China Desk — 15 Aug 2026, 04:40 UTC

Alibaba Cloud says autonomous driving’s bottleneck is the data pipeline

# GeekPark 18h

Alibaba Cloud is selling a full data-to-model-to-compute pipeline for Chinese autonomous driving, arguing that production engineering matters more than simply buying GPUs.

The platform connects multimodal vehicle data ingestion, cleaning, vectorization, automatic labeling, sample retrieval and training rather than treating annotation as a separate manual stage.
Alibaba says it can process millions of clips per day and improve processing speed by an order of magnitude over traditional self-built workflows, but these are vendor claims without an external benchmark.
The strategic shift is from model architecture alone to a closed-loop data factory: rare edge cases must be generated, labeled, evaluated and fed back into training quickly enough to matter.
The company positions its own foundation model as a possible VLA base and its world model as a generator of driving scenarios and tests, creating a vertically integrated cloud proposition.
Historical driving data may lose value as model paradigms change, while the production pipeline remains useful across architectures; that is Alibaba’s strongest strategic argument.
The forum’s location inside Alibaba’s data center is itself a signal that compute, data governance and model training are being sold as one operating system to automakers.
The key unanswered question is whether customers will accept dependence on one cloud provider for proprietary driving data and model infrastructure.

Open at GeekPark Ask Claude ChatGPT

What China’s AI talent market is really paying for

# Ginger River Review 19h

China’s AI hiring boom rewards scarce systems, post-training and product skills more broadly than celebrity researchers, revealing how the labor market is maturing beyond headline salaries.

The reported 3 million yuan, roughly $420,000, offer for a new PhD is real but described as exceptional rather than normal, with compensation concentrated among a small group of candidates.
Recruiters say demand spans chips, infrastructure, foundation models, embodied AI, robotics and agents, showing that China’s shortage is distributed across the stack.
The valuable profile is shifting from pure model research toward people who can connect data, evaluation, inference systems and products into a repeatable engineering loop.
High compensation also reflects organizational urgency: firms are buying time and execution capacity because internal training pipelines cannot quickly produce experienced post-training and systems engineers.
The conversation comes from a Chinese business podcast and recruiter, so it is qualitative market texture rather than a statistically representative wage survey.
For executives, the signal is that talent competition may increasingly be won through team design, compute access and research autonomy rather than salary alone.

Open at Ginger River Review Ask Claude ChatGPT

Chinese team entangles cold atoms across a record 420 kilometres

# SCMP 17h

Chinese scientists report 420-kilometre cold-atom entanglement, more than four times prior demonstrations and a step toward intercity quantum networks.

The experiment was published in Physical Review Letters, according to the report, and exceeds the distance at which direct transmission faces fundamental limits.
The result concerns entanglement over a long baseline, not a completed quantum repeater network or a demonstrated faster-than-light communication system.
Its engineering significance is as a test bed for distributing quantum states between separated nodes, where loss, synchronization and memory lifetime are the practical constraints.
A 420-kilometre span is relevant to China’s existing interest in wide-area quantum communications because it moves the experiment closer to network-scale geography.
The supplied report does not provide the fidelity, repetition rate or channel conditions needed to judge readiness for a deployable system.

Open at SCMP Ask Claude ChatGPT

BOE veteran’s second chip venture heads for Hong Kong IPO

# 36Kr 23h

Display-industry veteran Wang Dongsheng is taking Eesway Computing public after three years of losses totaling 4.9 billion yuan, roughly $690 million, to fund a RISC-V alternative bet.

The company raised 9.276 billion yuan, roughly $1.31 billion, across four rounds but ended 2025 with only 188 million yuan, about $26 million, in cash and a reported monthly burn near 96.8 million yuan.
Revenue grew from 1.752 billion yuan to 2.431 billion yuan between 2023 and 2025, yet 2025 gross profit of 453 million yuan did not cover 604 million yuan of administrative expense, let alone 1.042 billion yuan of research spending.
Its RISC-V market position is small in the overall market and becomes a domestic-leader claim only after foreign suppliers are excluded, illustrating how Chinese semiconductor rankings often depend on the definition of local content.
One customer supplied 82.1% of 2023 revenue, falling to 39.3% in the first quarter of 2026; the unidentified customer appears likely to be BOE, but the filing does not name it.
Cash recovered in early 2026 mainly through borrowing, asset sales and collections rather than a demonstrated operating turnaround.
The company is making a long-horizon architecture bet while its current economics remain fragile, creating a direct test of how far state and strategic capital will fund domestic substitution.
The IPO will need to persuade investors that RISC-V control and future scale can overcome concentrated customers, continuing losses and a thin gross-margin base.

Open at 36Kr Ask Claude ChatGPT

Two Chinese mathematicians win Fields Medals, sparking debate at home

# Culture China Digital Times 5h

Chinese coverage celebrates the first two Chinese nationals to win Fields Medals while online debate asks whether China’s academic system supports talent as well as it claims.

Wang Hong, now based at New York University and the Institut des Hautes Etudes Scientifiques, was recognized for work in harmonic analysis and geometric measure theory and is the third woman ever to win the medal.
Deng Yu, now at the University of Chicago, was recognized for work in partial differential equations and Hilbert’s sixth problem.
Both winners studied or completed major degrees abroad, complicating simple nationalist claims that the achievement belongs wholly to domestic universities.
The domestic reaction reportedly combines pride with concern about academic incentives, support for young researchers and whether institutions are better at celebrating talent after it succeeds than nurturing it beforehand.
The episode illustrates how scientific prestige in China is simultaneously a source of national pride and a prompt for criticism of the education and research system.

Open at China Digital Times Ask Claude ChatGPT

GLM-5.3 claims a major post-training leap in coding and security

# AI Leiphone 17h

Chinese coverage presents GLM-5.3 as a post-training breakthrough, while Western discussion focuses on comparable security benchmarks and restrictions on sensitive capabilities.

Also: Techmeme

Z.ai says the base model is unchanged and attributes the gains to post-training, reinforcement learning and its IndexShare and Slime training systems.
The company reports Terminal-Bench 3.0 rising from 4.6 to 28.3 and DeepSWE from 46.2 to 66.9, but the supplied article does not establish evaluation parity or independent replication.
Its security claims include finding 2,436 vulnerabilities across 269 projects, with 1,097 rated medium or high severity; these are vendor-reported results and need case-level validation.
The product’s defense architecture combines a lightweight classifier, intent monitoring during reasoning and model-level safety alignment, a notable design for open deployment where server-side filters disappear.
The strategic message is that post-training and realistic tool environments may now matter more than swapping the base architecture for each release.
The most important uncertainty is whether the gains hold outside the company’s own benchmark and workflow choices.

Z.ai says GLM-5.3 nearly matches a restricted security model Techmeme

Open at Leiphone Ask Claude ChatGPT

Alibaba releases Qwen 3.8 with 27 billion parameters

# AI Leiphone 13h

Chinese coverage emphasizes Qwen 3.8’s Apache 2.0 licensing and local deployment, while developer discussion focuses on whether the model’s benchmark claims translate into useful everyday work.

Also: Hacker News

The 27-billion-parameter model is described as natively multimodal and dense, with quantized versions intended to run on consumer GPUs.
Alibaba says Qwen models have exceeded 3 billion cumulative downloads and generated more than 300,000 derivative models, though these are company or public-platform totals rather than active-user measures.
An Apache 2.0 license lowers redistribution and commercial-use friction, making the model strategically important for firms that want local control over inference and data.
The source calls it superior to the previous Plus model in coding and office tasks, but supplies no independent benchmark table or hardware requirements.
The release continues Alibaba’s effort to make open weights a distribution channel for its model ecosystem rather than relying only on hosted APIs.

Hacker News debates Alibaba’s Qwen 3.8 release Hacker News

Open at Leiphone Ask Claude ChatGPT

DeepSeek Harness is an agent toolkit, not a finished product

# AI GeekPark 1d

DeepSeek Harness passed 50,000 GitHub stars in 12 hours; Chinese testing stresses its plugin architecture, local execution and unusually low cost, while warning that it is still developer-only roughware.

Also: Jiemian, V2EX

The system runs locally through a Node command, with the browser acting as a shell over local sessions, logs and data rather than as a hosted agent service.
Its deepest architectural choice is that the model, tools, interface, approval rules and main execution loop can all be replaced, making the Harness closer to an operating system than a fixed assistant.
The tool can reportedly complete multi-step website and API tasks, but the test also found rough interactions and a poor fit for non-programmers.
A plugin ecosystem appeared almost immediately, including design and interface extensions, supporting the thesis that agent value may come from composition rather than one monolithic model.
The reported token cost was below three dollars for the test tasks, but that figure is a local experiment and says little about reliability, review time or total engineering cost.
The project’s growth rate is a distribution signal, not proof of a stable platform; the interfaces and core plugins are explicitly expected to change rapidly.
If DeepSeek maintains compatibility while others build plugins, Harness could become a control layer for interchangeable models and tools; if not, the early ecosystem may fragment.

Open at GeekPark Ask Claude ChatGPT

China’s best music of 2026 so far

# Culture ChinaTalk 18h

A China-focused music roundup maps a lively 2026 scene across indie rock, avant-garde music and jazz, including a case where livestreaming drew police attention.

The selection spans acts from Shenzhen and other Chinese scenes, showing a culture ecosystem that is broader than the mainstream pop market.
The report notes that a Shanghai street livestream by a band triggered police trouble, a small but revealing reminder that public performance and online distribution remain regulated spaces.
The roundup is editorial rather than a market dataset, but it offers useful texture on how Chinese artists build audiences through Bandcamp, social media and touring communities.

Open at ChinaTalk Ask Claude ChatGPT

Chinese 3D compute-in-memory chip firm raises $282 million

# AI Leiphone 1d

NetEase-incubated Qianhe Yibang raised more than 2 billion yuan, roughly $282 million, to commercialize 3D DRAM compute-in-memory chips aimed at the memory wall.

The company combines stacked DRAM with compute in three dimensions, attempting to reduce the data movement imposed by conventional von Neumann separation of memory and processing.
Its pitch is system-level co-design: chip architecture, software stack and cloud-gaming or application-acceleration workloads are developed together rather than treating memory as a standalone component.
The financing includes state and industrial investors, including a national fund and a China Mobile-linked fund, suggesting strategic demand and capital support are part of the commercialization plan.
The company traces its work to a 2018 exploration of 3D DRAM compute-in-memory and says the new venture was established in 2024.
The announcement gives no silicon benchmarks, manufacturing node, yield, customer revenue or comparison with high-bandwidth memory, so the architecture remains an industrial bet rather than a demonstrated AI accelerator.

Open at Leiphone Ask Claude ChatGPT

Beijing worries Trump’s White House chaos could disrupt Xi’s visit

# Policy SCMP 5h

Chinese coverage stresses Beijing’s unease with Washington’s disorganization ahead of Xi Jinping’s planned visit, while the US-facing angle is the diplomatic risk to summit preparation.

The report says Xi’s planned Washington trip would be his first state visit to the US since 2015, giving procedural uncertainty unusually high political weight.
Chinese officials reportedly saw similar planning problems before Trump’s earlier China visit, so the concern is not only personality-driven but about whether the US system can execute commitments.
For Beijing, a chaotic counterpart complicates signaling: concessions may be temporary, staff-level agreements may not survive, and public promises may be reversed by the next factional dispute.
The article relies partly on unnamed sources, and the supplied text does not establish that the visit will occur or that a specific policy dispute caused the anxiety.

Open at SCMP Ask Claude ChatGPT

China and Brazil plan a 24-hour disaster-warning satellite

# Policy SCMP 14h

China and Brazil unveiled a satellite project intended to monitor all of South America around the clock, extending their long-running space partnership into regional climate infrastructure.

Brazil’s national space research institute says the satellite will provide weather and environmental monitoring across South America.
The project matters strategically because it turns Chinese space cooperation into a public-service capability with direct relevance to disaster response, agriculture and climate observation.
After nearly four decades of cooperation, the partnership shows how China builds influence through infrastructure that is useful to a partner’s whole region rather than only through bilateral trade.
The supplied report does not give a launch date, instrument package, financing structure or data-governance terms, so the operational impact remains prospective.

Open at SCMP Ask Claude ChatGPT

DJI gets another chance to challenge its Pentagon blacklist listing

# Policy SCMP 8h

A US appeals court sent DJI’s challenge back for reconsideration, reopening a legal route without removing the Chinese drone maker from the Pentagon list.

The ruling does not clear DJI; it requires the lower court to reconsider part of the case, leaving the blacklist designation in place for now.
The dispute shows how Chinese technology companies face a US national-security process that can impose commercial and reputational costs before the underlying allegations are finally resolved.
For DJI, the strategic issue is access to US government business and the legal precedent for challenging executive-branch company designations.
The case also demonstrates that litigation can slow, but not necessarily reverse, the fragmentation of the global drone market along security lines.
The supplied report does not state what evidence the lower court must reassess or predict the outcome.

Open at SCMP Ask Claude ChatGPT

Pikmin Bloom turns walking, flowers and strangers into a daily ritual

# Culture Initium 5h

Pikmin Bloom has surged in Taiwan and Hong Kong, where its walking mechanics and social encounters are reshaping ordinary routines and online communities.

The game’s social layer turns nearby players into anonymous collaborators, with shared flower trails and postcards creating lightweight ties between strangers.
Usage reportedly accelerated from late 2025 into early 2026, with online discussion rising as much as 94.4% month over month during the strongest period.
Players continued organizing walks and community events during a typhoon, while others debated how to complete step goals indoors, showing how game rules can structure behavior beyond the screen.
The story is a useful consumer-culture signal: a low-pressure game can become sticky when it links digital progress to place, movement and recurring social rituals.

Open at Initium Ask Claude ChatGPT

Chinese magnetic-sensor work could detect submarines 500 metres below

# Analysis SCMP 1h

Chinese research groups have built a Hall-effect magnetic sensor aimed at detecting weak underwater anomalies, with possible defense and navigation implications.

The work comes from institutes of the Chinese Academy of Sciences in Hefei and Ningbo, according to the supplied report.
The article’s central claim is a possible smartwatch-scale application for detecting a submarine about 500 metres below the surface, but the excerpt does not provide sensitivity, noise-floor or field-test data.
The strategic value would lie in distributed, low-cost magnetic sensing rather than a single exquisite platform, if the laboratory result survives marine interference and motion.
The supplied text is truncated before the architecture and measured performance are explained, so the operational claim should be treated as prospective rather than demonstrated.

Open at SCMP Ask Claude ChatGPT

SenseTime proposes Tokens Per Watt for AI infrastructure

# AI Leiphone 15h

SenseTime argues that data centers should optimize useful tokens per watt, not just PUE, and reports a 60% increase in effective IT capacity at an existing site.

The proposed TPW metric links electricity, GPU utilization, task scheduling and model output, attempting to give power and compute teams one optimization target.
Its architectural premise is to map physical power data to jobs, pods, nodes and model tasks so operators can see which workloads consume electricity and whether they can move in time.
The scheduling strategy places latency-tolerant training, evaluation and batch work into cheaper or less constrained periods while protecting online inference.
SenseTime says its system can raise effective IT capacity by about 60% without adding power capacity, but the supplied article does not provide a baseline, workload mix or independent validation.
TPW is best understood as a commercial operating framework proposed by a vendor, not yet an industry standard.

Open at Leiphone Ask Claude ChatGPT

Moonshot AI’s IPO needs a new story beyond model rankings

# Analysis 36Kr 1d

Moonshot AI is reportedly preparing for a Hong Kong IPO as Kimi K3 gives it a performance window, but scarce compute and fast model catch-up make timing more important than ever.

Earlier reports put potential fundraising near $3 billion, but Moonshot denied one filing rumor and the supplied article does not confirm a timetable.
Kimi K3 reportedly ranked near the top of Artificial Analysis while costing less than leading Western models, giving investors a clearer performance narrative than a generic low-price pitch.
DeepSeek’s newer model reportedly costs less than one-tenth as much per task, showing how quickly a Chinese rival can attack a temporary technical and commercial advantage.
Kimi temporarily paused new consumer subscriptions after demand approached cluster capacity, exposing compute availability as a constraint on turning benchmark strength into usage and revenue.
Recent fundraising by listed Chinese model companies demonstrates that an IPO can become a continuing financing channel, not merely a one-time cash event.
The market is likely to reward a company seen as near the frontier at the moment of listing, but the same mechanism creates pressure to list before the lead disappears.
The unresolved question is whether Moonshot can convert high-end model quality into durable distribution, enterprise contracts and sufficient inference supply.

Open at 36Kr Ask Claude ChatGPT

US coastguards train Taiwan amid increased Chinese activity

# Policy SCMP 16h

Chinese activity around Taiwan is prompting a new US coastguard training program focused on small-boat operations and coordination.

The two-week program is led by US instructors and involves Taiwan Coast Guard personnel, according to Taiwan’s agency.
The emphasis on law-enforcement-style maritime skills points to gray-zone competition, where coercion and presence operations may matter as much as naval combat.
For Beijing, greater US-Taiwan coastguard cooperation blurs the line between civilian maritime security and military signaling.
The supplied report gives no detail on the Chinese activity, the number of US personnel or whether the training is a recurring arrangement.

Open at SCMP Ask Claude ChatGPT

China confirms Y-20B can convert into an aerial tanker

# Policy SCMP 18h

China has confirmed that its Y-20B transport aircraft can accept palletized fuel tanks and become an aerial refuelling tanker, extending PLA air operations beyond the first island chain.

The Y-20B uses domestically developed WS-20 turbofans and can reportedly switch between transport and refuelling roles rather than requiring a dedicated tanker fleet.
A convertible design trades some mission specialization for fleet flexibility, allowing transport aircraft to be reconfigured as operational needs change.
The capability could extend fighter range, persistence and sortie generation in the western Pacific, but its value depends on tanker numbers, receiver compatibility and crew training.
The disclosure came through state broadcaster CCTV, making it both a technical announcement and a strategic signal intended for foreign observers.
The supplied report does not provide fuel capacity, conversion time, tanker inventory or evidence of operational deployment.

Open at SCMP Ask Claude ChatGPT

Apple and Alibaba train a China-specific model for Apple Intelligence

# Policy Solidot 21h

Chinese coverage says Apple and Alibaba trained a China-specific model, a major concession to local data and regulatory constraints as Apple prepares to bring generative AI to mainland iPhones.

Apple is reportedly using Alibaba as a local partner because its standard Apple Intelligence stack cannot simply be deployed unchanged in China’s regulatory and market environment.
China’s cyberspace regulator has reportedly registered Apple’s generative-AI service, clearing an important gate for a domestic launch.
The partnership lets Apple retain control over the device experience while outsourcing part of the model and compliance burden to a Chinese ecosystem company.
It also shows the strategic cost of China’s model rules: global products may need region-specific training, data arrangements and service architecture rather than a single worldwide AI backend.
Apple faces rising pressure from Huawei and other domestic phone makers whose AI features are already marketed as part of the device proposition.
The report does not disclose the model architecture, training data, division of responsibility or launch date, so claims about technical quality remain premature.

Open at Solidot Ask Claude ChatGPT

Chinese photonic-quantum company raises more than $14 million

# AI 36Kr 3h

A Chinese Academy of Sciences spinout raised more than 100 million yuan, roughly $14 million, to commercialize integrated photonic quantum computers and chips.

Silicon Zhen says it has built a programmable photonic quantum computer accessible through the cloud and internally validated a dedicated 5,000-bit photonic computing device.
The company reports stable on-chip generation of a four-photon, 16-qubit GHZ state and a four-qubit single-photon cluster state, plus a Grover search demonstration with a reported average identification probability of 98.7%.
Its near-term commercial anchor is a quantum random-number generator chip, with about 200,000 units ordered and deployments reported in SIM cards, utility meters and home cameras.
The architecture’s appeal is engineering-oriented: silicon photonics can operate at room temperature and potentially use mature manufacturing, but loss, fabrication variation and photonic error correction remain the scale bottlenecks.
The company’s claim to be China’s only general-purpose photonic quantum-computing system maker is a source claim, not an independently established industry ranking.

Open at 36Kr Ask Claude ChatGPT

SHEIN’s IPO is about a redemption clock, not fresh cash

# Analysis 36Kr 23h

SHEIN’s strong cash generation masks a $17.3 billion preferred-share redemption overhang, making an IPO a deadline-driven capital-structure event rather than a funding necessity.

The company generated $6 billion of operating cash flow across 2023 to 2025 and held $14.8 billion in cash at March 2026, while its historical equity funding totaled about $3.84 billion plus a smaller yuan round.
The pressure comes from $17.294 billion of convertible redeemable preferred shares; return rates for later investors reportedly rose from 8% to 12% in March 2026.
If the IPO is withdrawn, rejected or not relaunched within the specified period, redemption rights can revive, giving investors leverage even though the operating business is cash generative.
SHEIN’s operating engine remains unusually efficient: negative working-capital days, 36-day inventory turnover and consumer prepayments fund growth before suppliers are paid.
The risk is deceleration rather than immediate liquidity: US revenue fell 3.5% in 2025 and 14.3% year over year in the first quarter of 2026, while overall growth slowed sharply.
The company’s valuation already reset from a $98.2 billion pre-money round to $64 billion in the next financing, showing how a strong cash machine can still lose market confidence.
The IPO narrative therefore has to reconcile a profitable cross-border supply chain with slowing growth, US exposure and investor redemption mechanics.

Open at 36Kr Ask Claude ChatGPT

US students are questioning the standard China-threat playbook

# Analysis SCMP 2h

Chinese coverage highlights skepticism among future US policymakers toward treating Beijing mainly as an adversary, rather than simply repeating Washington’s tougher-China consensus.

The report points to classrooms at Georgetown, Rice and other US universities as an arena where the next generation of foreign-policy professionals is debating the default strategic frame.
This is an early elite-attitude signal, not evidence that US policy has shifted; students entering government will still face institutional, security and alliance pressures.
For Beijing, the distinction matters because it separates hostility toward the Chinese state from broader engagement with Chinese society and the consequences of interdependence.
The story is valuable chiefly as a view of how Chinese media interpret possible fractures in the US policy consensus.

Open at SCMP Ask Claude ChatGPT

ChangXin briefly overtakes Tencent in China market-value rankings

# Signal Solidot 18h

A reported market-value reversal puts memory maker ChangXin near Tencent, reflecting investor enthusiasm for AI hardware but also exposing how headline comparisons can mislead.

The source gives ChangXin a value of 3.58 trillion yuan, roughly $504 billion, and Tencent about 4.0 trillion Hong Kong dollars, roughly $510 billion; those figures do not clearly establish that ChangXin overtook Tencent.
ChangXin is described as the world’s fourth-largest DRAM supplier, placing a Chinese memory company directly in the investment narrative around AI servers and high-bandwidth memory demand.
The contrast with Tencent suggests capital is rewarding physical AI infrastructure over established internet platforms, at least in this market snapshot.
Market capitalization is a volatile investor signal, not evidence that ChangXin has higher revenue, profit or technological leadership than Tencent.
The comparison also depends on currency conversion, share-price timing and whether the figures refer to the same close, which the supplied item does not clarify.

Open at Solidot Ask Claude ChatGPT

Caption errors embarrass Chinese museums at Party-history exhibits

# Culture SCMP 19h

Chinese museums are facing criticism over caption mistakes in Communist Party-themed displays, exposing the gap between official historical messaging and institutional execution.

One Zhejiang museum reportedly rendered the Red Army as a red car in a Long March exhibit, an especially damaging error because revolutionary history receives high political attention.
The mistakes arrive during a wider push for red tourism, in which museums and attractions are expected to turn Party history into an educational and commercial experience.
The episode shows how centralized narrative control still depends on ordinary local institutions, translators, editors and contractors getting details right.
Public criticism is possible because the errors undermine official prestige, but the supplied report does not describe disciplinary action or a broader audit.

Open at SCMP Ask Claude ChatGPT

China’s travel apps try to turn intent into a bookable itinerary

# Culture 36Kr 14h

Chinese consumers want AI to handle travel logistics, but testing finds that itinerary agents still hallucinate hotels and schedules even as Alibaba’s Fliggy automates booking workflows.

The product approach is different from a chat assistant: it maps free-form requests and social-media guides onto transport, hotels, tickets, budgets and a confirmation step.
The reported test generated a three-day Xi’an itinerary, corrected route details and produced bookable options without requiring the user to switch apps.
The hard problem is not prose generation but temporal and spatial consistency: a plausible hotel, train or museum rule can still make the whole trip impossible when combined with other facts.
The article’s central consumer insight is that Chinese users are tired of fragmented platform labor and will value an agent only if it removes checking work rather than creating more of it.
The product remains cautious around payments and needs manual confirmation for obscure locations, so the current boundary is assisted execution, not full delegation.

Open at 36Kr Ask Claude ChatGPT

Xiaomi phone camera appears to turn the Sun into the Moon

# AI Solidot 19h

A Xiaomi phone reportedly added crater-like detail to a solar-eclipse image, reviving the question of whether computational photography is reconstructing scenes rather than recording them.

The report says a Xiaomi 17 Ultra image showed apparent lunar ridges and craters on the Sun during a total eclipse, suggesting the camera’s AI system inferred a familiar object incorrectly.
This is a classic generative-camera failure mode: enhancement that improves ordinary photos can become fabrication when the subject falls outside the training distribution.
Samsung has faced a similar controversy, indicating that the problem is systemic to computational photography rather than unique to Xiaomi.
The supplied report does not include the original file, processing settings or a controlled reproduction, so it establishes an allegation rather than a verified algorithmic diagnosis.

Open at Solidot Ask Claude ChatGPT

Microsoft is shrinking in China but keeping an engineering foothold

# Policy Solidot 1d

Microsoft has closed at least 15 China offices and joint ventures while retaining business serving Chinese firms abroad, revealing how regulation and geopolitics are reshaping a foreign technology footprint.

The company’s cloud and AI expansion is constrained by US export controls and China’s preference for domestic software, while broader economic growth offers less room than before.
Microsoft still sees value in serving Chinese companies that need Western technology to manage overseas operations, creating a narrower but commercially defensible niche.
The company also wants to retain access to China’s engineering talent, so the local presence is becoming more about people and cross-border enterprise services than domestic cloud dominance.
Microsoft says it operates under the same regulatory environment as other international suppliers and remains committed to the market, but the reported closures show that commitment is no longer synonymous with expansion.
The pattern illustrates a broader decoupling outcome: foreign firms may stay, but with smaller, more segmented operations and tighter separation between China-facing products and globally connected services.

Open at Solidot Ask Claude ChatGPT

Alibaba may sell Lingxi Games for more than $1.5 billion

# Culture 36Kr 18h

Alibaba is reportedly close to selling Lingxi Games to a private-equity buyer for more than $1.5 billion, underscoring the retreat from China’s internet conglomerate model.

The reported valuation exceeds earlier estimates of 7 billion to 9 billion yuan, roughly $1.0 billion to $1.3 billion, but neither Alibaba nor the buyer had confirmed the deal.
Lingxi’s cash engine is the long-running strategy game Three Kingdoms Tactics, whose durability makes the unit more like a mature cash-flow asset than a high-growth studio.
The sale would mark a reversal of the 2017 acquisition that brought a NetEase-linked team into Alibaba’s entertainment empire, illustrating how difficult content businesses are to integrate into a traffic-and-commerce conglomerate.
Alibaba’s broader asset reshaping suggests capital and management attention are moving toward cloud and AI, while non-core consumer and entertainment assets face sharper return requirements.
The story also shows the asymmetry of game development: one durable hit can justify a decade of ownership, but failure to create a second hit can turn a strategic division into a divestible asset.

Open at 36Kr Ask Claude ChatGPT

How Sanrio’s China licensing machine helped revive Hello Kitty

# Culture 36Kr 19h

Alibaba’s licensing platform helped Sanrio scale in China, where sales rose 83.2% and Kuromi overtook Hello Kitty, showing the power of localized IP distribution.

Alibaba’s Alifish holds exclusive mainland rights to 26 Sanrio characters and reportedly supported more than 200 brand partnerships and 30,000 product SKUs in 2024.
Sanrio’s China sales reached 31.44 billion yen, up 83.2%, while its broader character portfolio reduced dependence on Hello Kitty, whose group revenue share fell from 60.6% to 37.3%.
The model is relatively asset-light: licensing and channel operations let Sanrio expand rapidly without owning every product or store, while Alibaba supplies local commerce and partner coordination.
Pop Mart provides a useful Chinese comparison because it adds direct retail and blind-box mechanics to character IP, whereas Sanrio still relies more heavily on licensing.
The growth rate is unlikely to persist at the same scale as the China base expands, and Sanrio’s own filing warns that its China and North America strategy may not succeed.

Open at 36Kr Ask Claude ChatGPT

Anhui private industry grew 10.8% in the first half

# Signal Jiemian 30m

Anhui private industrial output rose 10.8%, with electronics up 60.9% and profits nearly doubling, offering a useful view of China’s uneven industrial recovery.

Private firms supplied 55.4% of the province’s above-scale industrial output, making them central rather than peripheral to Anhui’s manufacturing base.
High-tech private industrial output rose 40.6%, while reported profits reached 886.6 billion yuan, roughly $125 billion, up 98.4%.
Private-company imports and exports reached 3.307 trillion yuan, roughly $466 billion, up 39.2%; the number is large enough to show how strongly the province’s growth is tied to trade.
The figures are provincial official statistics, so they describe reported output and profit rather than independently audited productivity or demand.

Open at Jiemian Ask Claude ChatGPT

Venezuela’s Castro tribute tests its balancing act between Washington and Beijing

# Policy SCMP 5h

The episode shows Caracas trying to repair ties with Washington while preserving China’s economic interests, with a symbolic tribute exposing the political limits of that balancing act.

Venezuelan interim President Delcy Rodriguez’s praise of Fidel Castro drew a sharp rebuke from US Secretary of State Marco Rubio.
The China dimension is substantive because Beijing is described as seeking to preserve economic interests in Venezuela while Caracas navigates renewed pressure from Washington.
The story is less about a new Chinese policy than about the narrowing room for governments that want Chinese investment without appearing aligned against the United States.
The supplied report does not identify a specific Chinese project or negotiation affected by the dispute.

Open at SCMP Ask Claude ChatGPT

RuiPu Lanjun pairs battery growth with a sustainability message

# Culture Leiphone 1d

Battery maker RuiPu Lanjun says it returned to profit and is redirecting part of the improvement toward employee incentives and longer-term operating resilience.

The company forecasts first-half revenue of 14.5 billion to 15.2 billion yuan, roughly $2.0 billion to $2.1 billion, and net profit of 700 million to 850 million yuan, roughly $99 million to $120 million.
Its product announcements target data-center storage, heavy trucks, aircraft and embodied systems, showing how Chinese battery makers are widening beyond passenger electric vehicles.
For data centers, the company claims a 6.9 MWh storage system can reduce equipment count and site area relative to a 5 MWh design, but the figures are promotional and not independently tested.
The company says it will allocate about one-third of 2025 profit to employee incentives while cutting formalism and ineffective overtime, a notable response to China’s high-intensity technology labor culture.
The event is primarily brand marketing; it offers no audited evidence that the culture program will improve retention, safety or engineering output.

Open at Leiphone Ask Claude ChatGPT

China’s summer box office passes 9.5 billion yuan

# Culture IT Home 1h

China’s 2026 summer box office passed 9.5 billion yuan, roughly $1.34 billion, after 34 straight days above 100 million yuan.

Reported attendance exceeded 630 million people, showing that theatrical consumption remains a major mass-market habit despite streaming competition.
Domestic titles occupy two of the top three summer positions, while imported films remain part of the release mix rather than being absent from the market.
The total includes advance sales and comes from platform data relayed by state media, so it is a market indicator rather than a full audited revenue statement.
The concentration of releases in August suggests distributors are still using the summer window to manage demand and test the balance between domestic and imported content.

Open at IT Home Ask Claude ChatGPT

ChinAI	403	0/0		HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk	200	1/20
Pekingnology	200	0/20
Ginger River Review	200	1/20
Sinocism	200	0/20
QbitAI	ERR	0/0		ConnectionError: ('Connection aborted.', RemoteDisconnected('Remote end closed connection without response'))
36Kr	200	9/30
Solidot	200	11/20
Leiphone	200	9/20
GeekPark	200	3/30
IT Home	200	12/60	+44 over cap
SCMP	200	10/50	+6 over cap
Initium	200	4/15
Jiemian	200	8/30	+22 over cap
China Digital Times	200	2/15
The Wire China	200	0/50
Techmeme	200	3/15
Hacker News	200	1/20
V2EX	200	8/50	+34 over cap

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

A Hefei DRAM maker passed Tencent

Apple gets a licence, drones get a tariff

GLM-5.3 found a flaw forty years old

Alibaba sells the shortcut, ByteDance refuses one

What the filter blocks, and what it lets through

Same model, two verdicts on the weights

Threads we are pulling