China Desk — 16 Oct 2026, 08:07 UTC

Shanghai AI Lab model helps prove a long-standing dynamical-systems conjecture

# Leiphone 1d

Chinese coverage describes how the lab's Shusheng-S2 model supported a human-led proof of the three-dimensional Ghys conjecture, later checked using formal verification.

The researchers say the model produced a 400-page initial derivation, then supported seven rounds of follow-on reasoning as the problem graph expanded from 15 to 389 nodes.
The workflow paired long-horizon recursive reasoning with mathematicians who selected promising branches, rejected dead ends and redirected later exploration; the model did not independently deliver a finished proof.
The team says its training combined broad scientific pretraining with research-level mathematics data and verifiable-reward reinforcement learning, while a Harness system managed the extended reasoning process.
A separate team used the SHEAF framework to formalize and verify the proof, an important distinction between plausible mathematical prose and machine-checkable correctness.
The paper and formalization are reported as publicly available, allowing specialists to inspect both the argument and the verification artifacts.
If the result withstands peer scrutiny, it would be a notable demonstration of AI as a research collaborator; the broader claim that this is China's first world-class theorem solved with a domestic model is the outlet's framing.

Open at Leiphone Ask Claude ChatGPT

155 Chinese apps refuse Doubao's screen-control agent

# GeekPark 5h

Tencent, Alibaba, Meituan, Kuaishou and other platforms have blocked Doubao's system-level GUI automation, making app boundaries and control of user traffic an explicit dispute.

Doubao's screen automation can read on-screen content and simulate taps or text entry, letting an agent act across apps without each service building a direct API integration.
The refusal list is dominated by platforms with their own social, commerce, payment or content ecosystems; the article argues they fear losing control of user relationships and transaction flows.
ByteDance's proposed declaration protocol gives third-party developers a formal way to allow or reject automation, replacing some of the previous implicit friction through anti-bot controls.
This is a governance conflict as much as a technical one: an agent's ability to act through an interface challenges the app owner's authority over access, attribution and commercial rules.
The article reports that some refusals include other AI services, suggesting platforms are guarding data and distribution even against direct competitors in the agent market.
The key open question is whether operating-system vendors, app platforms or regulators will define a durable permission standard, and whether users can meaningfully override platform refusals.

Open at GeekPark Ask Claude ChatGPT

Xi renews anti-corruption warning ahead of a major Party meeting

# SCMP 4h

Chinese state media published fresh excerpts of Xi Jinping's speech calling for corrupt officials to be rooted out, while Chinese coverage presents the message as a warning to Party cadres ahead of a major conclave.

The speech was delivered in January to the Central Commission for Discipline Inspection, but its publication now gives the warning renewed political timing ahead of the coming Party meeting.
Xi's language about officials whose public loyalty differs from their actions frames corruption as both personal misconduct and disobedience to central policy.
The episode reinforces the Party's preference for discipline campaigns that police political reliability alongside financial corruption.
The excerpt is an official text released through the Party's flagship magazine, not independent reporting on a specific new investigation or purge.
The practical signal is continuity: cadres are being told that implementation and loyalty will be judged together, even when no new enforcement case is named.

Open at SCMP Ask Claude ChatGPT

China proposes export controls on AI, biotech and solar technologies

# IT Home 8m

Beijing's proposed export-catalog revision adds 40 technologies, including 23 AI areas; Chinese coverage foregrounds the state-control implications, while US coverage is absent from this batch.

The draft spans large models, agents, compute clusters and information processing, as well as cell and gene therapies, nucleic-acid drugs, polysilicon, wafers and solar-cell manufacturing.
The proposal excludes technologies already in the public domain through open-source releases or academic publication, according to the report; some licensing arrangements would receive streamlined treatment.
This is a consultation draft, not a finalized export ban. Its eventual reach depends on how authorities define covered technologies and how the revised catalog handles existing commercial and research activity.
The move fits China's broader use of technology-transfer rules as an industrial and security tool. Adding AI infrastructure and model-related capabilities could give authorities leverage over cross-border transfers of expertise as well as products.

Open at IT Home Ask Claude ChatGPT

StepFun open-sources a 600-billion-parameter model with a million-token context

# AI Zhidx 18h

Chinese lab StepFun released the weights for Step 5 Preview, a sparse MoE model with 27 billion active parameters and a reported top-three standing among open models.

The article describes a 92-layer, relatively narrow architecture intended to preserve depth for implicit reasoning while keeping active compute below that of much larger models.
Its attention system mixes gated sliding-window layers with sparse full-attention layers; the latter select relevant blocks from long histories rather than attending to every token.
StepFun reports that its sparse GQA method reduces selection overhead by grouping tokens and merging nearby selections, while retaining nearly the same information coverage as a denser reference.
The model is positioned for long-running agent tasks, where tool outputs create large contexts and explicit chain-of-thought tokens may be costly.
The release is a meaningful open-model development, but reported benchmark rankings and cost comparisons remain vendor- and evaluator-dependent.

Open at Zhidx Ask Claude ChatGPT

Huawei details chip-stacking and imaging work behind its Kirin roadmap

# AI Zhidx 19h

Huawei's technical briefing described camera, display and semiconductor work, including a claimed 1.5-micron hybrid-bonding pitch and a roadmap for future Kirin chips.

The chip discussion centers on dense hybrid bonding, with Huawei describing more than 50 million connections at a 1.5-micron pitch; the supplied report does not independently verify yields or production readiness.
Huawei also discussed running a 30-billion-parameter model on-device through memory and compute coordination, a relevant test of its phone platform under constrained power and bandwidth.
The imaging section describes a multispectral sensor intended to improve color reproduction and a modular 10x camera system, reflecting Huawei's strategy of differentiating through hardware-software integration.
A detailed public technical briefing after a six-year gap signals confidence in domestic engineering despite continuing semiconductor constraints.
The article is a long company-oriented account; several performance claims and comparisons are Huawei's own, not independent measurements.

Open at Zhidx Ask Claude ChatGPT

Ant Group's Ling-3.1-Flash targets long-running office-agent work

# AI QbitAI 1d

Ant Group says its open Ling-3.1-Flash combines sparse experts with linear attention to handle long contexts and multi-step tasks at lower deployment cost.

The model reportedly has about 560 billion total parameters, 25 billion active per token and a context window of up to one million tokens.
Its attention stack uses seven KDA linear-attention layers for every Gated MLA layer, retaining periodic global attention while reducing sequence-processing cost.
The company says mixed-precision quantization cuts model-weight memory from roughly 534 GB to 292 GB and lowers deployment needs from eight Hopper GPUs to four in its test setup.
The model is aimed at office agents that must gather files, use tools, produce spreadsheets and presentations, and verify results rather than merely answer prompts.
Ant Group cites internal long-running coding and office evaluations, but the supplied material is promotional and does not provide independent reproduction.

Open at QbitAI Ask Claude ChatGPT

Apple Intelligence gets China approval, with Qwen and a limited first release

# Analysis Zhidx 20h

Chinese coverage emphasizes Alibaba's Qwen partnership and the initial feature set, while the US roundup stresses approval and the late-October test; both say the new Siri AI is excluded.

Also: QbitAI, Leiphone, Techmeme

Tim Cook said Apple Intelligence and the related Qwen model had been approved and that a test version would arrive later in October.
The first features are reported to include live translation, writing tools and photo cleanup, while the more ambitious AI version of Siri is not included.
Apple's China launch depends on domestic model partners and regulatory approval, showing how local compliance shapes the product rather than simply delaying a global feature release.
The report says a later full release is expected with a future iOS update, but the timing and final feature set remain uncertain.

Open at Zhidx Ask Claude ChatGPT

Meta's 816-billion-parameter open model trails several Chinese rivals on coding tests

# AI Zhidx 7h

Chinese coverage measures Meta's Muse Spark Open against local open-weight models and highlights its weaker coding scores, while noting that it can run locally across two high-memory Mac systems.

Also: GeekPark

Muse Spark Open is a sparse mixture-of-experts model with 816 billion total parameters and about 42 billion active per token, with a claimed million-token context window.
The report cites third-party results placing it behind several Chinese models on software engineering and terminal-agent benchmarks, though it says Meta's model leads on some agent and multimodal evaluations.
Two Mac Studio systems can reportedly host the full model through Apple's unified memory and MLX, an unusual desktop deployment that favors memory capacity over conventional GPU-server scale.
Chinese coverage turns the release into a comparative account of domestic open-model competitiveness rather than simply repeating Meta's claim that it is the strongest US open-weight model.
Benchmark rankings are snapshots and depend on test selection; the article does not establish performance across production workloads.

Apple and Meta team up to run a huge open model on Mac hardware GeekPark

Open at Zhidx Ask Claude ChatGPT

China's envoy urges Manila to avoid retaliation and reject a forced US-China choice

# Policy Pekingnology 22m

At a Manila dialogue, China's ambassador urged restraint in maritime disputes and said the Philippines need not choose between Beijing and Washington, framing engagement as a way to prevent escalation.

Also: SCMP

Jing Quan described bilateral ties as being at a sensitive point but argued against letting maritime disputes define the entire relationship.
He cited renewed political dialogue, trade, tourism and people-to-people contacts as reasons to preserve channels of engagement.
The statement presents Beijing's preferred frame: Manila can maintain US ties while avoiding a zero-sum alignment against China.
The remarks also included criticism of Philippine actions and alleged anti-China forces, so the conciliatory appeal does not remove the underlying dispute.

China envoy's warning of an easy military victory draws Manila's rebuke SCMP

Open at Pekingnology Ask Claude ChatGPT

Chinese interconnect-chip maker Zhixingwei unveils a broad platform

# AI Zhidx 20h

Zhixingwei says its interconnect chips have shipped more than 800,000 units and span storage, networking, PCIe expansion and AI-cluster connectivity, challenging entrenched overseas suppliers.

The company's product range covers general server interconnects and higher-speed AI infrastructure, including PCIe switching, InfiniBand networking and an announced 800G AI network interface.
Interconnects can limit effective cluster performance even when accelerator peak compute is high, because GPUs wait for data or synchronization across devices.
Zhixingwei claims its PCIe switch can match a Broadcom product and its RAID configuration can reach higher write bandwidth in a specified test; the report provides no third-party validation.
The company's partnerships with local server makers, cloud providers and operating-system vendors point to the integration work needed to replace established components.
For China, the strategic question is whether domestic vendors can deliver reliability, software compatibility and sustained supply at scale, not only competitive silicon.

Open at Zhidx Ask Claude ChatGPT

Houmo outlines two compute-in-memory paths for edge AI

# AI Zhidx 3h

Chinese chip startup Houmo says it will develop both 2D and 3D SRAM compute-in-memory designs, aiming to balance model capacity, bandwidth, power and package constraints.

The company positions 2D CIM for workloads needing capacity and compute density, while its 3D path targets higher bandwidth through advanced packaging.
SRAM-CIM keeps weights in on-chip memory and performs multiply-accumulate operations in the memory array, reducing data movement; Houmo claims lower core area and power than a conventional implementation.
The split reflects a real architectural trade-off: 2D designs face bandwidth limits as models grow, while 3D stacking raises thermal and area constraints.
The report's comparisons are company figures, and it supplies no independent benchmark, production-volume data or customer deployment evidence.

Open at Zhidx Ask Claude ChatGPT

China rights roundup reports labor advocate questioning and tobacco-farmer protests

# Culture Initium 5h

Initium's monthly roundup reports prolonged questioning of a labor-rights observer and protests by Yunnan tobacco farmers over low purchase prices, connecting worker advocacy with rural bargaining power.

The farmer dispute is about the price paid for tobacco leaf, a livelihood issue in which growers have limited leverage against organized purchasing channels.
The headline also flags pressure on a labor-rights advocate, a reminder that independent labor monitoring can itself be politically sensitive in China.
The supplied text is incomplete, so it gives no fuller account of the questioning, protest size, official response or outcome.

Open at Initium Ask Claude ChatGPT

Tencent launches a WeChat mini-program to challenge ByteDance's RedNote dramas

# Culture GeekPark 5h

Tencent is promoting Green Bubble, a free short-drama service inside WeChat, using social sharing and friend activity to compete with ByteDance's fast-growing RedNote app.

RedNote's reported daily active users reached 168 million in July, with average viewing time around 125 minutes, giving ByteDance a major attention and advertising asset.
Tencent's strategy differs from a standalone app: users can watch within WeChat and discover shows through friends, groups and video accounts rather than relying only on algorithmic recommendations.
The launch gives producers another distribution channel after RedNote's ad-funded model concentrated audience and bargaining power in one platform.
The report says Green Bubble is early, with a smaller content library and many shows already released elsewhere; social distribution alone may not generate hit content.
The competitive stakes are consumer time and advertising inventory, especially among older and lower-tier-city users who have become important short-drama audiences.

Open at GeekPark Ask Claude ChatGPT

Alibaba's new AI glasses trade a display for lighter weight and always-on sensing

# AI InfoQ China 9h

Chinese coverage links the launch to a reported 8% domestic smart-glasses shipment decline, while describing Alibaba's response: lighter, screen-free glasses with a low-power sensing path.

The N1 Pro is reported at about 44 grams and uses a 5-milliwatt always-on sensing path that watches for basic scene changes before waking higher-power components.
The design avoids running a full video model continuously; a low-power chip monitors simplified grayscale frames, then escalates to communications, audio or imaging functions when needed.
Alibaba says persistent sensing requires user authorization and does not default to uploading the captured scene, addressing the privacy concern inherent in wearable cameras.
The trade-off is responsiveness: the reviewer reports one- to two-second delays for some voice and cloud interactions, with performance dependent on network access.
Chinese market data cited by the article show the category weakening, suggesting comfort and a convincing daily use case may matter more than adding sensors or translation features.

Open at InfoQ China Ask Claude ChatGPT

Douyin's local-services app cools after subsidies and traffic boosts

# Analysis Leiphone 3h

Douyin's standalone deal app reportedly peaked above 22 million daily users before growth slowed as the company cut subsidies, raised commissions and shifted toward advertising revenue.

The app was seeded with traffic from Douyin and discounts, demonstrating how a platform can quickly manufacture scale by redirecting users from its main feed.
The subsequent pullback exposes a familiar local-commerce trade-off: subsidized downloads are not the same as repeat purchase behavior or merchant economics.
The report says Douyin's service team is raising ad targets and reducing emphasis on gross merchandise volume, a shift from expansion toward monetization.
Reported active-user figures come from QuestMobile and company-side accounts; the article also notes disagreements over how much of the early growth was organic.

Open at Leiphone Ask Claude ChatGPT

Former Huawei AI lead starts a company focused on proactive physical AI

# Analysis QbitAI 1d

Former Huawei assistant-model lead Tang Duyu has founded Guanxing Shuzhi to build systems that infer when people need help and act through physical devices without waiting for a direct command.

The proposal changes the agent problem from executing a stated task to interpreting a continuous stream of events and deciding what to do, when to intervene and which device should act.
That adds a difficult safety layer: the same observed behavior can signal an emergency, a harmless choice or an ambiguous situation where intervention would be intrusive.
The founder's background spans assistant models, code intelligence and large-scale device deployment, but the report gives no product demonstration or company funding details.
Proactive behavior could make connected devices more useful, but it also raises questions of consent, liability and who sets acceptable intervention boundaries.

Open at QbitAI Ask Claude ChatGPT

China's low-speed electric cars split between a domestic crackdown and US demand

# Culture TMTPost 31m

Local bans are squeezing China's informal neighborhood-car industry, while export-focused maker Taotao is reportedly profitable selling low-speed vehicles in the US.

The article describes a market built around real mobility needs in counties and rural areas, especially for older residents poorly served by buses and uncomfortable with two-wheelers.
Beijing's safety and vehicle-approval rules are forcing the sector toward licensed cars; the report cites a sharp output fall after the regulatory shift and the collapse of former leader Reading.
Taotao's contrasting export strategy targets US low-speed-vehicle rules and retirement communities, illustrating how the same manufacturing base can serve a market that permits a category China is restricting.
The social trade-off is clear: cheap local mobility and large employment versus road safety and regulatory compliance. The article's historical market and accident figures should be treated as reported estimates.

Open at TMTPost Ask Claude ChatGPT

Chinese AI infrastructure engineers see sharp wage increases

# Culture Leiphone 1d

Recruiters and executives say pay for strong AI infrastructure engineers has risen sharply as labs compete to improve GPU utilization and reduce inference costs.

The article describes reported salary growth from about 300,000 yuan (roughly $42,000) in 2024 to around 1 million yuan (about $140,000) in 2026 for one candidate; exceptional packages can be much higher.
The work includes kernel optimization, inference scheduling and parallelism strategies such as expert parallelism and prefill-decode separation, where small utilization gains can save large compute budgets.
One source estimates typical GPU utilization at 25-27% and says strong teams can push it toward 40%; these are industry estimates, not an audited survey.
The change reflects a shift in AI competition from model quality alone toward the systems engineering that determines how much useful output each accelerator produces.
The article's salary anecdotes should not be read as market-wide averages, especially where equity makes up a large share of startup compensation.

Open at Leiphone Ask Claude ChatGPT

China buys half of Brazil's oil exports as Hormuz disruption shifts trade

# Signal SCMP 13h

Brazil-China trade reached a reported $143 billion in the first nine months of the year as China absorbed a large share of Brazilian oil exports during disruption to Gulf supply.

Brazilian exports to China rose 12% to $84.7 billion, while imports from China grew 9.9% to $58.6 billion, according to the Brazil-China Business Council.
The figures show how an energy shock can redirect China's commodity sourcing and deepen bilateral trade well beyond the oil sector.
A large increase in Brazilian crude does not by itself show that China can replace Gulf suppliers at equivalent scale or quality.
The source attributes the trade shift to the Iran war disrupting Hormuz; the supplied excerpt does not break out oil volumes or contract duration.

Open at SCMP Ask Claude ChatGPT

Beijing says China will not join nuclear-arms talks with the US and Russia

# Policy Jiemian 23m

China's foreign ministry rejected a proposed three-way denuclearization discussion, arguing that China's arsenal is far smaller and that Washington must first meet its special disarmament responsibilities.

The response reasserts Beijing's longstanding position that arms-control talks should reflect the large asymmetry between China's arsenal and those of the US and Russia.
The wording leaves little room for near-term trilateral negotiations, even as Washington has raised the idea of including Beijing in nuclear discussions.
China's argument emphasizes both fairness and sequencing: the largest arsenals should reduce first rather than setting equal obligations for unequal forces.
The report gives no new figures on China's stockpile or details of any parallel diplomatic channel.

Open at Jiemian Ask Claude ChatGPT

China's humanoid-robot rental prices fall as vendors chase real work

# Culture TMTPost 31m

A Chinese report describes robots moving from holiday demonstrations into retail, tourism and service roles as rental prices reportedly fall to a few hundred yuan per day.

The underlying business shift is from charging for spectacle to competing on measurable labor or sales output; low rental prices make operators test robots in ordinary commercial settings.
The article cites deployments in shops, tourist attractions and service stations, including robots handling product retrieval, delivery and customer interaction.
Its claims about unit economics, deployment volumes and labor substitution are based on selected company or outlet examples, not a broad independent survey.
The decisive constraint is still reliability: a robot must handle varied objects, interruptions and customer behavior cheaply enough to beat a human or a simpler machine.

Open at TMTPost Ask Claude ChatGPT

Kuaishou prepares Kling for independence while retaining control

# Analysis TMTPost 31m

Kuaishou is reorganizing video-generation unit Kling ahead of a reported $3 billion financing, exposing the governance and talent tensions in spinning a major AI asset out of a public company.

The proposed financing could dilute Kuaishou to about 68%, while leaving it the controlling shareholder and consolidating Kling in its accounts, according to the article.
Kling's new CEO is a Kuaishou executive rather than a founding researcher; the report highlights how responsibility for results may diverge from control over capital and resources.
A special voting arrangement reportedly gives the CEO greater voting influence than his economic stake alone would imply, illustrating how pre-IPO structures allocate authority among parent, managers and investors.
The article also points to senior researcher departures as a risk to continuity, but does not establish how much of Kling's model work or roadmap left with them.
The key test is whether the unit gains enough operational autonomy to compete for talent and investment without losing the distribution and infrastructure advantages of its parent.

Open at TMTPost Ask Claude ChatGPT

Beijing says Visa's China approval followed the Xi-Trump summit

# Policy SCMP 10h

China's embassy says Visa received approval to establish a bank-card clearing institution as a direct outcome of Xi Jinping's US visit, presenting market access as a summit deliverable.

A local clearing institution can let Visa process domestic transactions under China's regulatory framework rather than relying only on cross-border networks.
The official statement links a commercial approval to high-level diplomacy, suggesting Beijing wants summit outcomes to be visible and attributable.
The report does not detail the approval's scope, operating conditions or when Visa could begin service.

Open at SCMP Ask Claude ChatGPT

Beijing values summit pageantry even when policy results are modest

# Policy SCMP 6h

SCMP argues that the lavish welcome for Xi Jinping in Washington signaled that sustained engagement remains important despite the strategic rivalry and limited near-term results.

The article treats ceremony as political communication to Beijing, US allies and global markets, not merely diplomatic theater.
The framing suggests China values visible parity and respect as evidence that Washington recognizes its status and strategic weight.
The report cautions that improved optics do not erase structural rivalry or guarantee durable agreements.

Open at SCMP Ask Claude ChatGPT

China orders faster coal-mine restarts ahead of winter heating season

# Policy Jiemian 27m

Central agencies told Shanxi to restart suspended mines and raise output while maintaining safety, as Beijing prepares to secure coal supplies for winter.

The directive calls for mine-by-mine restart plans, faster inspections and daily production monitoring, alongside stronger fulfillment of long-term power-coal contracts.
The state is prioritizing supply security and stable prices over a rapid reduction in coal dependence as the heating season approaches.
The episode shows how central agencies coordinate production in a major coal province when energy reliability becomes an immediate political and economic concern.
The report gives no revised output target or estimate of how many mines are affected.

Open at Jiemian Ask Claude ChatGPT

Sinopec starts China's first million-ton low-concentration CCUS demonstration

# Analysis Jiemian 25m

Sinopec has begun building a large carbon-capture, utilization and storage project at Shengli oilfield, pairing capture equipment with dedicated wind power and storage.

The plan calls for injecting captured carbon dioxide through more than 70 wells into several types of oil reservoirs, with operations targeted for late 2028.
A 98.8-megawatt wind project is intended to supply the capture facility directly, linking renewable generation to energy-intensive carbon capture.
The demonstration matters as an industrial test of whether low-concentration CO2 capture can be integrated with existing oilfield infrastructure at million-ton scale.
The project is under construction, so the reported design is not yet evidence of achieved capture rates, lifecycle emissions reductions or commercial economics.

Open at Jiemian Ask Claude ChatGPT

Douyin sues a JD.com employee over alleged false claims about platform pressure

# Policy Leiphone 23h

Douyin e-commerce has sued a JD procurement manager over social-media allegations that it pressured merchants and promoted negative coverage of JD; a Beijing court has accepted the case.

The dispute grew from a viral screenshot during the 618 shopping festival, amplified by accounts whose posts reportedly appeared unusually similar and concentrated.
The article says the screenshot linked to a media report about platform price practices, not evidence that Douyin manipulated a news ranking as alleged.
The case reflects how China's platform rivalry now plays out through both merchant policies and public narratives, with defamation law used to contest claims made during promotions.
The court has accepted the case, but the allegations have not been adjudicated in the supplied report.

Open at Leiphone Ask Claude ChatGPT

BYD opens its first overseas flash-charging network in Indonesia

# Analysis IT Home 25m

BYD launched its second-generation Blade battery and flash charging in Indonesia, beginning with chargers at 10 Denza dealerships and a plan to expand public access.

BYD claims a 5-minute charge to a usable level and a 9-minute charge to full, though the report supplies no independent test conditions or vehicle-specific details.
The first overseas network is dealership-based, giving BYD a way to pair vehicle launches with infrastructure rather than relying entirely on local third-party charging.
BYD says it has built more than 12,000 domestic flash-charging stations and plans 6,000 overseas; delivery and utilization will determine whether the network supports international sales at scale.
The technical and commercial test is whether fast charging can be deployed economically where grid capacity, land and local service partners differ from China.

Open at IT Home Ask Claude ChatGPT

Oppo's wearable recorder tries to turn conversations into searchable memory

# Culture GeekPark 14h

Oppo is positioning its small clip-on recorder as a phone-integrated memory assistant, with automatic transcription, topic timelines and no retained original audio, according to a hands-on review.

The device reportedly records continuously for up to 24 hours and uses voice identification to separate the wearer from other speakers.
Its software groups conversations across settings into project timelines, trying to make recordings useful for follow-up rather than leaving isolated audio files.
Oppo says it retains text rather than original audio and encrypts data through its private cloud, but the review does not explain consent practices for other people being recorded.
The product illustrates the privacy trade-off of ambient AI: lowering friction to capture information also makes clear indicators and bystander consent more important.

Open at GeekPark Ask Claude ChatGPT

Chinese robotics firms weigh the data gap as OpenAI enters embodied AI

# Analysis TMTPost 31m

A Chinese robotics founder argues that OpenAI's reported progress in manipulation reflects years of data, compute and engineering investment, raising the bar for domestic competitors.

The article says one Chinese startup, Yuanli Lingji, used 150,000 hours of data and aims to reach one million hours by year-end; those figures are company-provided.
The strategic argument is that scaling robot learning depends on the collection pipeline and compute infrastructure as much as on model architecture.
This is a competitive assessment from a vendor with a stake in embodied AI, not an independent comparison of OpenAI and Chinese systems.
The report does not provide enough detail to verify the claimed multi-million-hour OpenAI dataset or compare data quality and task coverage.

Open at TMTPost Ask Claude ChatGPT

On Running says premium pricing in China depends on more than shoe technology

# Culture 36Kr 2h

In an interview, On Running describes its premium strategy and expansion into football and golf while acknowledging that its China revenue target remains unmet.

The company says it wants Chinese product design and consumer understanding to support growth, while its regional sales are still below a previously stated goal of 10% of global revenue.
On argues its advantage lies in coordinating product innovation, athlete endorsement, owned retail and pricing discipline rather than relying on a single feature that rivals can copy.
The report places the brand between mass-market sportswear and luxury goods, a position that depends on Chinese consumers continuing to pay for premium athletic identity.
The company's lowered global outlook and delayed China target are a reminder that brand narrative does not guarantee demand or sustained full-price sales.

Open at 36Kr Ask Claude ChatGPT

Nike courts Chinese consumers with local products and a channel reset

# Culture TMTPost 31m

Nike's China chief is highlighting local product design and sports communities as the company confronts a prolonged sales decline and stronger specialist-brand competition.

The report says Greater China revenue fell 26% on a constant-currency basis in the latest quarter and has declined for nine consecutive quarters.
Nike is building a China-specific product team and plans to take back online sales from distributors, aiming to control pricing and connect digital and physical retail.
The article frames the CEO's visits to football and fitness events as an attempt to rebuild brand relevance among urban consumers, not simply a publicity tour.
Moving Greater China into a broader Asia-Pacific reporting region could make the business harder to assess from public financial disclosures, even if Nike says local decision-making will increase.

Open at TMTPost Ask Claude ChatGPT

China policy roundup flags a new anti-corruption message before the Fifth Plenum

# Analysis Sinocism 9h

Sinocism's briefing highlights Xi Jinping's warning about officials who publicly profess loyalty but privately resist central directives, alongside upcoming China policy and diplomacy items.

The excerpt frames political discipline as a concern about implementation and obedience, not only personal enrichment.
The briefing's other listed themes include China's relations with the EU and Russia and substitution of foreign wind-power components, but the supplied text gives little detail.
As a newsletter roundup, it is useful as a map of issues under discussion, not as original reporting on each listed development.

Open at Sinocism Ask Claude ChatGPT

ChinAI	403	0/0		HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk	200	0/20
Pekingnology	200	1/20
Ginger River Review	200	0/20
Sinocism	200	1/20
QbitAI	200	9/10
36Kr	200	6/30
Solidot	200	12/20
Leiphone	200	18/20
GeekPark	200	8/30
IT Home	200	12/60	+43 over cap
SCMP	200	10/50	+11 over cap
Initium	200	4/15
Jiemian	200	8/30	+21 over cap
China Digital Times	200	0/15
The Wire China	200	0/50
Techmeme	200	5/15
Hacker News	200	0/20
V2EX	200	8/50	+35 over cap
Zhidx	200	9/20
InfoQ China	200	8/20	+12 over cap
New Wisdom	200	8/15	+2 over cap
TMTPost	200	8/17	+9 over cap

China's biggest apps just decided an AI agent has to be invited in

151 apps told ByteDance's phone agent no

Beijing waves through Apple's easy features

Exports surged and Brussels hardened

Zhipu's cyber bet, and idle GPUs

Harbin's informer, Luckin's algorithm

Threads we are pulling