China Desk — 19 Aug 2026, 08:49 UTC

Chinese lab deploys humanoid robots for autonomous experiments

# QbitAI 2h

A national-level Chinese laboratory is using Monte2 robots for long sequences of liquid handling, cell culture and toxicity testing, moving AI for science from computation into physical execution.

Monte2 combines two arms with object-centered physical modeling so it can recognize reagents, vessels and instruments and coordinate actions across equipment rather than repeat one fixed motion.
The reported workflows include nucleic-acid preparation, cell toxicity testing and cell passaging, where small errors in pipetting, force or timing can corrupt an entire experiment.
The system uses vision, force and touch signals and claims submillimeter-level manipulation in changing liquid and soft-material conditions; the report provides no independent accuracy or failure-rate data.
The laboratory plans to expand the fleet to about 100 robots by the end of 2027 and use an AI scheduler to coordinate them.
The architectural direction is a closed loop: a large model plans experiments, embodied systems execute them, instruments produce observations and the next experiment is selected from the results.
If the claims survive reproducible validation, the bottleneck shifts from robot demos to laboratory integration, calibration, safety and the economics of unattended experiment throughput.

Open at QbitAI Ask Claude ChatGPT

China's AI bubble may be industrial policy by another name

# ChinaTalk 22h

Chinese financial language calls the AI funding rush a pass-the-parcel bubble, while this analysis argues that inflated valuations and mass company exits may be deliberate costs of state-led industrial policy.

The article says Chinese investors are applying the phrase to an early-stage market where valuations outrun revenue, money clusters in fashionable subsectors and many firms lack a path to profit.
It cites roughly 80,000 Chinese AI companies dissolving or deregistering over two years, but the excerpt does not define the database or distinguish failed startups from normal corporate cleanup.
The counterargument is historical: Beijing used concentrated state capital and overcapacity to build solar and electric-vehicle industries, so a bubble can coexist with the creation of a globally competitive supply base.
The analysis says national AI funds and hundreds of billions of yuan have accelerated capital formation; those figures and policy details need verification before being treated as a precise spending total.
The key question is allocation quality: can public money fund infrastructure, talent and durable companies, or will it mainly recycle valuations among investors before demand materializes?
For executives, the implication is that Chinese AI competition may remain intense even after a financing correction, because industrial policy can preserve capacity and strategic suppliers after private capital retreats.
The article's conclusion is an argument, not a forecast: the same mechanism that creates waste can also produce scale, lower costs and a few globally important firms.

Open at ChinaTalk Ask Claude ChatGPT

LandSpace lands a Zhuque-3 booster after orbital launch

# Solidot 1h

Chinese coverage stresses the engineering milestone: LandSpace recovered Zhuque-3's first stage on land after reaching orbit, giving China a second demonstrated reusable-launch path alongside maritime net recovery.

Also: SCMP, 36Kr

The booster launched the Honghu 03 satellite and soft-landed six minutes later at a designated site in Gansu, according to the report.
Zhuque-3 is 66 meters tall, 4.5 meters in diameter and uses nine Tianque-12 engines on its first stage and one Tianque-15 on its second stage.
This was China's first orbital-class booster recovery using landing legs on land; the national Long March 10B has used maritime net recovery, so China now has two distinct demonstrated paths.
The report places Zhuque-3 alongside Falcon 9, New Glenn and Long March 10B as orbital launch systems with flight-validated partial reusability.
The strategic test is not the landing alone but whether the stage can be inspected, refurbished and reflown at useful cadence and cost.
Until repeat flights and turnaround data appear, the result is a major engineering demonstration rather than evidence of a mature low-cost launch business.

Open at Solidot Ask Claude ChatGPT

US advisers urge Washington to treat data as an economic asset

# SCMP 9h

A US-China advisory commission says the United States risks losing AI and robotics leadership unless it develops a national data strategy, drawing an explicit lesson from China's data commercialization push.

The report describes China's approach as treating large data collections as economic infrastructure rather than merely as privately held byproducts of online services.
The comparison is strategically important because data policy affects model training, robotics, industrial optimization and the ability to set technical standards, not only privacy compliance.
The US recommendation is not simply to collect more data; it implies national rules for access, stewardship, sharing and commercial incentives.
The account offers no detailed US policy design and does not establish that China's data assets are broadly accessible or high quality, so the comparison should not be mistaken for proof of Chinese advantage.
The story matters to technology executives because data governance can become industrial policy: permissions and interoperability may determine which companies can build durable AI systems.

Open at SCMP Ask Claude ChatGPT

Shanghai shuts a 30-storey ghost-delivery operation

# IT Home 27m

Chinese coverage stresses food-safety enforcement in a Shanghai case where one barbecue shop used forged licenses to run more than 30 delivery storefronts across seven districts.

Investigators say the shop copied or digitally altered real business licenses, then used them to pass platform checks and collect orders under addresses that did not prepare the food.
The actual kitchen was more than 10 miles from the listed Huangpu storefront, creating a direct mismatch between platform identity, delivery geography and food accountability.
The operation generated tens of thousands of yuan per month and more than 1 million yuan, roughly $141,000, overall, according to the report.
The case shows why platform-level merchant verification is a public-health control, not merely an onboarding inconvenience.
Authorities have transferred the suspects to police on alleged forgery of official documents; the case remains under investigation.
The report places the case beside a national crackdown in which seven platforms were fined a combined 3.597 billion yuan, roughly $507 million, showing that ghost kitchens are being treated as a systemic platform problem.

Open at IT Home Ask Claude ChatGPT

Unitree shares surge after its Shanghai debut

# IT Home 49m

Chinese coverage frames Unitree's debut as a national robotics milestone, while market reaction turned the listing into a speculative spectacle as shares jumped 460% at the close.

Also: Techmeme

Unitree became the first humanoid-robot company to list on China's Star Market, according to the report, and its shares closed at 845 yuan.
At that close, the report puts market capitalization at 341.772 billion yuan, roughly $48.1 billion, with Sequoia China holding a reported stake worth about 21.9 billion yuan, roughly $3.1 billion.
The Chinese framing emphasizes a domestic robotics champion and investor celebration; the price action instead raises questions about valuation discipline and the supply of investable robotics stories.
The company's core technical capabilities and future revenue are not established by a first-day move, and the report contains no operating forecast.
The event matters because a public-market price now becomes a reference point for suppliers, venture investors and rival robot makers across China's embodied-AI sector.

Unitree's robot IPO surges in Shanghai debut Techmeme

Open at IT Home Ask Claude ChatGPT

Chinese team releases a cross-body world simulator for robots

# AI QbitAI 22h

Current Robotics introduced CurrentWorld-0, a simulator that predicts action, multi-camera views and force-touch outcomes across different robot bodies instead of merely generating plausible video.

The system keeps each robot's own action subspace while modeling shared physical consequences, so pushing a cup should produce the same object behavior whether the actor is a fixed arm, mobile platform or humanoid.
It generates multiple viewpoints consistently and tracks object state through occlusion, addressing a common failure where one camera shows success while another reveals that the object never moved.
Force and touch prediction are included for tasks such as gripping chips or peeling a cucumber, where visual alignment alone cannot establish contact or load.
The design targets evaluation and training: a failed action must remain failed in simulation rather than being visually corrected into a successful trajectory.
This is a technically meaningful distinction from ordinary video models, but the report supplies demos rather than quantitative rollout error, contact accuracy or transfer results.
If validated, a controllable cross-body simulator could reduce the cost of resetting physical scenes and allow policies to be compared under repeatable intermediate states.

Open at QbitAI Ask Claude ChatGPT

A deleted Beijing security account reveals censorship by collateral damage

# Policy China Digital Times 3h

Chinese coverage documents exhaustive security checks at Beijing's Qianmen gate and the deletion of a matter-of-fact account, showing how politically sensitive locations can make ordinary travel writing subject to broad enforcement.

The post reportedly described identity scans, searches and physical checks at Qianmen during a crowded summer visit, without openly criticizing the authorities or discussing historical taboos.
China Digital Times says the original post was deleted and replaced by a notice citing the Cybersecurity Law, suggesting that location, keywords or automated review may have mattered more than explicit political intent.
The episode illustrates a familiar censorship pattern: content can be removed not only for advocacy but also for describing the friction of visiting a politically charged site.
The account's restrained tone is part of the story; ordinary experience is rendered politically legible by the surrounding security architecture.
The report cannot establish whether human review, keyword scanning or a specific complaint caused removal, so the mechanism remains uncertain.
For companies operating in China, this is a reminder that moderation risk often comes from context and place rather than from a clearly prohibited sentence.

Open at China Digital Times Ask Claude ChatGPT

AI video creation adopts a previs-first workflow

# AI QbitAI 1h

A Chinese AI-video tool uses AI-generated 3D blockouts to set cameras, paths and actor movement before rendering with video models, addressing a real control problem that prompts alone cannot solve.

The proposed pipeline separates spatial and temporal control from appearance: creators build a rough 3D scene, position cameras and characters, record a previs pass, then feed it to video generators.
That architecture attacks information loss between an intended shot and a text prompt, especially for depth, occlusion, actor distance, camera speed and end framing.
The approach is an accessible version of established film previsualization, with AI lowering the cost of scene blocking rather than eliminating the need for directorial judgment.
The report says the tool can generate a simple blockout from one to three reference images in roughly four to seven minutes, but provides no controlled comparison of retries, consistency or final-video quality.
For technical teams, the interesting layer is the interface between structured 3D constraints and generative rendering: the blockout becomes a spatial control signal while texture and lighting remain model-generated.

Open at QbitAI Ask Claude ChatGPT

Dongfeng and Huawei unveil a 300,000-yuan-class family flagship SUV

# AI QbitAI 7h

Chinese coverage presents the Yijing X9 as a 5.3-meter family SUV combining Dongfeng manufacturing with Huawei's ADS 5, HarmonySpace 6 and vehicle-control stack.

Also: Leiphone

The vehicle uses Huawei's WEWA 2.0 architecture, a 896-line dual-optical-path lidar and a 32-sensor suite; the report says the lidar targets stable detection of small obstacles beyond 120 meters.
Huawei's positioning emphasizes defensive driving and risk fields, shifting the sales pitch from merely completing a route to proactively choosing lower-risk behavior.
HarmonySpace 6 adds a native multi-agent cockpit that can coordinate navigation, vehicle control and entertainment, while an interior sensing system watches for children, pets and occupants left behind.
The car extends Huawei's role beyond software and sensors into chassis control, communications and cloud services, increasing the strategic depth of the supplier relationship with Dongfeng.
The 299,800-yuan starting price is roughly $42,200, but the report gives no independent safety data, delivered volume or evidence that the full stack performs as advertised.

Huawei and Dongfeng open presales for the Yijing X9 at $42,000 Leiphone

Open at QbitAI Ask Claude ChatGPT

China bids farewell to former premier Zhu Rongji

# Policy Sinocism 9h

Chinese coverage of Zhu Rongji's funeral shows senior leaders following a tightly managed protocol, while the wider commentary treats his death as the closing of an earlier reform-era chapter.

The roundup says members of the top leadership and Han Zheng paid respects at Babaoshan, with Hu Jintao sending a wreath, presenting the event as institutionally orderly and politically scripted.
The source's interpretation is that Zhu's death marks the end of an era whose political and economic conditions had already disappeared years ago.
Funeral protocol for senior leaders is itself a governance signal: the Party can honor a consequential predecessor while controlling the public frame around nostalgia, reform and succession.
The excerpt also ranges across foreign affairs and economic policy, but those sections are not needed to understand the funeral's political meaning and are excluded here.
The report provides no public debate over Zhu's legacy, so it is clearer about official ceremony than about how ordinary citizens are responding.

Open at Sinocism Ask Claude ChatGPT

Hong Kong's heat is becoming a public-health crisis

# Culture Initium 9h

Chinese-language Hong Kong coverage reports record heat, deaths and growing risks for outdoor workers, arguing that extreme-heat policy has lagged behind the city's changing climate.

Hong Kong recorded 36.9 degrees Celsius on August 9 and a 30.2-degree overnight low on August 12, according to the report, both records for the relevant historical series.
The article describes heat as a larger killer than typhoons and heavy rain, especially for older people and outdoor workers, while noting that humidity prevents the body from cooling efficiently.
The social burden is uneven: construction workers, street vendors and hikers face direct exposure, while dense housing and limited ventilation can keep indoor temperatures dangerous after sunset.
The report combines interviews with emergency physicians, disaster-preparedness officials and an architecture professor, framing heat response as a labor, housing and urban-design issue rather than a personal-responsibility problem.
It asks whether warnings, work protections, cooling infrastructure and neighborhood support are adequate, but the excerpt does not provide a full assessment of government measures.
The broader signal is that climate adaptation in a dense Chinese city will require institutional changes, not only more fans, umbrellas and individual precautions.

Open at Initium Ask Claude ChatGPT

DeepSeek Harness evolves from plugin kit into agent marketplace

# AI GeekPark 6h

Chinese coverage says DeepSeek Harness grew from a rough, developer-first release into a community ecosystem with 149,000 GitHub stars and more than 5,100 plugins within five days.

The architecture reportedly makes the model adapter, tools, filesystem, sandbox, scheduler and even the agent loop replaceable plugins, with no privileged core that users cannot patch.
That fragmentation explains both the initial rough edges and the rapid community response: desktop shells, image support, IDE interfaces and migrated skills appeared within days.
The system's hard boundary is audit logging: anything sent to the model must be reconstructable, which gives a potentially useful foundation for debugging and governance.
The community has already contributed more than 5,100 plugins from over 3,500 authors, according to the report, but GitHub stars are adoption interest rather than production reliability.
The broader signal is architectural: an open agent runtime can outsource product completion to its users, but it also inherits supply-chain, permission and compatibility risks from an uncontrolled extension ecosystem.
For enterprise use, the key question is whether plugin reversibility and complete logs can become enforceable security properties rather than attractive design claims.

Open at GeekPark Ask Claude ChatGPT

China's space-computing startups confront a four-week upload problem

# Signal GeekPark 4h

A Chinese report says more than 30 domestic space-computing ventures are chasing orbital AI, but uploading an 800MB model to a satellite can take nearly four weeks because low-Earth-orbit uplinks remain slow.

The report distinguishes processing satellite-generated data in orbit from sending ground-generated data upward for computation; the latter requires a business case that does not yet have dependable infrastructure.
At roughly 1 Mbps and with short ground-station windows, a model must be fragmented across many passes, making experiments painfully slow relative to a satellite's roughly three-year design life.
The asymmetric link problem matters: laser downlinks benefit from large ground receivers and adaptive optics, while high-rate uplinks remain technically and operationally immature.
The Chinese startup Space Byte is reportedly pursuing ground-to-space storage first, using it as a stepping stone toward computing and as a way to create demand for better uplinks.
The report says a domestic company has raised 140 million yuan, roughly $20 million, and plans to place two nodes in orbit, but it does not establish that the economics beat terrestrial data centers.
This is a useful reality check against space-computing narratives: launch capacity and orbital power are not enough if data cannot reach the processors at operational speed.

Open at GeekPark Ask Claude ChatGPT

Nvidia positions Cosmos 3 as a world model for robots

# AI Leiphone 2h

A Chinese analysis of Nvidia's Cosmos 3 highlights a unified text, video, audio and action model aimed at moving physical-AI intelligence from data centers to edge devices.

The reported architecture unifies multiple modalities in one Transformer-based model and treats world understanding, future simulation and action as a single control stack.
Audio is included because collision and friction cues can reveal physical state that video alone misses; first-person data is presented as especially valuable for learning interaction.
The model also uses inverse dynamics to infer actions from unlabeled video, which could turn large stores of passive footage into training data for robot policies.
Nvidia's strategic position is broader than GPU supply: a common world-model and edge runtime could become an application layer that standardizes how diverse robot bodies consume perception and control.
The report is based on a conference presentation and Nvidia's own claims, with no independent deployment results, latency figures or robot failure rates.
The critical test is whether a unified model can preserve physical causality and run within the power and latency limits of real robots rather than merely generate plausible video.

Open at Leiphone Ask Claude ChatGPT

A Chinese edge SoC pairs an NPU with a CUDA-compatible GPGPU

# AI Leiphone 1d

Chengheng Micro's CH3715 combines a 48-TOPS NPU with a 1-TFLOPS FP32 GPGPU, CPU, DSP, FFT and image processing to replace multi-chip edge-vision systems.

The architectural choice reflects heterogeneous workloads: the NPU handles neural inference, the GPGPU supplies programmable high-precision parallel compute, DSP and FFT handle signal processing, and the CPU schedules the system.
The four-core GPGPU is reported as CUDA-compatible, which could reduce migration costs for customers with existing parallel algorithms while the NPU targets INT8 and FP16 inference.
The SoC also integrates dual ISPs, a rendering GPU, high-speed interfaces and real-time control cores, aiming to reduce board area, data movement and cross-chip latency in robots and industrial vision.
The company claims more than 50% higher system throughput, 30% lower power and 40% smaller equipment, but those figures are vendor claims and lack workload details.
The design lesson is that edge AI is not a pure TOPS contest: perception, geometry, signal processing and control often need different numerical precision and software models.
Chengheng says its next chip will remove less-used blocks, improve memory bandwidth and target roughly half the cost, suggesting that customer deployments will determine whether the broad first-generation integration was justified.

Open at Leiphone Ask Claude ChatGPT

China's AI race shifts from bigger models to embedded systems

# Analysis Leiphone 4h

Chinese AI coverage argues that the next contest is putting models inside complex products, with Alibaba, ByteDance, Baidu, Tencent and Huawei pursuing different agent architectures.

Alibaba's reported route decomposes agent capability into computer use, process rewards, tool calling and reusable skill libraries, making the model less of a black box.
ByteDance is focusing on generating GUI-agent trajectories through exploration, intent reasoning, reflection and error recovery, treating interaction data as the scarce resource.
Baidu, Tencent and Huawei are described as embedding models into maps, recommendations, knowledge graphs and infrastructure rather than replacing those systems with a chatbot.
The strategic distinction is between building what an agent knows and building how it learns from environments; both point toward trajectory data and reinforcement learning as constraints beyond parameter count.
The article uses upcoming conference work and company research as evidence, so it maps direction more reliably than it proves product performance.
For a technical executive, the implication is architectural: durable advantage may sit in tools, state, evaluation and domain data around the model, not in the model endpoint alone.

Open at Leiphone Ask Claude ChatGPT

Momenta, XHEART and QNX build a safety-certified driving stack

# AI Leiphone 5h

Chinese autonomous-driving company Momenta, chipmaker XHEART and BlackBerry's QNX combined algorithms, an X7 automotive SoC and QNX SDP 8.0 into an ASIL D-certified production platform.

The stack separates the major production layers: Momenta supplies full-stack driving algorithms, XHEART supplies the automotive SoC and QNX supplies the safety-oriented operating system.
The partners say the platform passed ISO 26262 ASIL D certification, the highest functional-safety level, and is intended to support international vehicles including those subject to European UN R171 requirements.
The technical significance is less a new autonomy model than the productization of interfaces among neural perception, vehicle compute and a certifiable operating environment.
Certification provides a compliance foundation, not proof of autonomous-driving safety in traffic; the report gives no miles, disengagements or production customer.
The collaboration also shows Chinese autonomy suppliers pairing with foreign embedded-software infrastructure when global homologation matters.

Open at Leiphone Ask Claude ChatGPT

China studies autonomous robotic fortresses for disputed islets

# Policy SCMP 6h

A Chinese coastguard proposal envisions coordinated robot and drone swarms defending South China Sea outposts, replacing some human presence with autonomous systems.

The proposal appears in a Chinese-language journal supervised by state shipbuilder CSIC, so it is a policy and military-technology signal rather than an announced deployment.
The concept responds to the asymmetric threat posed by cheap drone swarms with a network of sensors, unmanned vehicles and automated defenses operating from contested islands.
Robotic outposts could reduce personnel exposure and sustain surveillance, but they also create difficult questions about command authority, communications resilience and escalation when autonomy fails.
The report does not specify hardware, budgets, deployment sites or legal authorization, so the proposal should not be read as an operational plan.
Its importance lies in how Chinese security institutions are imagining maritime control: persistent machine presence may complement or replace visible troop deployments in gray-zone competition.

Open at SCMP Ask Claude ChatGPT

Gree opens a company-linked vocational school for teenagers

# Culture Leiphone 8h

Chinese coverage highlights Gree's new school, led by 72-year-old chairwoman Dong Mingzhu, which promises direct links from secondary vocational training to jobs at Gree and its suppliers.

The first cohort will take 300 students aged 15 to 18, with annual tuition of 9,000 to 15,000 yuan, roughly $1,270 to $2,110.
The model promises that students enter an employer-linked pipeline and graduate ready for specific roles, reviving debate over whether Chinese vocational education should be tightly coupled to industrial employers.
Online reactions in the roundup range from praise for practical employment to comparisons with an older era of assigned work, revealing mixed feelings about security versus personal choice.
The school is a corporate workforce strategy as much as an education project: Gree gets earlier access to trained labor and can shape curricula around factory and supply-chain needs.
The report does not establish guaranteed employment, accreditation quality, wage outcomes or how students can exit the Gree pipeline.

Open at Leiphone Ask Claude ChatGPT

A Chinese war film turns the Iraq invasion into soft power

# Culture SCMP 7h

A hit Chinese film about the Iraq war follows a debt-bound Chinese chef during the US invasion, using an overseas story to present Beijing's worldview to foreign audiences.

The film avoids naming the country directly but clearly depicts a 2000s American invasion, making the geopolitical message legible without a literal political slogan.
Its Chinese protagonist is an ordinary worker caught in war, a narrative choice that turns national critique into a human story about vulnerability, debt and survival.
SCMP frames the film as part of a broader soft-power effort to communicate Beijing's interpretation of US intervention to international audiences.
The report does not provide box-office numbers, foreign distribution or audience-response data, so its actual reach remains unclear.
The cultural signal is important because contemporary Chinese international messaging increasingly travels through commercially legible genres rather than only through official documentaries.

Open at SCMP Ask Claude ChatGPT

Nokia plans a major retreat from China

# Analysis Solidot 1h

Nokia reportedly plans to cut most China employees and close factories by year-end, retaining mainly after-sales service across a market where it still had about 7,200 regional employees at the end of 2025.

The report says Nokia's China footprint once covered research, marketing, services, delivery and supply, including a Beijing research center founded in 1995.
The planned retreat would move the company from local industrial participation toward a much narrower support presence, though the affected headcount has not been disclosed.
The story is strategically relevant to foreign technology suppliers: China operations can remain broad for decades and then contract rapidly as market access, competition and global supply chains change.
The report is based partly on a former employee and does not include a Nokia statement confirming the timetable or compensation terms.

Open at Solidot Ask Claude ChatGPT

Baidu says AI now contributes more than half of core revenue

# Signal Leiphone 19h

Baidu reported 31.3 billion yuan, roughly $4.4 billion, in quarterly revenue, with AI making up half of general-business revenue for a second consecutive quarter.

The report says general-business revenue was 25.2 billion yuan, roughly $3.5 billion, and that AI represented 50% of it, a useful measure of mix rather than total-company revenue.
The result suggests Baidu's AI monetization is moving from a product narrative into a material portion of the operating business, though the excerpt does not define which AI products are included.
Two US investment firms reportedly opened or increased Baidu positions, providing market validation but not evidence of operating quality.
The item gives no segment margins, recurring revenue, cloud contribution or cash-flow detail, so it cannot show whether AI is profitable or merely growing.

Open at Leiphone Ask Claude ChatGPT

A physics discussion explores materials that could compute

# Analysis QbitAI 2h

A Chinese science program uses foam, sand and jamming transitions to explain research into low-power soft materials with memory and learning-like behavior.

The discussion describes jamming as a continuous transition in which disordered particles acquire rigidity when packing exceeds a critical volume fraction of about 64% for random spheres.
The technical idea is to use emergent collective behavior in soft matter as a tunable physical substrate, potentially computing through material states rather than conventional transistor switching.
The report links this direction to nonequilibrium statistical physics, scaling laws and replica-symmetry-breaking theory, but does not identify a working device or benchmark.
The value is conceptual and educational: it shows how Chinese science media is connecting frontier physics to post-silicon, low-energy computation without claiming that such materials are ready for deployment.

Open at QbitAI Ask Claude ChatGPT

Unitree investor warns most humanoid startups will fail

# Analysis 36Kr 5h

A veteran Chinese hardware investor says the embodied-robotics market may yield three clear winners, a handful of survivors and many failures, urging founders to prove technology, engineering and business in that order.

The investor's forecast is opinion, not a market fact, but it captures a recurring concern in Chinese coverage: capital and valuations are arriving faster than supply chains and customer demand can mature.
He says Unitree's early advantage came from electric actuation, obsessive engineering and unusually tight cost control, while warning that financing alone cannot compress hardware iteration cycles.
The interview frames the next moat as an ecosystem of suppliers, industrial partners and capital rather than a robot body alone.
His account is especially relevant after Unitree's listing because a public valuation can amplify pressure to grow before reliability, service economics and recurring orders are proven.
The report does not provide Unitree's current revenue, margins or order backlog, so the investor's three-winner scenario should be read as a disciplined thesis, not guidance.

Open at 36Kr Ask Claude ChatGPT

South China Sea destroyer breakdown prompts questions about US readiness

# Analysis SCMP 49m

Chinese coverage uses a disabled US destroyer to ask whether prolonged operations are consuming America's reserve capacity, while the cited US analyst stresses an unusual chain of equipment and process failures.

The report says a US guided-missile destroyer lost power and drifted in the South China Sea last month, but the excerpt gives no investigation findings or confirmed cause.
Former US Navy captain Brad Martin describes the incident as highly unusual and says multiple equipment and process failures would likely be required.
The China-facing value is the strategic framing: a mechanical failure becomes evidence to examine force-generation depth, maintenance and operational tempo rather than a standalone mishap.
One incident cannot establish a readiness trend, and the report does not provide fleet-wide failure, deployment or maintenance data.

Open at SCMP Ask Claude ChatGPT

Douyin dismantles porn funnel using fake single-parent personas

# Policy IT Home 4m

Douyin says it removed 209,000 violations and helped police arrest 29 people in three criminal groups using AI content, coded language and fake dating personas to funnel users to overseas porn sites.

The platform says it acted against 209,000 pieces of content and 78,000 accounts, including 2,384 accounts with more than 10,000 followers; these figures are Douyin's own enforcement claims.
The cases show a shift from obvious sexual content toward social-engineering personas, euphemistic symbols and traffic migration to external services.
Douyin says it is upgrading large-model detection and expanding its dictionary of coded sexual language, while reporting suspected criminal evidence to authorities.
The AI-generated cases are notable because generative tools are being used to mass-produce borderline content, forcing the platform to police both semantics and production scale.
Chinese platform governance increasingly combines automated moderation, account sanctions and police referral rather than treating content enforcement as a purely private rules matter.

Open at IT Home Ask Claude ChatGPT

Mongolian official apologises after dispute at Chinese oil camp

# Policy SCMP 2h

A Mongolian official apologised after China's embassy complained that protesters harassed workers at a Chinese-invested oil exploration site, exposing the political sensitivity of overseas Chinese projects.

The protest occurred at an oil exploration area operated by Mongolia Foison Energy, where Chinese workers and investment were involved, according to Mongolian media cited by the report.
Chinese diplomatic pressure converted a local protest into a bilateral issue, with the apology signaling that Ulaanbaatar was managing the relationship as well as the immediate dispute.
The report does not establish whether the exploration project violated local law, what protesters demanded or whether operations resumed.
The episode illustrates the recurring tension around Chinese investment abroad: host-country resource politics can quickly acquire an embassy and sovereignty dimension.

Open at SCMP Ask Claude ChatGPT

Thirty US universities ordered to audit China research ties

# Policy SCMP 11h

The US Defense Department ordered 30 universities, including Harvard and MIT, to audit foreign research relationships; universities say officials implied wrongdoing without publishing evidence.

The action makes China-linked research collaboration a compliance and funding issue for major US universities, even though the excerpt does not name the specific partnerships under review.
The dispute is strategically relevant to Chinese science because it can narrow the channels through which researchers, laboratories and companies exchange talent and knowledge.
The report does not establish that any university violated a rule or that the audited relationships involved sensitive technology.
The episode illustrates how national-security review can create uncertainty before facts or allegations are publicly specified, affecting institutional risk decisions beyond the immediate audits.

Open at SCMP Ask Claude ChatGPT

US Democrats challenge easing of China and Hong Kong sanctions

# Policy SCMP 10h

US Democratic lawmakers objected to the administration's recent easing of sanctions on Chinese and Hong Kong officials, arguing it weakens pressure over Hong Kong's autonomy.

The four-page letter says the July decision raises questions about the durability of US Hong Kong policy, according to the report.
The story is about US policy toward China, not Chinese domestic action, so its relevance is the potential change in bilateral enforcement rather than a new Beijing measure.
The excerpt does not identify the officials, sanctions or administration rationale in enough detail to assess the practical effect of the easing.
The dispute signals that China policy remains vulnerable to domestic US political swings even when sanctions are presented as part of a longer-term strategy.

Open at SCMP Ask Claude ChatGPT

Chinese chip-equipment maker ships new dicing tool to memory producer

# Signal IT Home 21m

Huahai Qingke shipped the first Versatile-DT300D dual-worktable wafer dicing system to a leading Chinese memory maker, a small but concrete advanced-packaging supply-chain milestone.

The tool combines dual automatic worktables, high-cleanliness washing, metrology and defect inspection rather than serving only as a cutting machine.
Its software is designed to feed traceable process data into a factory manufacturing platform, linking dicing to yield management and production analytics.
The company positions the system for stacked packaging, chiplets, memory and image sensors, where thinning, edge quality and particle control are increasingly important.
The shipment is evidence of customer adoption, not proof that the equipment matches the throughput, precision or installed base of leading foreign tools.
Huahai Qingke is controlled by Sichuan provincial state assets, illustrating how China's semiconductor-equipment push mixes private engineering with state-linked ownership.

Open at IT Home Ask Claude ChatGPT

China's tech roundup spotlights Gree's school, Cambricon pay and Unitree

# Analysis 36Kr 8h

Chinese roundup coverage links a company-run vocational school, a huge Cambricon employee stock award and Unitree's IPO, showing how labor, capital and industrial policy are converging around technology.

The roundup says 124 Cambricon employees are due shares worth about 690 million yuan, roughly $97 million, at the cited market price, or about 5.57 million yuan, roughly $785,000, per employee on average.
Its Gree item presents a direct school-to-employer pipeline, while the Unitree item treats a robotics listing as an industrial investment catalyst; together they show different ways firms are building talent and capital ecosystems.
The roundup also mentions AI agents operating Windows desktops and enterprise chat interfaces, but those are product announcements rather than independently measured breakthroughs.
Because this is a compilation, the figures and claims belong to the cited underlying reports and should not be read as one coordinated policy announcement.
The useful signal is cultural and economic: Chinese technology coverage is tying employee wealth, vocational training and robot-company financing into a single industrial-development narrative.

Open at 36Kr Ask Claude ChatGPT

Chinese property manager loses money after a metal-trading bet

# Signal Jiemian 26m

Qifu Life Services expects a first-half loss after an 83.4 million yuan, roughly $11.7 million, fair-value loss on silver and gold investments reversed an earlier windfall.

The property manager says its estimated first-half loss is 2.4 million to 4.4 million yuan, roughly $0.3 million to $0.6 million, versus a 75.1 million-yuan profit, roughly $10.6 million, a year earlier.
The loss is unrealized: the company still held the metals at the reporting date and attributes the change to lower market prices.
In January, the company had sold about 680,000 ounces, or 21.15 tonnes, of silver for roughly 247 million yuan, about $35 million, highlighting how far its treasury activity had moved beyond property services.
The case exposes a governance risk in Chinese listed companies: a non-core speculative gain can become large enough to dominate reported earnings and investor attention.
Nothing in the announcement establishes that the company has abandoned metals investing, so the risk may persist beyond this accounting reversal.

Open at Jiemian Ask Claude ChatGPT

Beijing reportedly approves small H200 shipments to ByteDance and Tencent

# Policy Techmeme 5h

US coverage focuses on export-control pragmatism: sources say Beijing allowed ByteDance and Tencent to receive about 10,000 Nvidia H200 chips each at mainland facilities.

The item is a Techmeme aggregation of Financial Times reporting, so the shipment figures and authorization are attributed to unnamed sources rather than independently established here.
If accurate, the policy would represent selective access to advanced foreign accelerators for leading Chinese firms rather than a broad reopening of the mainland market.
The apparent goal is to give national champions enough hardware to narrow the gap with US rivals while retaining leverage over supply and deployment.
The strategic uncertainty is whether these are one-off permissions, a controlled quota system or the beginning of a wider licensing channel; the excerpt does not answer that.
For infrastructure planners, small batches would not erase China's access constraints, but they could materially affect benchmarking, distillation and engineering learning at the largest labs.

Open at Techmeme 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	0/20
Sinocism	200	1/20
QbitAI	200	10/10
36Kr	200	6/30
Solidot	200	15/20
Leiphone	200	15/20
GeekPark	200	2/30
IT Home	200	12/60	+46 over cap
SCMP	200	10/50	+11 over cap
Initium	200	3/15
Jiemian	200	8/30	+9 over cap
China Digital Times	200	1/15
The Wire China	200	0/50
Techmeme	200	3/15
Hacker News	200	1/20
V2EX	200	8/50	+31 over cap

Inner Mongolia is turning inference into an export commodity, and that makes the AI bubble a policy choice

Six departments make tokens an export good

Half-mast for Zhu, loyalty essays from the generals

Trump trims the drills, Wang Yi books Seoul

DeepSeek is now billing for memory, not maths

Honor bet 9,999 yuan on a camera that moves

Threads we are pulling