China Desk

China News, Summarized 15 Aug 2026, 01:44 UTC 34 stories 18/19 sources

Today's brief

China's most valuable company now sells memory, and Tencent's own earnings explain the swap

A Hefei DRAM maker passed Tencent

CXMT is now the biggest listed company in China. Seventeen days after its Shanghai debut, the Hefei memory maker closed Friday at roughly 3.58 trillion yuan of market value against Tencent's HK$4.0 trillion — about 3.44 trillion yuan, which is the ranking Chinese outlets are actually computing. Bloomberg put Thursday's crossover at $524bn to $511bn.

The crossover was Tencent falling, not CXMT rising. CXMT's own shares fell 1.2% on the day; the swap happened because Tencent dropped further after disclosing a 176% surge in AI capital spending. Allspring's Gary Tan gave the line of the week: "chips are the new clicks."

Tencent published the bill on page seven. Marketing services revenue hit 43.565bn yuan (~$6.1bn), up 22% against a Chinese internet ad market that grew 4.6% last year on QuestMobile's count — and gross margin on that line slipped from 58% to 57%. Ad serving used to carry near-zero marginal cost. Now every impression burns GPU.

Strip the AI products out and the company looks untouched. Remove Yuanbao, CodeBuddy, WorkBuddy and the rest and non-IFRS operating profit goes from 75.6bn to 86.1bn yuan — 19% growth instead of 12%. That 10.5bn yuan (~$1.5bn) gap is one quarter's cover charge. Depreciation inside fintech and business services rose 44%.

The tell is who Tencent buys from. It signed a $3bn server DRAM agreement with CXMT in June; ByteDance signed a five-year deal worth more than $7bn in July, and server products went from 8.4% of CXMT's revenue in 2024 to 26.5% in 2025 — investors are penalising the buyer of AI hardware while rewarding its supplier, over the same purchase orders. Wednesday's note on Tencent hoarding memory now has a scoreboard.

Go deeper on this section: ClaudeChatGPT

Apple gets a licence, drones get a tariff

Apple trained a model for China only. Reuters reports Apple built a China-specific LLM with Alibaba's support, which would make it the first foreign company to offer a proprietary AI model in the country. The Cyberspace Administration has registered Apple's generative service, clearing Apple Intelligence for mainland iPhones after two years of watching Huawei eat its share with AI handsets.

The registration is the story, not the model. Beijing has spent that time insisting every model filed and approved; the first foreign firm through the gate got there with a domestic partner attached. Buy the licence, share the training.

Washington answered with drones. The White House imposed tariffs on unmanned aircraft and components, including a 100% levy on "particularly sensitive" models. Second day running that the trade move is American and the mainland feed doesn't mention it.

And Beijing quietly shut the offshore trust door. New Ministry of Finance and tax administration rules issued 24 July tax offshore trusts at asset transfer, at income generated inside, and at distribution, against an industry estimate of 2.3 to 2.7 trillion yuan (~$324–380bn) in such structures. Li Daokui, Tsinghua finance professor and dean of its economic-practice centre, welcomed them — then noted capital income tops out at 20% while labour income reaches 45%: "That's clearly unfair" — and proposed capping all personal income tax at 20%. The fairness argument runs both ways, and the second way is a tax cut for the people the first way just caught.

Go deeper on this section: ClaudeChatGPT

GLM-5.3 found a flaw forty years old

Zhipu shipped GLM-5.3 overnight on the same base model as 5.2 — 744bn total, 40bn active, IndexShare architecture — with everything else coming from post-training. Terminal-Bench 3.0 went from 4.6 to 28.3 and DeepSWE v1.1 from 46.2 to 66.9; on Zhipu's own Code Bench it beat Claude Opus 4.8 at the hardest tier, 31.4% to 29.5%, using about 50,000 tokens per task where Opus needs 120,000.

The security numbers are the actual release. CyberGym rose to 84.5% from 5.2's 77.2%, above Anthropic's Mythos 5 and OpenAI's GPT-5.6. Working with Tsinghua, Nankai and Chinese security labs, Zhipu says it found 2,436 vulnerabilities across 269 projects, some up to 40 years old, 1,097 of them medium-to-high severity, now in the national vulnerability database's repair queue. The oldest sits in the design of DNS itself: a few crafted requests amplify server load nearly 80,000-fold.

The other case reads like fiction. A tracking engine flagged an odd temporary file server in Brazil; with the model in the loop, researchers reconstructed an autonomous AI attacker they call Neo — four servers, seven malicious domains, 60,000 mailboxes, over 100 attack scripts and 18,567 phishing emails aimed at accounting and tax firms, all inside two months. Defence AI unpicking offence AI, and the Chinese write-ups know exactly how that sounds.

Go deeper on this section: ClaudeChatGPT

Alibaba sells the shortcut, ByteDance refuses one

Alibaba Cloud brought thirty carmakers into its own data centre. The Ulanqab site carries 60% of China's intelligent-driving R&D compute, and the pitch was that model, chip and cloud used to be three suppliers at three different conferences. What they are selling is not GPU hours but a pipeline: raw clips in from the fleet, automatic cleaning, vectorisation, model-assisted labelling, millions of clips a day, model iteration measured in days.

The argument underneath is the good one. Stored data depreciates when the architecture switches from end-to-end to vision-language-action; the pipeline doesn't. Meanwhile T-Head's line on superpods — that most on sale are Ethernet server stacks and the only real test is unified addressing with synchronous memory semantics — is a fair shot, and their 64-card single-layer symmetric design is the answer they want judged.

ByteDance is still paying to avoid the shortcut. 36Kr's researchers describe distillation as a cold start that carries a model to 90 so you can work the last 10. Seed banned it at founding and Zhang Yiming reaffirmed that in late July even if it means trailing domestic rivals — Seed-2.1 Pro sits 16th on Code Arena's WebDev board against Kimi K3 at 2. DeepSeek's V4 report describes distilling a dozen domain experts into the generalist; K3 used nine teachers.

The split is by modality, and it explains everything. Seed's video model reached the frontier without distillation because generated video can't substitute for real footage. Language is precisely where borrowing works — which is precisely where ByteDance is behind. Nvidia's Jensen Huang calls learning from other models the foundation of intelligence; Zhang is betting a year of leaderboard humiliation that it's a ceiling.

Go deeper on this section: ClaudeChatGPT

What the filter blocks, and what it lets through

A WeChat essay on censorship is arguing that censorship is incompetent. The post, translated by China Digital Times, makes a careful case: no public directory of blocked terms exists and regulations don't require platforms to restrict ordinary commentary; platforms over-block because automated filters can recognise characters but not context, so they err on the side of caution and keep expanding the lists. Users invent homophones, platforms block those too, and the vocabulary keeps shrinking.

It is still online, which is the point. Six weeks after the Jieyang dog case, the sayable critique is that the machinery is clumsy, not that it is wrong. Initium meanwhile followed the grief across the border, to a Buddhist rite held in Hong Kong for the dog and the invisible limits of cross-border solidarity.

Elsewhere, the filter had a day off. Zhejiang Provincial Museum's Long March display turned the Red Army into a "red car" amid the red-tourism push. And during Wednesday's total eclipse, Xiaomi 17 Ultra photos of the sun came back with ridges and craters — the camera's reconstruction model knew what a bright disc in a dark sky is supposed to look like, and it was wrong.

Go deeper on this section: ClaudeChatGPT

Same model, two verdicts on the weights

Chinese and Western coverage of GLM-5.3 disagree on what happened. Leiphone's three labels — strongest security model, strongest coding model, strongest open-source model — and QbitAI's framing of Tang Jie's promise to Musk moving forward both treat open weights as a settled fact.

They aren't. The weights ship in two weeks, once security reviews wrap up — the first GLM release held back explicitly for safety review. English coverage led there, and on Zhipu adding vulnerability-discovery environments expecting better bug-spotting and getting a model that reasons across whole exploit chains, which it says was not the intended outcome. That admission is the most interesting sentence in the launch, and it's absent from the Chinese write-ups I read.

Go deeper on this section: ClaudeChatGPT

Threads we are pulling

  • DeepSeek's price rise still lands 16:00 UTC Sunday. Harness passed 50,000 GitHub stars in twelve hours, and a third-party comparison circulating among developers claims the same V4-Flash burns roughly three times the tokens in DeepSeek's harness as in a rival open one. Worth testing before Monday's invoice.
  • I under-weighted memory. I read the price rise as load-shedding by electricity. CXMT's re-rating and ByteDance's $7bn DRAM contract say the binding shortage is DRAM at least as much as power.
  • Zhu Rongji: still no funeral grade. SCMP ran a seven-story retrospective; Initium's podcast keeps the frame on centralisation, not liberalisation.
  • Alibaba's promised 27bn-parameter dense Qwen3.8 companion: day three, nothing.
  • New watch — Moonshot. It has reported fake "friend funds" and "reserved allocations" to police, its second such notice since June, after closing an F round at a $35bn valuation and opening a pre-IPO round early at $50bn pre-money. When the counterfeit share market gets busy, the listing is close.

Go deeper on this section: ClaudeChatGPT

The river · 34

Chinese coverage frames DeepSeek Harness as an Android-like base for agents: local-first, MIT-licensed and built so models, tools, UI and sandbox can all be replaced.

Also: V2EX

More
  • The developer preview runs locally through a command-line launcher, leaving sessions, logs and data on the user's machine rather than making the product a hosted service.
  • Its central design is that everything is a plugin, including the model, tools, interface, approval policy and agent loop; this makes the framework closer to an operating substrate than a finished assistant.
  • DeepSeek's own test reportedly used less than three dollars of tokens to redesign a website and plot the repository's growth, but the account also calls the interface rough and unfriendly to non-programmers.
  • The first twelve hours produced more than 50,000 GitHub stars, an attention signal rather than proof of sustained usage or ecosystem quality.
  • The strategic move is to open the control plane around an agent, allowing developers to assemble private workflows and swap models as capability and pricing change.
  • The missing pieces are governance, stable plugin interfaces and safe execution defaults; without them, extensibility can become an integration burden rather than a platform advantage.

New Chinese rules tax offshore trusts at transfer, income and distribution stages, tightening enforcement as fiscal pressure makes wealthy taxpayers more valuable.

More
  • The Ministry of Finance and tax authority rules target offshore trusts used by Chinese tax residents to defer or reduce personal income tax; the supplied account says domestic offshore-trust assets are estimated at 2.3 trillion yuan (roughly $324 billion).
  • The policy is designed to tax the lifecycle of a structure rather than wait for a final distribution, reducing the room for trusts to function as indefinite deferral vehicles.
  • A Tsinghua finance professor argues the move improves fairness but also notes that labor income can face a top rate of 45 percent while capital income tops out at 20 percent.
  • That framing reveals the political tradeoff: Beijing wants more compliance from wealthy households without necessarily making capital income tax rates match labor income.
  • The measure matters beyond the wealthy because it signals a broader shift toward taxing cross-border assets and using information and reporting rules to strengthen revenue collection.

Chinese researchers report a fourfold extension of the cold-atom entanglement distance, pushing an experiment toward the scale needed for intercity quantum networking.

More
  • The team reports entanglement between cold atoms separated by 420 kilometers, more than four times previous demonstrations, according to a paper in Physical Review Letters.
  • The distance exceeds the range at which direct transmission faces fundamental limits, making the result relevant to quantum-repeater and distributed-network research.
  • This is not yet a functioning quantum network: the experiment establishes a physical test bed, while memory lifetime, photon loss, synchronization and repeaters remain major engineering barriers.
  • The result matters for China because it combines laboratory quantum information research with the country's broader emphasis on long-distance secure communications infrastructure.
  • Its value will be clearer if later work demonstrates higher rates, longer coherence and a repeatable multi-node architecture rather than a one-off link.

Alibaba Cloud is positioning itself as the production line for Chinese autonomous driving, arguing that data processing and model iteration now matter as much as chips.

More
  • The company says its pipeline handles multimodal ingestion, cleaning, vectorization, automated labeling and sample retrieval, with capacity of about one million driving clips per day.
  • The architectural pitch is a closed loop from vehicle data to training and evaluation, rather than a standalone storage or GPU service; Alibaba says distributed scheduling improved processing speed by an order of magnitude over traditional self-built systems.
  • Alibaba is also offering Qwen as a possible vision-language-action base and Wanxiang as a world-model engine for generating rare scenarios and evaluating responses.
  • This shifts the competitive bottleneck from owning GPUs to converting messy real-world data into high-quality training samples at a repeatable cadence.
  • The numbers are vendor claims, and the key proof will be whether customer models improve in safety and release speed rather than merely in pipeline throughput.

A China-linked collider result challenges the simple quark-and-gluon picture by indicating a structure that carries baryon number without electric charge or color charge.

More
  • The researchers inferred the structure from precise measurements of net baryon number and net electric charge in high-energy nucleus collisions, according to the report.
  • The finding does not mean the standard model has been overturned; it means the simplified picture used to describe nucleon interiors may be incomplete.
  • A baryon-number-carrying, electrically neutral and color-neutral component would force theorists to revisit how composite structure emerges from quarks, antiquarks and gluons.
  • The result is important for China’s research system because it shows domestic and international teams contributing to a foundational physics question through large experimental infrastructure.
  • Independent replication and theoretical reconciliation will determine whether this becomes a durable new description or an anomaly in the measurement or interpretation.

Chinese coverage stresses Beijing’s unease with Washington’s internal disorder, while the broader story concerns the risks facing a planned Xi-Trump summit.

More
  • The report says Chinese officials were frustrated by disorganized US planning before an earlier Trump visit and now see similar uncertainty ahead of Xi's planned Washington trip.
  • The concern is operationally important for diplomacy: leadership-level meetings require sequencing, security and policy commitments that cannot be improvised without raising political costs.
  • For Beijing, American institutional coherence is itself a strategic variable; uncertainty can delay commitments even when both sides want a summit.
  • The account is based partly on unnamed sources and describes a planned trip, so the timing and political significance remain contingent.

A second venture by BOE founder Wang Dongsheng is seeking a Hong Kong listing after losing 4.9 billion yuan (roughly $690 million) in three years to pursue domestic RISC-V and display-interface chips.

More
  • Esi Computing raised 9.276 billion yuan (roughly $1.31 billion) across four rounds, yet reported annual losses of 1.837 billion, 1.547 billion and 1.516 billion yuan from 2023 through 2025.
  • Its 2025 gross profit of 453 million yuan (about $64 million) did not cover 604 million yuan in administrative expenses, before 1.042 billion yuan in research spending.
  • The company reported 24.31 billion yuan (about $3.4 billion) in 2025 revenue but remained loss-making, with an adjusted loss rate of 47.7 percent and an expected loss in 2026.
  • One customer supplied 82.1 percent of revenue in 2023, declining to 39.3 percent in the first quarter of 2026; the unnamed profile appears consistent with BOE, but that identification is an inference rather than a disclosed fact.
  • The case shows the cost of building a domestic architecture alternative: national strategic importance can attract capital long before product economics or customer diversification are proven.
  • Its cash position and monthly burn make the IPO a financing necessity as much as a growth narrative.

Chinese coverage highlights a university spinout’s full-stack photonic quantum strategy, from quantum random-number chips to programmable machines.

More
  • Silicon Zhen Chip raised more than 100 million yuan (roughly $14 million) in a Series B led by industrial investors, with proceeds aimed at a 10,000-bit specialized photonic cluster and universal quantum chips.
  • The company says its programmable photonic computer is cloud-accessible and the only domestic photonic system it knows of that runs universal quantum algorithms.
  • Its technical path uses silicon photonics, integrated devices and room-temperature operation, shifting the challenge from cryogenic physics toward optical loss, fabrication and packaging.
  • The company reports stable on-chip generation of a four-photon, 16-qubit GHZ state and a four-qubit cluster state, plus a Grover search demonstration with a reported 98.7 percent average identification probability.
  • Its near-term commercial anchor is a quantum random-number chip already ordered at about 200,000 units, while photonic computing remains a long-horizon engineering bet.
  • The financing is strategically important because the founders say reaching million-qubit systems requires deep cooperation with a wafer foundry, not just more laboratory research.

China’s new environmental code enters force, consolidating more than 30 laws and a large body of administrative and local rules into one framework.

More
  • The code contains five sections and 1,242 articles, incorporating 30-plus environmental laws, more than 100 administrative regulations and over 1,000 local regulations.
  • Ten major stand-alone laws, including rules on air, water, soil and marine pollution, are folded into the code and cease to operate separately.
  • This is a governance move as much as a legal one: it gives regulators a single organizing framework for environmental enforcement across jurisdictions and media.
  • Codification can reduce contradictions and make compliance easier to map, but its practical effect will depend on implementing rules, local enforcement and the agencies' willingness to impose costs.
  • The law also signals that environmental protection remains a formal state priority even as China balances industrial growth, energy security and local fiscal pressure.

Chinese coverage highlights GLM-5.3’s coding and vulnerability-finding claims, while US discussion focuses on its narrow CyberGym edge over Anthropic’s restricted model.

Also: Leiphone, IT Home, Techmeme

More
  • Z.ai reports an 84.5 percent CyberGym score, up from 77.2 percent for GLM-5.2 and slightly above Mythos 5; the figures are vendor claims awaiting broader replication.
  • The model reportedly found 2,404 deduplicated vulnerabilities across 220 projects, including 1,088 medium- or high-risk cases, with some legacy bugs dating back decades.
  • Its coding gains come from post-training in realistic long-horizon environments, where the model must inspect code, use tools, run tests and close a task rather than answer a static question.
  • The model weights are scheduled to open after safety evaluation and hardening, while the most sensitive security functions are restricted to verified users.
  • The release is a notable Chinese attempt to turn open source, cybersecurity and agentic coding into one competitive narrative.

Chinese coverage stresses Qwen 3.8’s 27-billion-parameter multimodal design, long context and local deployment, while US discussion focuses on developer reaction and model discoverability.

Also: Leiphone, Hacker News

More
  • Alibaba says the dense model supports a native 262,000-token context window, extendable to 1 million tokens through YaRN, and adds a reasoning-effort control for trading quality against compute.
  • The weights use Apache 2.0, allowing commercial deployment, and the company positions quantized inference on consumer GPUs as a practical route to local use.
  • The architectural story is less about a larger base model than about packaging multimodality, long context and adjustable inference depth into a deployable size.
  • Alibaba claims gains over Qwen 3.6 27B in coding and office tasks, but the supplied item does not provide independent benchmark results.
  • The release is strategically important because a permissive, locally runnable model can reduce dependence on hosted APIs and expand the global developer surface for a Chinese model family.

A Chinese robotics firm says its compute demand has risen five- to tenfold in two years, while cloud workstations are becoming infrastructure for simulation, data and deployment.

More
  • The bottleneck in embodied AI is not only model size: teams must coordinate teleoperation data, simulation, training, hardware testing and multi-user development environments.
  • The company uses a cloud workstation with Isaac Sim and related tools so teams can obtain standardized GPU environments without buying idle local machines.
  • The technical pain is systems integration: CUDA, ROS2, simulators, drivers and data permissions create version and access problems that multiply as teams grow.
  • The cloud approach turns the workstation into a reproducible environment that can scale by task, which is more valuable for robotics than simply renting an isolated GPU.
  • The demand figures and claimed deployment improvements come from the company and cloud provider, but they capture a genuine shift from laboratory experiments to continuous engineering pipelines.

A Chinese humanoid-model company and dexterous-hand maker are defining data standards that include torso, arm and gaze coordination rather than hand motion alone.

More
  • The partners say current manipulation datasets often record only hand degrees of freedom, making it difficult for models to learn stable whole-body coordination.
  • Their proposed standard will align capture formats, labels and evaluation, with a training baseline intended to make results reproducible across hardware.
  • The technical importance is representational: a hand action is conditioned by posture, reach, visual attention and contact forces, so isolated hand trajectories omit the control problem that matters in the real world.
  • The companies plan to validate the loop from data gloves to model training to deployment on direct-drive dexterous hands, targeting tasks such as connectors and threaded fasteners.
  • A shared baseline could reduce repeated integration work across Chinese robotics teams, although the announcement supplies no released dataset or independent performance result.

Chinese coverage argues that blunt platform filters, not only explicit state bans, force users into coded language and make ordinary online speech hard to publish.

More
  • The translated essay says users routinely replace ordinary words with homophones, initials and coded terms after posts or comments are silently blocked by automated moderation.
  • It attributes much of the overblocking to platforms managing legal and political liability, rather than to a publicly enumerated list of forbidden terms.
  • That distinction matters: companies have some discretion in how they meet broad content obligations, so the everyday experience of censorship is shaped by risk-averse product and moderation choices.
  • The result is an online language layer optimized for filter avoidance, which reduces clarity and makes social discussion harder even when the underlying topic is not illegal.
  • The essay itself remains online and frames the problem as platform negligence, a depoliticized argument that may be more publishable than a direct challenge to the censorship system.

What China's AI labor market is actually paying for

# Culture Ginger River Review 16h

Chinese hiring demand is shifting from star credentials to scarce abilities in post-training, infrastructure, embodied AI and product delivery, according to a Chinese recruiting discussion.

More
  • The discussion says million-yuan packages for new PhDs exist but are exceptional; the more valuable signal is which skills companies repeatedly struggle to hire.
  • Recruiters see demand across chips, infrastructure, foundation models, robotics and agents, indicating that China's AI labor market is broadening beyond model researchers.
  • Post-training, evaluation, data construction and systems engineering command attention because they convert a base model into a usable product and are harder to import through a single star hire.
  • The conversation also shows how Chinese companies are trying to hire people who bridge research and deployment, rather than preserving a strict boundary between academia and engineering.
  • This labor-market shape helps explain why companies continue to reorganize teams and recruit from one another even when headline compensation becomes less sustainable.

Chinese social media celebrates the first two Chinese nationals to win Fields Medals, while debate turns to elite migration and domestic academic support.

More
  • Wang Hong and Deng Yu, both educated partly in China and now based at US or European institutions, received the prize for work in analysis and partial differential equations.
  • The awards generated national pride online, but discussion also questioned how much credit belongs to Chinese universities when the winners completed major parts of their training and careers abroad.
  • Wang is the first Chinese woman to win the medal and only the third woman overall, giving the event a gender dimension beyond national achievement.
  • The reaction exposes a recurring tension in Chinese science policy: official narratives celebrate global distinction while public discussion asks why top researchers often need foreign institutions to reach the frontier.
  • The story is therefore about academic mobility and institutional confidence as much as mathematics.

Qianhe Yibang raised more than 2 billion yuan (roughly $280 million) to commercialize 3D DRAM in-memory computing, targeting the memory wall with a vertically integrated architecture.

More
  • The company combines stacked DRAM with compute in three dimensions, aiming to reduce the data movement that limits conventional von Neumann systems.
  • It says the team began work on a 3D DRAM compute-in-memory chip in 2018 and was incubated by NetEase before becoming an independent chip company.
  • The financing includes state and industrial funds, including a China Mobile-linked fund, giving the company potential customer access as well as capital.
  • The architecture is promising for bandwidth-bound workloads, but the report provides no independently verified performance, yield or production metrics.
  • The strategic bet is that co-designing memory, compute and software can matter more for AI serving and cloud gaming than simply scaling conventional accelerators.

A US appeals court sent DJI’s blacklist challenge back for reconsideration, keeping a Chinese drone maker inside a long-running US national-security dispute.

More
  • The court did not remove DJI from the Pentagon list; it gave the company another opportunity to contest the basis for its inclusion in lower court.
  • DJI's case shows how US restrictions can persist through administrative and judicial processes even without a new sanctions announcement.
  • For Chinese technology companies, the practical issue is not only access to US markets but also the reputational and financing effects of being labeled as linked to the Chinese military.
  • The ruling matters if the lower court demands a stronger evidentiary record or clarifies how much deference the Pentagon receives when making company designations.

Chinese coverage treats ChangXin’s market-cap jump above Tencent as a vote for AI memory hardware over internet platforms, though the comparison is volatile and currency-dependent.

More
  • Solidot reports ChangXin at roughly 3.58 trillion yuan (about $504 billion) after an A-share close, while Tencent stood near 4.0 trillion Hong Kong dollars (about $507 billion); the two figures are close and exchange-rate sensitive.
  • ChangXin is described as the world's fourth-largest DRAM supplier, placing memory manufacturing at the center of China's AI infrastructure narrative.
  • The reversal reflects investor preference for scarce AI hardware exposure, but a stock-market ranking is not evidence that ChangXin has surpassed Tencent in operating scale or profitability.
  • The signal matters if domestic DRAM capacity continues to attract capital and narrows China's dependence on foreign memory suppliers; otherwise it may be a speculative listing effect.

Chinese state television confirms that the Y-20B can be converted between heavy transport and aerial-refueling roles, extending the PLA Air Force’s reach beyond the first island chain.

More
  • The Y-20B uses domestically developed WS-20 turbofans and can accept pallet-ready fuel tanks and removable refueling equipment, according to the report.
  • A convertible airframe gives China more flexibility than a tanker fleet designed for one mission, although conversion time, sortie availability and crew training will determine its real utility.
  • The capability supports longer-range fighter operations and logistics in the western Pacific, where basing access and tanker capacity constrain sustained air activity.
  • It also reflects a broader Chinese military-industrial pattern: extracting multiple operational roles from a common airframe to improve production and deployment economics.
  • The evidence confirms technical potential, not the size or readiness of an operational tanker force.

Apple and Alibaba reportedly trained a model for China, showing how local regulation and fierce handset competition are reshaping Apple’s global AI architecture.

More
  • The report says Alibaba supported training a model intended for Apple Intelligence features on devices sold in China, where Apple must work within domestic generative-AI rules.
  • The arrangement lets Apple preserve product integration and user experience while outsourcing part of the model layer to a Chinese partner with local data and compliance infrastructure.
  • Chinese rivals such as Huawei have used AI features to compete more aggressively in smartphones, increasing the commercial pressure on Apple to launch a credible domestic experience.
  • China's registration of Apple's generative-AI service is described as a step toward availability, but the item does not establish a launch date or the final division of technical responsibility.

China's best music of 2026 so far

# Culture ChinaTalk 15h

A China-focused music roundup highlights a lively independent scene spanning jazz, emo, post-punk and avant-garde collaboration.

More
  • The selection ranges from Shenzhen indie rock and Chinese post-punk offshoots to a Brooklyn-formed jazz group, showing how local scenes absorb and remix global forms.
  • One anecdote about a Shanghai street livestream being interrupted by police also points to the uneven space available for informal performance and public music-making.
  • The value here is cultural rather than commercial: it offers a map of current Chinese music communities that mainstream platform charts often obscure.

Chinese analysts warn that the new World Artificial Intelligence Cooperation Organisation should not become an anti-US bloc, revealing Beijing’s concern about competing technology coalitions.

More
  • The analysts say the organization is not intended as an alternative to the US-led Pax Silica partnership, but worry that outsiders may interpret it that way.
  • The advice reflects a practical diplomatic goal: attract countries that want access to Chinese technology and standards without forcing them to choose between Washington and Beijing.
  • For China, an international AI body is useful only if it gains membership and legitimacy beyond existing political partners; an explicitly rival identity would narrow that market.
  • The episode shows AI governance becoming an arena for coalition-building, standards influence and dependency management rather than only safety rules.

SenseTime argues that data centers selling model output need a metric linking electricity to useful tokens, not just the conventional power usage effectiveness ratio.

More
  • The proposed TPW metric, Tokens Per Watt, combines compute efficiency, energy use, model efficiency and time-of-use electricity pricing into one business-oriented measure.
  • The company describes an eight-level telemetry system linking jobs, GPUs, racks and building power so operators can attribute electricity to specific workloads.
  • An orchestration agent could shift latency-tolerant training, evaluation and batch jobs toward cheaper or less-constrained periods while preserving real-time inference.
  • SenseTime reports that its Shanghai Lingang facility increased effective IT load capacity by about 60 percent without adding power capacity; this is a company claim and is not independently substantiated in the item.
  • TPW is useful only if teams agree on what counts as a useful token and measure output quality, not merely raw token volume.

Chinese coverage says Microsoft has closed at least 15 offices or joint ventures, while its China business survives through multinational-facing services and access to engineering talent.

More
  • Microsoft says it remains committed to China and operates under the regulatory environment applied to international suppliers, while the report describes a five-year contraction in local operations.
  • Export controls make cloud and AI expansion harder, and Chinese customers increasingly prefer domestic software, narrowing the market for Microsoft's direct product strategy.
  • The remaining niche is strategically unusual: Chinese companies such as ByteDance still need Western technology management for overseas operations, creating demand even as domestic use declines.
  • Microsoft also values China's engineering talent, showing why a foreign technology company may retain a smaller local presence despite geopolitical and commercial headwinds.
  • The story is a model of partial decoupling rather than clean withdrawal: sales, operations and research can move in different directions.

SHEIN generated about $6 billion in operating cash over three years, yet its IPO is driven partly by $17.3 billion of preferred shares and looming redemption rights.

More
  • The company's operating cash flow was $1.757 billion, $1.405 billion and $2.838 billion in 2023, 2024 and 2025, while cash resources reached $14.831 billion by March 2026.
  • Its cash conversion is unusually strong for retail: customers pay before fulfillment, and net cash conversion days stayed around negative 12 to negative 13, with inventory turning in 36 days in 2025.
  • The hidden pressure is contractual rather than operational. Convertible redeemable preferred shares totaled $17.294 billion, and certain investors' return rate rose from 8 percent to 12 percent in March 2026.
  • The IPO therefore functions as a way to convert private claims into a public-market path and prevent redemption rights from becoming a liquidity problem, not simply as a fundraise.
  • The business is still growing in orders and customers, but revenue growth and US sales have slowed, making the public valuation more sensitive to durability than to raw scale.

Chinese coverage emphasizes mainland activity around Taiwan, while the US account focuses on a new coastguard training partnership.

More
  • The two-week program covers small-boat operations and team coordination, placing the cooperation below the threshold of a formal military deployment while still improving maritime interoperability.
  • Beijing is likely to read such training through the lens of cumulative presence and signaling rather than the narrow content of the exercises.
  • The development matters because coastguards operate in the gray zone between civilian law enforcement and military coercion, where escalation can be managed or amplified through routine encounters.
  • The account does not establish that the training changes the regional balance by itself; its significance lies in the pattern of recurring US-Taiwan operational ties.

The reported sale of Alibaba’s Lingxi game unit illustrates Chinese internet groups treating mature content businesses as cash-flow assets while redirecting capital toward cloud and AI.

More
  • Bloomberg sources say private-equity firm Centurium Capital is the leading bidder at a valuation above $1.5 billion; neither side had confirmed the deal in the supplied account.
  • Lingxi's main asset is the long-running strategy game Three Kingdoms Tactics, whose reported monthly revenue was roughly 180 million to 250 million yuan (about $25 million to $35 million) in 2025.
  • The business has stable cash generation but has not produced a second product on the same scale, making it easier to value as a mature asset than as a growth platform.
  • Alibaba's internal reporting change and broader disposal of non-core holdings fit a larger retreat from ecosystem expansion toward AI, cloud and fewer strategic priorities.
  • For China's game industry, the lesson is that capital and traffic cannot substitute for a durable content-production system; a single hit can support a business but not necessarily a conglomerate strategy.

Moonshot AI’s rumored Hong Kong IPO would turn Kimi’s temporary frontier-model lead into capital, but the company’s own compute limits expose the fragility of that advantage.

More
  • Market reports have placed a possible offering at about $3 billion, though the company denied one reported filing timetable and no filing is confirmed in the supplied account.
  • Kimi K3 briefly ranked near the top of an external model leaderboard with lower reported task costs, giving the company a stronger public-market narrative than its earlier releases.
  • Demand then approached cluster capacity, prompting a temporary pause for new consumer subscriptions; that episode shows the difference between benchmark leadership and serviceable capacity.
  • The core funding use would likely be compute and product expansion, not survival alone: a model that cannot serve demand cannot turn technical attention into recurring revenue.
  • The risk is rapid commoditization. Competitors can copy validated post-training and agent techniques, while a cheaper model such as DeepSeek can erode the pricing premium before an IPO closes.

Chinese-language coverage traces Pikmin Bloom’s rapid rise in Taiwan and Hong Kong to social-media momentum and routines built around walking.

More
  • The game uses walking, flower planting and encounters with nearby players as its core loop, turning ordinary movement into a shared social activity.
  • The report says online discussion began accelerating in late 2025 and grew sharply through the spring, with one monitored month rising 94.4 percent.
  • Players kept completing walking and planting tasks even during a typhoon, while others negotiated how to preserve the ritual indoors, showing how game rules can reshape daily behavior.
  • Its rise illustrates an Asian consumer pattern in which social platforms, location-aware play and lightweight routines can revive an older mobile title without a conventional blockbuster launch.

China and Brazil are preparing a satellite intended to provide continuous environmental and disaster monitoring across South America, extending their long-running space partnership.

More
  • Brazil's space research institute says the satellite will monitor the whole continent and improve weather and environmental information.
  • The project gives China a civilian infrastructure role in South America that is harder to frame as a purely commercial export or military partnership.
  • For Brazil, the value is regional observation capacity; for China, it is an opportunity to make Chinese space hardware and data services part of public-sector infrastructure abroad.
  • The supplied account does not give a launch date, financing structure or instrument specifications, so the operational capability remains prospective.

Baidu is packaging a large document corpus and enterprise workflow agents into a standalone office product, with claimed monthly office users above 25 million.

Also: Leiphone

More
  • Baidu says the product connects 1.8 billion professional documents, 700 million academic resources and users' private drive data to a mixture-of-experts agent system.
  • The release emphasizes parallel Office agents, long-term memory, cross-device execution and enterprise auditing rather than a single chat benchmark.
  • Its finance mode includes research reports, market data and spreadsheet skills, showing how Chinese vendors are using proprietary content and workflow templates to defend office AI.
  • The 25 million monthly-user figure is company-reported, and the account does not distinguish active task completion from casual feature use.

Chinese museums are being criticized for basic caption errors in Communist Party history displays, undercutting an official push for revolutionary tourism.

More
  • One Zhejiang museum caption reportedly turned the Red Army into a red car, a small error with outsized political sensitivity because the exhibition concerns the Long March.
  • The mistakes reveal the gap between top-down historical messaging and the mundane quality-control work needed to make that messaging credible.
  • They also show how online criticism can target state institutions through administrative competence rather than overt ideological dissent.
  • The embarrassment is greater because museums and red tourism sites are expected to serve as trusted infrastructure for patriotic education.

Chinese demand and Alibaba’s licensing platform have become central to Sanrio’s turnaround, with Kuromi overtaking Hello Kitty in China.

More
  • Alibaba's IP platform holds exclusive mainland rights to 26 Sanrio characters and helped more than 200 brands launch over 30,000 licensed products in 2024.
  • Sanrio's China sales reached 31.44 billion yen (roughly $210 million) in its latest fiscal year, up 83.2 percent, while stores in China reached 59 by the end of 2025.
  • The market is not simply importing a Japanese character: Alibaba supplies local licensing, channels and merchandising operations, turning a foreign IP catalog into a Chinese retail system.
  • Kuromi's lead over Hello Kitty shows how character portfolios can reduce dependence on a single legacy asset and align better with local consumer taste.
  • The model contrasts with Pop Mart's heavier direct-retail approach; Sanrio remains more dependent on licensing, which scales faster but leaves less control over execution.
Source ledger
ChinAI 403 0/0 HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk 200 1/20
Pekingnology 200 1/20
Ginger River Review 200 1/20
Sinocism 200 0/20
QbitAI 200 8/10
36Kr 200 9/30
Solidot 200 11/20
Leiphone 200 8/20
GeekPark 200 4/30
IT Home 200 12/60 +41 over cap
SCMP 200 10/50 +5 over cap
Initium 200 3/15
Jiemian 200 8/30 +19 over cap
China Digital Times 200 3/15
The Wire China 200 0/50
Techmeme 200 2/15
Hacker News 200 1/20
V2EX 200 8/50 +40 over cap