponement, and space experts told SCMP it does not change the contest over a permanent lunar presence. Consolation: China established a two-way laser link over 400,000 km, 1.25Mbps up and 100Mbps down — an 8K lunar image in about 12 seconds against four to five minutes on a 5Mbps microwave link.
  • The garment order is the humanoid number that matters. Zhejiang Humanoid Robot Innovation Center's 2,000-unit deal with Jack Technology rests on 98% fabric-piece separation, ±2mm placement, 35 seconds per pocket and roughly 1,000 garments a day, by its own account at WRC. Cloth deforms; that is the whole difficulty.
  • Doubao Work shipped, and Feishu users are already annoyed. I flagged the launch Tuesday. Today a V2EX thread complains that Feishu now carries two sidebars and two Doubao chat boxes that cannot be deleted or hidden in software you pay for annually. Distribution advantage, meet distribution resentment.

Go deeper on this section: ClaudeChatGPT

The river · 34

China’s robot games begin to look like job interviews

# Zhidx 23h

At China’s humanoid-robot games, expanded workplace scenarios and heavier autonomy requirements shifted the test from running and dancing toward reliable work, with Agibot taking 18 golds.

More
  • Agibot won 18 gold, 16 silver, and 12 bronze medals, while the event expanded scenario competitions from six to 21 across factories, libraries, hotels, medicine, and emergency response.
  • Autonomous completion received full scoring weight while remote control received half, turning the competition into a rough benchmark for deployment readiness.
  • Agibot’s emergency-response stack combines a vision and instruction model, a task-planning agent, and reinforcement-learning tools for navigation, manipulation, and correction.
  • The library challenge required mapping, localization, book-spine recognition, transport, shelving, and error correction, a much more revealing workload than a single scripted movement.
  • The company says its G2 robots already conduct factory inspection and testing, linking competition performance to production data and operational experience.
  • The strategic advantage claimed by vertically integrated firms is the loop between body, real-world data, model, and deployment; the event is designed to reward exactly that loop.
  • The remaining question is whether performance in controlled competition transfers to long-term reliability, maintenance economics, and safety in customer sites.
Open at Zhidx Ask Claude ChatGPT

China-Nepal border flash flood leaves hundreds dead or missing

# Initium 7m

A glacial-collapse disaster at the Gyirong border area killed and displaced people on both sides, with Chinese and Nepalese casualty counts changing as rescue work continued.

Also: SCMP, Solidot

More
  • Chinese and international reports give different tallies and time stamps, but all describe a major mudslide and flood affecting the border crossing and downstream settlements.
  • The event disrupted roads, communications, and power on the Chinese side and destroyed infrastructure in Nepal.
  • Early scientific accounts point to a glacial collapse and meltwater conditions rather than a tectonic earthquake, though the exact trigger remains under investigation.
  • The disaster shows how border logistics, tourism, climate-sensitive mountain systems, and emergency coordination intersect in the Himalayas.
Open at Initium Ask Claude ChatGPT

LiveEdit brings causal diffusion video editing to 12.66 frames per second

# Leiphone 1d

Researchers at Tsinghua and Hong Kong University of Science and Technology built an open streaming editor that uses four denoising steps and cached unchanged tokens to reach 12.66 FPS.

More
  • LiveEdit replaces full-video bidirectional attention with chunk-wise causal attention, allowing output before future frames arrive.
  • A three-stage distillation process transfers high-quality editing behavior into the causal model and compresses 100 diffusion steps to four.
  • Its autoregressive mask cache predicts which spatial tokens changed between chunks and reuses self-attention features for the rest, pruning about 70% of redundant tokens in the reported configuration.
  • The choice to cache self-attention rather than feed-forward features reflects a systems tradeoff: temporal repetition is high in attention features, while feed-forward activations preserve high-frequency detail.
  • The framework starts from Wan2.1-T2V-1.3B and two thousand? The source says 20,000 high-quality pairs selected from Ditto-1M, making the training recipe relatively compact.
  • The result is important for live effects, video conferencing, and augmented reality because it addresses latency and continuity together rather than merely making offline generation faster.
Open at Leiphone Ask Claude ChatGPT

SenseTime reports first IFRS profit as token output scales

# Leiphone 22h

SenseTime reported first-half IFRS profit of 620 million yuan (about $87M), linking profitability to a 24-fold increase in daily token service and a broader agent platform.

More
  • The company says its token factory reached 2.4 trillion daily tokens in July and total operated compute reached 48,000 petaflops, though both are company metrics.
  • Its infrastructure stack combines model adaptation, inference optimization, heterogeneous scheduling, energy management, and delivery rather than treating compute as a simple GPU rental business.
  • SenseTime reports more than 1,000 enterprise customers, 4.5 billion? The source says 4,500 ten-thousand? Actually 45 million users for personal agents, and over 20 industries; the customer and user figures are not directly comparable.
  • The company says domestic-chip model utilization can reach 2.5 times the chip vendor baseline and that an energy-management agent forecast load with 96% accuracy.
  • The profit was helped by fair-value gains from ecosystem assets, so operating profitability should not be inferred solely from the headline result.
  • SenseTime’s long-term thesis is to charge for completed tasks rather than raw tokens, but that pricing transition remains ahead.
Open at Leiphone Ask Claude ChatGPT

China’s internet giants make a $169B annual AI infrastructure bet

# TMTPost 3h

Chinese internet companies are committing hundreds of billions of yuan to AI infrastructure, accepting near-term cash-flow pain in exchange for control of models, compute, and future applications.

More
  • Alibaba and Tencent spent more than 120 billion yuan (about $16.9B) in second-quarter capital expenditure, while ByteDance’s 2026 plan reached 200 billion yuan (about $28.2B).
  • Tencent’s 13.8 billion yuan (about $1.9B) negative free cash flow was its first reported negative result and was largely tied to 51.4 billion yuan (about $7.2B) in computing prepayments.
  • The companies represent four business models: Alibaba sells cloud capacity, Tencent internalizes compute into its ecosystem, Baidu sells model and token services, and ByteDance combines internal use with external infrastructure sales.
  • Alibaba’s AI cloud and computing-services revenue reportedly grew 45% to 48.4 billion yuan (about $6.8B), while adjusted profit fell 38%, offering the clearest early monetization but also showing investment drag.
  • Domestic chip substitution is being accelerated by export restrictions, with industry research cited as expecting Tencent’s domestic-chip procurement share to exceed 65%.
  • The key risk is utilization: if consumer AI does not produce a mass-market breakout, internalized compute at Tencent and strategic spending at ByteDance could face a long payback period.
  • The comparison with US cloud providers is about timing rather than simple technological superiority; China is building at high intensity while US infrastructure businesses are further along the harvest curve.
Open at TMTPost Ask Claude ChatGPT

China’s trade-in subsidies shift from old appliances to smart products

# TMTPost 2h

China has allocated 187.5 billion yuan (about $26.4B) for consumer trade-ins, and the 2026 program adds smart products such as glasses while narrowing support for older categories.

More
  • The Ministry of Finance says the funds have driven about 1.32 trillion yuan (roughly $186B) in sales and reached 178 million person-uses, implying more than seven yuan of sales per yuan of central funding.
  • Smart glasses receive a national 15% subsidy capped at 500 yuan (about $70) per item, while smart-home products are left more heavily to local experimentation.
  • The change from digital products to digital and smart products signals that Beijing is willing to subsidize market education for emerging devices, not only replacement demand.
  • The first month of smart-glasses subsidies produced only 7.2 million yuan (about $1.0M) in sales, but cited retail data show much faster growth over longer periods.
  • The program also narrows appliance coverage and standardizes national rules, indicating a preference for concentrated, measurable stimulus over broad support.
  • For manufacturers, the policy combines a demand boost with a test of whether AI-enabled devices can move beyond technology enthusiasts into ordinary households.
  • The important uncertainty is whether subsidy-driven trial converts into repeat usage after the discount disappears.
Open at TMTPost Ask Claude ChatGPT

Zhipu’s viral Ox Alpha model is GLM-5.3-Flash

# AI QbitAI 16h

Zhipu says the mysterious Ox Alpha model was its open-source GLM-5.3-Flash, which attracted heavy early usage and pairs native multimodality with a much lower price.

Also: Zhidx

More
  • The model has 320B total parameters and reportedly outperformed Zhipu’s 753B GLM-5.2 on selected tasks, illustrating that active-parameter efficiency can matter more than headline size.
  • The report says it reached 57 on an Artificial Analysis index and matched Claude Opus 4.8, but the comparison is not independently detailed in the supplied text.
  • Its ability to subtitle multi-speaker video, summarize a full film, and build interactive interfaces shows the product is being evaluated as an agentic tool rather than a chat model.
  • The claim that more than 62 trillion tokens ran on domestic accelerators makes supply-chain autonomy part of the launch narrative.
  • The anonymous test phase served as an unusual market experiment: external developers validated demand before the company disclosed the model’s identity.
Open at QbitAI Ask Claude ChatGPT

DeepSeek revenue reportedly reaches 475 million yuan in seven months

# Signal Zhidx 1d

DeepSeek reportedly generated 475 million yuan (about $67M) in January-to-July revenue while losing 715 million yuan (about $101M), illustrating the gap between model usage and infrastructure economics.

Also: Techmeme

More
  • The reported overall gross margin is 44.6%, with API services at 82.9%, suggesting that efficient inference can be profitable even while the company invests heavily in capacity.
  • AI infrastructure spending reportedly reached 11 billion yuan (about $1.5B), nearly ten times the 2025 figure, so margin strength does not remove the need for capital.
  • DeepSeek is also reported to be seeking 50 billion yuan (about $7.0B) at a 500 billion yuan (about $70.4B) valuation and preparing for a possible IPO.
  • Its August API price increases and peak/off-peak pricing indicate a move from growth-at-any-price toward monetizing scarce inference capacity.
  • The numbers are based on sources cited by The Information and were not confirmed by DeepSeek, so the financing and profitability conclusions remain provisional.
  • The strategic test is whether DeepSeek can turn low-cost model adoption into enough recurring revenue to fund infrastructure without losing its price advantage.
Open at Zhidx Ask Claude ChatGPT

Alibaba and Zhipu ignite another Chinese model price war

# AI Zhidx 16h

Alibaba’s Qwen3.8-Flash-Next and Zhipu’s GLM-5.3-Flash were released within hours of each other, combining open weights, sparse architectures, and prices far below premium foreign models.

More
  • Qwen3.8-Flash-Next is described as a 125B model with 6B active parameters, 262,144-token native context, and an extension path to one million tokens.
  • Alibaba reports training costs around one-ninth those of Qwen3.7-Plus and API pricing of 1 yuan input and 3 yuan output per million tokens.
  • Zhipu’s 320B GLM-5.3-Flash targets native multimodality and claims pricing as low as one-fortieth of Claude Opus 4.8 during a promotion.
  • Both releases emphasize agent coding, tool use, long-context work, and multimodal computer interaction, showing where Chinese model competition is moving.
  • The benchmark claims come largely from the vendors or affiliated testing, so the durable signal is architectural and pricing direction rather than a settled global ranking.
  • Open release of the Qwen architecture is also a community-validation strategy before the larger Qwen4 family.
Open at Zhidx Ask Claude ChatGPT

Meitu reports 39.5% profit growth as AI subscriptions expand

# Analysis Zhidx 20h

Meitu’s first-half adjusted profit rose 39.5% to 650 million yuan (about $92M), while AI productivity subscriptions and overseas usage became the clearest growth engines.

Also: QbitAI, TMTPost

More
  • Image and design products still produced 17.7 billion yuan (about $2.5B), or 80% of revenue, while paying subscribers reached 18.4 million.
  • Productivity-app monthly active users rose 43.5% to 33 million, and AI compute-point consumption increased about 113% year on year.
  • Vmake’s annual recurring revenue reached about $5M, while new products MVLAND and Picchi each exceeded $500,000 in recurring revenue.
  • The strategic question is whether Meitu can turn its aesthetic data and vertical workflows into durable professional willingness to pay rather than simply higher compute consumption.
  • Overseas growth is important because the company says new paying users are disproportionately coming from higher-ARPPU markets.
Open at Zhidx Ask Claude ChatGPT

Guming lets database agents handle alerts before calling a DBA

# AI GeekPark 1d

Chinese tea chain Guming put agents between more than 100 database instances and its small DBA team, cutting low-risk change approvals by 30% to 40% and reducing one incident to under five minutes.

More
  • The system combines metadata, slow-query history, lineage, risk scoring, and operational recommendations so a DBA can choose among indexing, throttling, and temporary scaling actions.
  • A data gateway parses and rewrites SQL, masks unauthorized fields, blocks dangerous operations, and records actions by session rather than granting one broad agent account.
  • The organizational change is as important as the automation: senior DBAs spend less time responding to alerts and more time on capacity planning, architecture, and upgrades.
  • The company reports that autonomous or semi-autonomous database changes now represent more than half of such work, but the source does not provide error or rollback rates.
  • Guming’s stated rule is gradual delegation: standardize first, keep people in the decision loop, and expand agent autonomy only after repeated evidence of safe behavior.
Open at GeekPark Ask Claude ChatGPT

Qwen3.8-Flash claims frontier performance with only 6B active parameters

# AI Leiphone 20h

Alibaba’s new open model combines sparse attention, gated residuals, and large N-gram embeddings to cut training and inference costs while targeting coding and agent workloads.

More
  • The model is described as a 125B multimodal mixture of experts with 6B active parameters per token and a context window expandable to one million tokens.
  • Alibaba reports 1 yuan input and 3 yuan output per million tokens, with training costs roughly 90% below Qwen3.7-Plus.
  • The architecture’s important detail is that the 51B embedding component serves as an efficient lookup structure without entering the per-token matrix computation.
  • The release makes efficiency and price a direct competitive weapon against both DeepSeek and premium Western APIs.
  • The benchmark comparisons are vendor-reported, so production adoption and independent evaluation will determine whether the claimed frontier position lasts.
Open at Leiphone Ask Claude ChatGPT

Pinduoduo’s 40% R&D increase signals a shift from traffic to supply chains

# Analysis InfoQ China 0m

PDD’s slower advertising growth and 40% jump in R&D spending point to a platform investing in traceability, manufacturing, compliance, and rural fulfillment rather than only traffic acquisition.

More
  • PDD reported second-quarter revenue of 112.4 billion yuan (roughly $15.8B), up 8%, while online marketing revenue grew only about 3% and transaction services grew 13%.
  • R&D rose to 4.6 billion yuan (roughly $650M), spread across governance, agricultural standards, factory upgrades, logistics, overseas compliance, and local delivery.
  • A lychee traceability deployment links farms, batches, testing, packing, shipping, product listings, and after-sales handling; the hard problem is data consistency behind the QR code.
  • A cited merchant saw repeat purchases rise from about 15% to 30%, but the source correctly notes that one merchant cannot establish a platform-wide causal effect.
  • The strategic shift is from matching buyers and sellers to shaping production and fulfillment, which creates durable operational leverage but produces less visible short-term revenue.
  • The model only matters if demand signals remain stable enough for factories and farms to justify process changes, and if governance rules become embedded in workflows rather than periodic cleanup campaigns.
Open at InfoQ China Ask Claude ChatGPT

Qwen Office rolls out Qwen3.8-Flash as its standard mode

# AI QbitAI 7h

Qwen Office is making Qwen3.8-Flash the default for routine work, claiming a 100% speed increase and 75% lower token consumption after model-harness co-optimization.

Also: InfoQ China

More
  • The product separates standard and advanced modes, reflecting an application-layer attempt to route most tasks to a cheaper model rather than expose users to a large model menu.
  • The reported improvements are based on internal real-office testing, with no public task mix or independent reproduction in the supplied report.
  • The broader significance is Chinese model providers optimizing for delivered work and token economics, not only benchmark intelligence.
Open at QbitAI Ask Claude ChatGPT

Siemens says industrial agents need plant knowledge, not a model wrapper

# AI QbitAI 6h

Siemens is positioning its industrial AI stack in China around engineering data, industrial protocols, and controlled workflows, arguing that factory agents must understand operations rather than merely chat.

More
  • Its Eigen Engineering Agent can read ECAD formats, identify conflicts, and generate PLC labels from hardware topology, linking electrical design to automation software.
  • The Intelligence Center X sits above PLM, ERP, MES, CRM, and operational technology systems to manage models, agents, skills, workflows, and decision traces.
  • The source claims two- to fivefold execution improvements and deployments in more than 100 companies, but those figures are vendor-reported.
  • The China relevance is deployment and localization: Siemens says the engineering agent is already being piloted with Chinese companies, where older equipment and fragmented protocols complicate rollout.
  • The strategic barrier is not model intelligence alone; it is the cost of mapping plant-specific semantics, permissions, safety boundaries, and validation into reusable components.
Open at QbitAI Ask Claude ChatGPT

Feihe deploys 436 agents across its dairy factories

# AI Leiphone 2h

Chinese dairy producer Feihe says 436 agents now support maintenance, process control, and personnel management, with a TPM system cutting average fault recovery from 104 to 47 hours.

More
  • The agents run on ByteDance’s HiAgent platform and are deployed across more than ten smart factories with over 90% digital production equipment.
  • The company reports more than 240,000 active users in three months and nearly 23 billion tokens consumed, but these are vendor-reported adoption figures.
  • Its AI TPM connects equipment data, spare-parts planning, budgets, and decisions across factories rather than offering a generic chatbot.
  • A membrane-cleaning project reportedly saves more than 500,000 yuan (about $70,000) annually, saves 144,000 tonnes of water, and avoids more than 3 million yuan (about $423,000) in equipment loss.
  • The organizational lesson is to let frontline workers identify repeatable problems while platform and security teams control deployment, permissions, and reuse.
Open at Leiphone Ask Claude ChatGPT

Four ministries target vehicle reliability, hype, and autonomous-driving claims

# Policy Jiemian 11m

A joint Chinese campaign will inspect vehicle quality and curb exaggerated marketing, while testing automated-driving systems, cloud security, AI safeguards, and accident logs.

Also: IT Home

More
  • The action runs nationally for one year and covers cars and motorcycles, with special attention to production consistency, reliability, durability, and validation of new technology.
  • Manufacturers must examine their own processes and suppliers, report defects, and initiate recalls when problems are found.
  • Regulators can publicly name companies and testing bodies, suspend product approvals or certification, block registrations, and impose fines for serious violations.
  • The policy explicitly treats marketing claims as part of product safety, an important distinction for driver-assistance systems whose capability boundaries are often communicated commercially.
  • The campaign also formalizes pressure on third-party testing organizations to resist non-rational competition and maintain objective, traceable evaluations.
Open at Jiemian Ask Claude ChatGPT

Wang Yi urges Washington to clear obstacles before Xi-Trump talks

# Policy SCMP 3h

China’s foreign minister called for fewer obstacles to high-level exchanges and a positive agenda before an expected Xi-Trump meeting, signaling diplomacy aimed at stabilizing ties while managing disputes.

More
  • Wang made the appeal to US ambassador David Perdue in Beijing, according to China’s foreign ministry.
  • The official framing emphasizes implementing existing leader-level consensus, managing differences, and keeping relations stable, healthy, and sustainable.
  • The statement is preparatory diplomacy rather than evidence that contentious issues such as trade, technology, or security have been resolved.
  • Its significance lies in the Chinese preference for treating summit logistics and political atmosphere as part of the strategic negotiation itself.
Open at SCMP Ask Claude ChatGPT

China’s robot expo shifts from demos toward human-machine trust

# Analysis Leiphone 1d

The World Robot Conference in Beijing added a buyer day and reported more than 200 million yuan (about $28M) in on-site sales, while speakers framed trust and real work as the next physical-AI bottlenecks.

More
  • The exhibition covered factories, homes, logistics, and service work, reflecting a Chinese industry moving from spectacular demonstrations toward deployment scenarios.
  • The article argues that cars may reach physical-AI adoption before humanoids because they already combine energy, sensors, actuators, compute, and repeated human interaction.
  • AIVA’s proposed vehicle concept uses the headlights as expressive interfaces, illustrating how Chinese product teams are treating interaction and trust as part of hardware architecture.
  • The event’s buyer day is a useful institutional signal: organizers are trying to connect demonstrations directly to procurement rather than relying on media attention.
  • The hard problem remains longitudinal trust—safe behavior, predictable boundaries, privacy, and recoverability over years of use—not whether an agent can complete one staged task.
Open at Leiphone Ask Claude ChatGPT

OceanBase leads China’s distributed database market with 20.2% share

# Signal Leiphone 2h

A Sullivan report puts China’s distributed transactional database market at 5.95 billion yuan (about $838M) and OceanBase first with 20.2%, as core industries move beyond pilots.

More
  • The report defines the category technically as horizontally scalable storage and compute with cross-node transaction consistency, strong consistency, high availability, and zero recovery-point loss.
  • Alibaba Cloud, Tencent Cloud, and Huawei Cloud follow with 17.9%, 16.5%, and 14.3%, making the market competitive rather than dominated by one vendor.
  • The market grew from about 1.69 billion yuan (roughly $238M) in 2021 to 5.52 billion yuan (about $778M) in 2025, with a projected 12.93 billion yuan (about $1.82B) by 2030.
  • Financial services and communications are entering scaled deployment, while government, energy, and transport are moving from peripheral to core systems.
  • The next competitive boundary is AI data infrastructure: real-time transactions, multimodal data, hybrid retrieval, analytics, and agent-facing services.
  • The market-share figure comes from an industry report and should be read as a vendor-defined category rather than a universal database census.
Open at Leiphone Ask Claude ChatGPT

Chinese coverage frames Meta’s child-safety settlement as a product reset

# Policy GeekPark 1h

Chinese analysis stresses Meta’s forced limits on recommendation, scrolling, and teen access, while US discussion frames the same settlement as Meta’s bid to lead child safety without admitting wrongdoing.

Also: Techmeme

More
  • The article says Meta may pay up to $18B over ten years, with most directed to states and youth-safety programs, though the final structure is complex.
  • Its most important point is that the settlement mandates product changes: time limits, night blocking, reduced notifications, private defaults, age detection, and a non-algorithmic feed.
  • Chinese coverage treats these constraints as an attack on the growth engine itself, while the US aggregator emphasizes Meta’s positioning and refusal to concede liability.
  • The proposed terms would matter beyond Meta if similar obligations reach TikTok and YouTube, because unilateral restrictions could simply shift teen usage to competitors.
  • The case is a precedent for regulating addictive design through enforceable product architecture rather than only fines or disclosure rules.
Open at GeekPark Ask Claude ChatGPT

Shenzhen faces a strategic test in the AI-for-science race

# Analysis SCMP 57m

The next AI contest may be shortening the full scientific loop from hypothesis to experiment, and Shenzhen is positioned in the Chinese coverage as a candidate to build that stack.

More
  • The article frames AI for science as a systems problem involving simulation, laboratory equipment, data collection, and iterative learning rather than a model leaderboard.
  • Shenzhen’s opportunity is to combine industrial capacity, research institutions, hardware, and deployment environments, though the excerpt does not identify a specific program or funding commitment.
  • The strategic comparison is with the United States’ effort to connect national laboratories into an integrated discovery platform.
  • Success would require durable links between models and physical instruments, not simply access to larger training clusters.
Open at SCMP Ask Claude ChatGPT

Chinese podcast listeners shift from AI capability to AI collaboration

# Culture QbitAI 23h

Xiaoyuzhou’s AI report says Chinese podcast discussion is moving from what models can do toward how people work with them and what remains uniquely human.

More
  • AI creators grew 187% and AI programs 239% in 2025, while searches for collaborative AI rose 227% in 2026, according to platform data.
  • The platform says discussion of AI and creativity grew sharply, with listeners focusing on taste, responsibility, experience, and judgment as generative tools become ordinary.
  • The data reflects one platform’s audience, not China’s entire AI workforce, but it offers a useful cultural signal about how early adopters are processing the transition.
  • The report also shows a very young technology audience, with users aged 19 to 25 making up nearly 40% of tech podcast listeners.
Open at QbitAI Ask Claude ChatGPT

Moonshot negotiates revenue sharing for Kimi K3 on US clouds

# Policy Solidot 23h

China’s Moonshot AI is reportedly discussing revenue-sharing deals with Microsoft, Amazon, and Google to host Kimi K3, a potential bridge between Chinese models and US cloud distribution.

More
  • The proposed arrangements could give Moonshot up to 30% of revenue from K3 services on Azure, AWS, and Google Cloud, though negotiations are early and may fail.
  • If completed, the deal would make US cloud platforms distributors of a Chinese flagship model rather than only infrastructure suppliers.
  • The arrangement highlights a market asymmetry: Chinese models can be attractive abroad on price and capability even while geopolitical controls complicate hardware access.
  • The main uncertainties are export-control review, liability, support, and how cloud providers would separate model access from restricted compute or data flows.
Open at Solidot Ask Claude ChatGPT

Huawei reaches the global top tier in enterprise storage

# Signal Zhidx 17h

A Gartner report places Huawei among the global top three enterprise-storage vendors and names it the only Chinese leader as AI workloads reshape file storage and data resilience.

More
  • The report emphasizes Huawei’s OceanStor architecture, including active-active file storage, high availability, ransomware detection, and AI-oriented data handling.
  • The strategic shift is from capacity and hardware price toward AI throughput, resilience, energy use, and supply-chain control.
  • Analyst rankings are not the same as market share, and the promotional article does not provide Gartner’s underlying scores or methodology.
Open at Zhidx Ask Claude ChatGPT

Chinese academic calls for maritime readiness in a future US rivalry

# Policy SCMP 1h

A prominent Chinese scholar argues that China needs a national consensus on sea power and should channel overseas investment returns into defense as maritime competition with the US intensifies.

More
  • The argument starts from China’s dependence on sea-borne trade and the United States’ existing maritime advantage.
  • It represents strategic commentary, not a government policy announcement, but its language reveals how some Chinese academics frame commercial shipping and naval power as one problem.
  • The policy direction implied is a broader fusion of economic security, overseas investment, trade-route protection, and defense spending.
Open at SCMP Ask Claude ChatGPT

Taiwan charges nine people over alleged AI-server resale to China

# Policy Initium 10h

Chinese coverage highlights Taiwan’s prosecution of nine people allegedly involved in reselling AI servers to China, while US coverage focuses on a Singapore logistics investigation in the same enforcement chain.

Also: Techmeme

More
  • The supplied excerpt is a translated headline and page shell, so it does not provide the charges, hardware, or evidence behind the prosecution.
  • Read with the US report, the event shows export-control enforcement reaching brokers, logistics providers, and intermediaries rather than stopping at formal chip sales.
  • The allegations should not be treated as proof that a specific Chinese organization received restricted hardware.
Open at Initium Ask Claude ChatGPT

China’s week: border disaster, summit diplomacy, and a new AI model

# Analysis Sinocism 10h

A China-focused newsletter groups the Gyirong disaster, Xi’s upcoming diplomacy, India relations, economic inspection, and a new domestic model, offering a useful agenda map rather than a single report.

More
  • The newsletter says Xi is expected to attend an SCO summit and visit Egypt, while a Politburo meeting may address the economy.
  • It also flags the border disaster, China-India relations, northeast economic inspection, and a new Chinese AI model as the week’s central files.
  • The item is valuable as editorial selection: it shows which events a China-focused briefing considers connected across politics, disaster response, and technology.
  • Because the supplied text is an introduction rather than the underlying stories, details beyond that agenda should not be inferred.
Open at Sinocism Ask Claude ChatGPT

Japan fast-tracks anti-drone procurement as China’s military grows

# Policy SCMP 5h

Japan selected a domestic interceptor drone in under three months, with analysts linking the procurement speed to the need to respond to China’s expanding military capabilities.

More
  • The system was chosen after competitive trials by Japan’s acquisition agency, showing a move toward shorter defense procurement cycles.
  • The China dimension is strategic rather than commercial: Tokyo is adapting acquisition processes to a faster regional threat environment.
  • The source does not establish the system’s battlefield effectiveness or whether rapid procurement will become a permanent Japanese defense model.
Open at SCMP Ask Claude ChatGPT

China’s AI boom exposes a gap between revenue, compute, and labor

# Culture Leiphone 9h

A Chinese tech roundup pairs DeepSeek’s fast-growing but loss-making business with telecom layoffs, Tencent’s compute shortage, and disputes over pay, revealing the labor and capital stress behind the AI race.

More
  • The DeepSeek section reports 475 million yuan (about $67M) in first-seven-month revenue against a 715 million yuan (about $101M) loss, with heavy infrastructure spending.
  • Tencent executive Dowson Tong says Hunyuan was rebuilt repeatedly and that insufficient compute slowed training and product development, arguing that endurance may matter more than an early start.
  • The roundup also carries unverified employee accounts of layoffs and pay cuts at China Telecom’s cloud unit and a dispute over shifting part of salaries into performance pay at SF Express.
  • The article mixes reported disclosures, employee claims, and commentary, so each item has a different evidentiary status.
  • Taken together, the selection shows AI investment competing with cost-cutting and weaker labor protections inside Chinese technology firms.
Open at Leiphone Ask Claude ChatGPT

China launches a bamboo-framed floating solar platform

# Signal SCMP 8h

China deployed a 100-kilowatt floating solar demonstration platform using composite bamboo frames off Shandong, linking renewable generation to hydrogen, ammonia, and methanol production.

More
  • The platform is a demonstration project rather than a commercial-scale power plant, so durability, maintenance, and lifecycle economics remain unproven.
  • Using bamboo composites aims to reduce material impact and potentially create a lower-carbon alternative to conventional floating structures.
  • Its integration with chemical-fuel production makes the project relevant to China’s broader push to couple renewable power with industrial energy carriers.
Open at SCMP Ask Claude ChatGPT

Chinese web fiction debates whether pro-women writing becomes a new orthodoxy

# Culture Initium 20h

A Chinese podcast examines the rise of fiction centered on women’s agency and asks when shared standards for avoiding misogyny become another form of creative censorship.

More
  • The trend rejects misogynistic language and romance plots centered on women’s submission, while foregrounding female characters as agents of the story.
  • The debate is not simply progressive versus conservative: readers and writers are also arguing over whether women’s desire, power, and conflict have one acceptable expression.
  • The topic reveals how online literary communities turn moral and political expectations into genre conventions, moderation norms, and reputational pressure.
  • The podcast format is interpretive rather than a survey, so it illustrates discourse and argument more than prevalence.
Open at Initium Ask Claude ChatGPT

Jimeng launches a film label to fund and distribute AI-native productions

# Culture Leiphone 1d

Chinese AI video platform Jimeng is creating a film label that offers selected projects compute, technology, funding, and distribution support, trying to build a full commercial path for AI-native cinema.

More
  • The program targets feature films and series rather than short demonstrations, with tiered support based on project quality, production needs, and distribution prospects.
  • Support may include model access, technical guidance, copyright protection, theater or streaming introductions, revenue sharing, sponsorship, and derivative IP development.
  • The initiative addresses a bottleneck that generation tools do not solve: who owns, distributes, markets, and monetizes a finished AI production.
  • Early partners include Chinese animation and film organizations, but no completed project has yet demonstrated audience demand or sustainable economics.
  • The move suggests Chinese model platforms are seeking defensibility through content ecosystems and rights relationships, not only through model quality.
Open at Leiphone Ask Claude ChatGPT

Greentown’s legacy property inventory falls below half its stock

# Policy 36Kr 7h

Greentown China’s first-half profit collapsed while older inventory fell below 50% of unsold value, showing a developer trading near-term margins for balance-sheet cleanup.

More
  • Revenue fell 26% to 39.5 billion yuan (about $5.6B), while attributable profit dropped 61% to only 81.7 million yuan (about $11.5M).
  • The company held about 221.5 billion yuan (roughly $31.2B) of unsold value, including 106 billion yuan (about $14.9B) from projects acquired in 2021 or earlier.
  • Greentown recorded 19.1 billion yuan (about $2.7B) in impairment and fair-value losses, including 11.9 billion yuan (about $1.7B) on inventory.
  • Management says liquidity and leverage are comparatively safe, but margins remain under pressure and another round of impairments depends on selling conditions and local price competition.
  • The company still targets about 100 billion yuan (roughly $14.1B) of new saleable value, creating tension between cleaning old stock and bidding for expensive new land.
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 8/10
36Kr 200 10/30
Solidot 200 13/20
Leiphone 200 17/20
GeekPark 200 3/30
IT Home 200 12/60 +44 over cap
SCMP 200 10/50 +18 over cap
Initium 200 6/15 +1 over cap
Jiemian 200 8/30 +16 over cap
China Digital Times 200 0/15
The Wire China 200 0/50
Techmeme 200 4/15
Hacker News 200 0/20
V2EX 200 8/50 +25 over cap
Zhidx 200 9/20
InfoQ China 200 8/20 +12 over cap
New Wisdom 200 8/15 +3 over cap
TMTPost 200 8/17 +8 over cap