China Desk

China News, Summarized 12 Oct 2026, 17:45 UTC 34 stories 22/23 sources

Today's brief

Beijing agreed to halve a forecast, which is much cheaper than halving a shipment

A halving measured against a prediction

The number finally arrived, and it is about the future. After two days with commerce minister Wang Wentao in Beijing, EU trade and economic security commissioner Maroš Šefčovič said the understanding on hybrid and plug-in hybrid exports opens the prospect of cutting Chinese shipments by more than half over four years — 700,000 fewer vehicles next year alone and up to €14bn trimmed, per SCMP.

Half of what, though. Šefčovič's own phrasing is half of the pace the EU predicts. Chinese hybrid and PHEV exports to Europe run around 50,000 a month, so four years at today's rate is roughly 2.4 million cars; halving that gives 1.2 million, not "millions". The baseline is a growth curve nobody has published, as Initium's anatomy of the deal lays out.

The curve exists because of a tariff gap. Since October 2024 Chinese battery-electric cars have faced combined duties up to 45.3% in the EU while plug-in hybrids pay the ordinary 10%. Monthly hybrid exports went from about 3,800 in October 2024 to some 50,000 by July 2026. Brussels calls that circumvention; Chinese makers call it product-market fit.

What Beijing banked is nearer-term. Sixteen consensus points, faster rare-earth and magnet licensing — Europe sources about 98% of its permanent magnets from China, and the October 2025 controls are suspended only to 10 November 2026 — advance notification of export controls, and roughly €225m of tariff relief for European exporters. Next round: March 2027.

And Brussels keeps the knife out. The goods deficit was about €360bn in 2025 and already €234bn through July; François Chimits, who runs the Europe programme at Institut Montaigne, expects it to pass €400bn and calls it a second China shock. A December toolbox is drafted: emergency import powers triggerable in 24 hours, three-supplier rules for critical sectors, a solidarity fund. Leaders meet Thursday.

Go deeper on this section: ClaudeChatGPT

The clean room got demolished in six days

REA is the loudest thing on GitHub. Reverse Engineer Anything, an MIT-licensed TypeScript project from a developer going by morluto, went from under 10,000 stars on 6 October to over 110,000 by today. It wires Ghidra, IDA, Hopper and JADX into coding agents over MCP, so one sentence gets you a decompiled answer with evidence, per GeekPark's long read.

Practitioners are unimpressed, and right. Elias Bachaalany, co-author of Practical Reverse Engineering and once at IDA's maker Hex-Rays, answered the hype wave with the word "hype". Duncan Ogilvie, who wrote the IDA Pro MCP plugin and the x64dbg debugger, says he cleared this year's Flare-On malware-analysis contest in about an hour. Virtualising packers still hold; server-side logic is untouchable.

The new thing is who is holding the tool. Reverse engineer Maurice Heumann ran a dozen Claude Code and Codex agents in parallel and rebuilt 99% of a shooter's functions as C++, 83% byte-identical — on Hex-Rays' own MCP, not REA. The skill floor, not the ceiling, moved.

Chinese coverage went straight to liability, and that is the better story. The industry's defence has been clean-room development: one team reads the binary and writes a spec, a second team that never saw the code implements it, and the wall between them is the evidence at trial. REA puts both teams inside one agent. Hacker News calls it IP laundering.

The statute cuts both ways. China's 2020 Copyright Law lists software reverse-engineering research among the permitted circumventions of technical measures — then bars supplying circumvention tools to others, which is precisely what a one-line installer does. In the US, Bowers v. Baystate upheld a licence's no-reverse-engineering clause; in the EU such clauses are void. Rockstar, Activision and Bethesda have already pulled browser ports.

The licensed version of the same idea raised money today. ModCraft, founded by Li Shuo — a Tencent and Youzu veteran who produced Game of Thrones: Winter is Coming — took a seed round in the tens of millions of yuan from SenseTime's Guoxiang Capital and L2F. Natural-language game mods across 28 Steam titles, and the run data sold to ByteDance as world-model training material, via 36Kr. Mods as a sanctioned interface into a compiled game.

Go deeper on this section: ClaudeChatGPT

Two ministers wake up on the sanctions list

Paraguay's and Belize's cabinets got personal. From Monday Beijing froze the Chinese assets of Gustavo Villate, Paraguay's ICT minister, and Francis Fonseca, Belize's foreign and education minister, after both governments signed agreements with Taipei, SCMP reports. Both countries still recognise Taipei; the signal is that individual ministers now carry the cost.

Meanwhile the diplomatic calendar is pointed at Europe's edges. Foreign minister Wang Yi attends the Baku summit of the Conference on Interaction and Confidence-Building Measures in Asia on Thursday as Xi Jinping's special representative, then Croatia, Montenegro and Bulgaria — per SCMP. The same days EU leaders sit down to judge the hybrid understanding.

Go deeper on this section: ClaudeChatGPT

Heterogeneous plumbing beat model-building for funding

Infinigence AI raised 2.3bn yuan (~$324m). The Tsinghua electronic-engineering spinout, founded 2023 by professor Wang Yu with chief executive Xia Lixue, is now at 4.3bn yuan (~$606m) cumulative and reportedly preparing an IPO. It sells neither models nor chips — it sells the conversion of mismatched silicon into tokens, per Zhidx.

The claims are specific enough to argue with. A 70-billion-parameter model trained across Ascend, Iluvatar, MetaX and Moore Threads plus AMD and Nvidia at 97.6% utilisation; two working days to adapt a new chip; cross-domain training between Hami in Xinjiang and Shenzhen, some 2,500 miles apart, up 165%; a disaggregated prefill/decode architecture cutting first-token latency 51.5% and per-token cost 37.5%; daily token calls up 40x from December to July.

THE TELL: the Nvidia-shaped version of this business — Fireworks AI at $17.5bn, Baseten at $13bn, Together AI at $8.3bn — is worth roughly thirty times as much. Neutrality across sixteen chip families is a worse business and a more necessary one.

Huawei published the most interesting engineering of the day. Noah's Ark Lab, with UCL, cleared all 25 public ARC-AGI-3 games and 183 levels using 7,518 actions against 17,135 for humans — a perfect relative-efficiency score, on frozen Claude Opus 5 weights with no fine-tuning, per QbitAI. The same model under the official harness averages about 40.7.

The method is a versioned, executable guess about the world. Memento 3 keeps a rule book and a Python simulator of the game, replays its own actions and compares 64-by-64 frames cell by cell to falsify them, and keeps the whole thing in Git so wrong theories can be rolled back. One rule book went through eleven revisions.

Then it stopped calling the model. On Atari Pong it inferred the bounce geometry, handed control to the learned code, and won 21-0 in three different openings off roughly 9,500 frames. Recursive self-improvement with the weights nailed shut, and the artefact is auditable — which is more than most agent stacks can say.

Go deeper on this section: ClaudeChatGPT

A lawyer named the letters he can no longer read

Cai Yaqi posts other people's lives. He is a criminal-defence lawyer and a well-known instructor for China's national legal qualification exam, with 1.59m followers on Bilibili, and he regularly reads out messages strangers send him about their work. On 8 October he posted a 39-second video, now near 434,000 views, described by Pekingnology.

The content was an instruction. People working in the prison system, judicial administration or administrative reconsideration should stop submitting their accounts, because he cannot read them aloud. He would not say why. Everything else was still welcome — "as long as they don't go too far, at least for now." Scrolling comments compared it to pleading guilty for leniency.

Elsewhere the argument is about requisitions nobody files. McKinsey's August survey has 39% of respondents expecting AI to shrink headcount within a year while only 14% say it already has; last year's cohort predicted 32% and under-delivered. Gartner found around 80% of large firms piloting autonomous tech cut staff, with no link between deeper cuts and better returns, per TMTPost.

The pricing is where the shift is legible. Tencent's WorkBuddy still charges 198 yuan (~$28) per seat per month; Salesforce's Agentforce charges $0.10 per standard agent action and Intercom's Fin $0.99 per qualifying resolution. Software is starting to bill for work rather than for logins — against a budget line that used to be payroll.

One person at the sharp end. Xiao Chaohu quit a communications job at 52, crammed for about 100 days, scored 398 and is now a third-year master's student in tea processing at Fujian Agriculture and Forestry University, where he used AI to write a jasmine-scenting monitoring system with two patents in examination. At 55 he applied to be a brand director and was turned down fast, TMTPost reports.

Go deeper on this section: ClaudeChatGPT

Threads we are pulling

  • The Claude ban thread finally has a denominator. I have been running forum anecdote for a week. Anthropic's transparency report: 11.4m accounts banned in the first half of 2026 against 1.45m in the second half of 2025, 398,000 appeals, about 42,000 overturned — a 10.6% success rate.
  • Microsoft's Qwen base is now a company-wide test. Following yesterday: Satya Nadella says the firm is trialling Microsoft-Decision-1 across incident response, quality control and scientific discovery. Of 129 open-source decision models on HuggingFace, 84 are built on Qwen, per Leiphone.
  • CXMT answers the memory squeeze. Following Sunday's piece on the dying budget handset: president Cao Kanyu says a 4F² architecture with vertical surrounded-gate transistors ships as DDR5 server RDIMM by year end — the domestic line moving toward the contracts that caused the squeeze, via Leiphone's morning brief.
  • The hybrid cap does not touch brand entry. At Paris, Huawei-and-Seres brand AITO said it enters Europe in autumn 2027 and wants a Euro NCAP five-star rating this year, per IT Home; BYD's Denza will push its assisted-driving stack to European Z9 GTs by over-the-air update next year pending UN R171 approval.
  • The brake story is now a marketing line. Zhao Yongpo, who runs Great Wall's Wey brand, on the four-ministry design rules: the greatest luxury is not the sofa, it is the brakes.

Go deeper on this section: ClaudeChatGPT

The river · 34

AI infrastructure startup Wuwen Xinqiong raises $324 million

# Zhidx 15h

Wuwen Xinqiong raised 2.3 billion yuan, roughly $324 million, taking cumulative funding to 4.3 billion yuan, roughly $606 million, on a platform for heterogeneous domestic and foreign AI chips.

More
  • The company says it has trained models across four Chinese chip families plus AMD and Nvidia, reporting 97.6 percent utilization in a 70-billion-parameter mixed-chip run.
  • Its platform targets the operational problem of fragmented compute: adapting to a new chip in two working days, pooling clusters across regions and scheduling different models over different accelerators.
  • A production system reportedly spans 37,000P of heterogeneous compute and 16 chip types, while a cross-cluster inference design cut first-token latency 51.5 percent and per-token cost 37.5 percent in the cited test.
  • The strategic positioning is deliberately neutral: Wuwen Xinqiong does not build the model or chip, but acts as the translation layer between fragmented hardware and rising token demand.
  • The financing and reported IPO preparation show Chinese capital treating AI infrastructure as a strategic industrial layer, not merely a supporting software category.
  • The numbers are company disclosures, so the key validation will be whether cost reductions and utilization hold across customers rather than controlled demonstrations.
Open at Zhidx Ask Claude ChatGPT

Huawei agent beats 25 unfamiliar games with a self-updating world model

# AI QbitAI 11h

Huawei's Memento 3 reportedly solved 183 ARC-AGI-3 levels with 56 percent fewer actions than humans by writing and revising a game simulator instead of changing model weights.

More
  • The system uses a code-as-model approach: it maintains a rulebook and executable simulator, tests hypotheses against observed frames and revises both when predictions fail.
  • It achieved a perfect 100 relative human action-efficiency score, using 7,518 actions against the human reference of 17,135, according to the supplied report.
  • The agent retained a history of rulebook versions with Git-like rollback, making recursive self-improvement auditable rather than an opaque sequence of prompts.
  • On Pong, it reportedly learned a control policy and then played three games without further large-model calls, suggesting that distilled local models can take over repetitive execution.
  • The result is impressive but benchmark-bound: pixel games provide a clean feedback loop, while real-world robots face partial observability, safety constraints and expensive physical mistakes.
  • The most transferable architectural idea is separating slow hypothesis formation from fast model-based action, a pattern relevant to embodied agents and tool-using software systems.
Open at QbitAI Ask Claude ChatGPT

A Chinese legal celebrity quietly redraws the limits of public testimony

# Culture Pekingnology 2h

A criminal lawyer with 1.59 million Bilibili followers says he can no longer read submissions from prison and judicial workers, offering a rare glimpse of censorship through an ordinary creator's workflow.

More
  • The lawyer, Cai Yaqi, publishes case-related experiences and follower accounts across Bilibili and WeChat, making his channels an informal window into lives that rarely reach national media.
  • His 39-second notice did not name a rule or censoring body; its matter-of-fact wording and warning that submissions should not go too far reveal how boundaries are often communicated indirectly.
  • The restriction is significant because it reaches a legal educator and criminal-defense practitioner rather than a conventional political commentator.
  • Chinese online speech often survives through euphemism, selective omission and creator self-censorship, so the absence of an explanation is itself part of the evidence.
  • The supplied article is incomplete, but it clearly contrasts the platform's continued openness to personal stories with a narrowing around state, prison and judicial administration topics.
Open at Pekingnology Ask Claude ChatGPT

China offers voluntary hybrid-export limits to avert an EU tariff fight

# Policy Initium 6h

Chinese and European negotiators reached an ambiguous understanding that could cut future Chinese hybrid exports by more than half, marking Beijing's first preemptive export restraint in this dispute.

Also: SCMP

More
  • The EU says the arrangement could reduce projected hybrid and plug-in hybrid shipments by more than half over four years, while China has not published the baseline or enforcement mechanism.
  • The talks also produced 16 points on rare-earth licensing, market access, export controls and a possible review of electric-vehicle trade measures.
  • China previously called voluntary export limits a WTO violation but now says the understanding is WTO-compliant, preserving room to interpret the arrangement without conceding a formal quota.
  • The strategic move trades some export volume for time and avoids an immediate EU safeguard or tariff escalation, while testing whether Brussels will reciprocate on access and licensing.
  • The unresolved baseline matters: the reduction may apply to projected future growth rather than current shipments, so the headline concession may be smaller than it sounds.
  • European officials describe this as a first step, not a settlement; implementation will show whether dialogue can replace the usual investigation, tariff and retaliation cycle.
Open at Initium Ask Claude ChatGPT

Shanghai multimodal startup MOSI raises more than $141 million

# AI Zhidx 12h

MOSI raised more than 1 billion yuan, roughly $141 million, in a Pre-A round led by state-linked funds, backing a real-time multimodal stack descended from Fudan research.

Also: New Wisdom

More
  • The company says it has released more than 20 models across speech, audio, transcription, vision and generation since the start of 2026.
  • Its planned MOSS Live system separates continuous front-end perception and conversation from back-end reasoning and action, targeting glasses, cars, home devices and robots.
  • The round is MOSI's second financing announcement this year; it previously raised several hundred million yuan, roughly $30-60 million, according to the supplied report.
  • State, industrial and private investors are funding a research-to-product path focused on multimodal interaction rather than a general chatbot race.
  • The technical ambition is substantial, but the commercial test is whether real-time sensing and action can be delivered within the latency, power and data constraints of edge devices.
Open at Zhidx Ask Claude ChatGPT

China turns embodied-AI competitions into simulated job auditions

# Policy Leiphone 3h

Shanghai's embodied-intelligence contest will score robots on real job tasks across 23 events, revealing a state-backed push to convert demonstrations into comparable industrial performance.

More
  • The contest covers manufacturing, inspection, maintenance, logistics, retail, healthcare, safety, emergency response and tourism across more than 1,000 expected teams.
  • Its four-part scoring framework measures task execution, embodied-data collection and feedback, autonomy and innovation, with different coefficients for humanoid, quadruped and wheeled systems.
  • Including teleoperated systems alongside autonomous robots acknowledges that near-term commercial value may come from hybrid labor rather than full autonomy.
  • The event is backed by national industrial agencies and Shanghai organizations, tying competitions to scenario data, financing and procurement pathways.
  • The strategic goal is a repeatable pipeline from simulated validation to field training, orders and overseas showcases; the open question is whether contest metrics predict reliability outside curated scenes.
Open at Leiphone Ask Claude ChatGPT

A former Robotaxi operator bets on open-world physical agents

# AI Leiphone 14h

Chinese startup Femo One is applying lessons from map-free driving to robots that operate in offices, homes and commercial districts, prioritizing transferable data over tightly scripted demos.

More
  • Founder Liu Nianqiu argues that high-definition-map driving locked systems into bounded environments, while map-free autonomy forced tighter coupling among perception, planning and data production.
  • Femo One calls its robots physical agents that can fetch items, buy coffee or handle waste rather than perform one fixed job, an ambitious but difficult product definition.
  • The company plans a deploy, collect, train and redeploy loop, with initial data collection this year and multi-site deployment next year.
  • Open environments expose the gap between a robot that succeeds once in a demo and one that can reliably act whenever a user asks, especially under weather, crowds and ambiguous instructions.
  • The interview is valuable because it explains the architecture migration: removing maps changes sensor selection, data economics and the boundaries between perception and planning.
Open at Leiphone Ask Claude ChatGPT

China is using AI microdramas as a public-opinion tool

# Policy SCMP 9h

Beijing is scaling AI-generated short dramas for ideological messaging, adapting a commercial format built around romance and revenge into a cheaper channel for policy and cultural narratives.

More
  • The campaign aligns with directives that treat technology as a tool for strengthening ideological work and guiding public opinion at home and abroad.
  • The format matters because microdramas already have high-frequency distribution, serialized attention and low production costs on Chinese platforms.
  • The supplied account contrasts state messaging with the sensational genres that dominate commercial AI content, showing how the same production stack can serve both entertainment and propaganda.
  • AI lowers the cost of producing variants for different audiences, but the excerpt does not establish reach, audience response or whether viewers distinguish official content from ordinary fiction.
  • This is less about model quality than about institutional adoption: publicity departments can use generative video as a scalable content pipeline while platforms retain distribution and moderation control.
Open at SCMP Ask Claude ChatGPT

AI reverse engineering lowers the barrier to copying software behavior

# Policy GeekPark 5h

Chinese coverage of the REA reverse-engineering tool stresses a China-specific legal boundary: technical analysis may be lawful for study, while copying product functionality can still trigger copyright, contract or trade-secret risk.

More
  • REA connects AI coding agents to tools such as Ghidra, IDA and Hopper through MCP, allowing a model to inspect binaries and return evidence-backed explanations.
  • The article says the technical novelty is limited because similar MCP integrations already existed; the meaningful change is making reverse engineering accessible to users who are not experts.
  • The supplied discussion notes that packed or virtualized software, dynamic behavior and server-side logic remain difficult, while long tool descriptions can consume the agent's context.
  • Chinese law may permit reverse engineering for learning or research under specified conditions, but commercial reuse and license restrictions create separate risks; the article also compares this with US case law.
  • The broader signal is defensive and competitive: agents can shorten analysis of legacy clients, proprietary interfaces and games, while simultaneously lowering the cost of cloning.
Open at GeekPark Ask Claude ChatGPT

QCon Shanghai turns from AI coding to production-grade agent loops

# AI InfoQ China 25m

QCon Shanghai frames the next AI engineering problem as verification: agents must plan, execute, observe and recover across real business workflows, not merely generate text or code.

More
  • The conference program spans agent harnesses, evaluation, observability, infrastructure, SRE and embodied systems, showing that Chinese engineering teams increasingly treat agents as production systems rather than chat interfaces.
  • The most substantive idea is the move from coding to commerce: travel and other transactions need real-time supply, service and fulfillment signals that can verify whether an agent actually succeeded.
  • The proposed architecture resembles a closed-loop control system: planning, tool execution, monitoring, validation, feedback and re-execution replace one-shot inference.
  • Chinese coverage presents this as an industrialization challenge, with emphasis on turning energy, compute and tokens into measurable business output rather than celebrating model capability alone.
  • The event is promotional, so the claims are agenda-setting rather than independent evidence of deployment scale; the useful signal is which engineering problems Chinese companies now consider bottlenecks.
Open at InfoQ China Ask Claude ChatGPT

Zhipu quietly tests a model claiming nearly 500 tokens per second

# AI Zhidx 14h

Zhipu is gray-testing GLM 5.3 HighSpeed in its coding product, with one user reporting more than 490 tokens per second; the real test is whether that speed survives production scale.

Also: QbitAI

More
  • The model reportedly retains a million-token context window and up to 131,072 output tokens, making throughput especially relevant to long agentic coding tasks.
  • The reported speed is an unverified gray-market observation, not a published service-level guarantee; prior Zhipu high-speed versions reached about 400 tokens per second in limited access.
  • The account attributes the gains to persistent GPU kernels, dynamic batching, request merging and KV-cache scheduling, illustrating that inference systems, not only model weights, determine agent latency.
  • This is the same model event as item 91; this account provides the infrastructure explanation.
Open at Zhidx Ask Claude ChatGPT

Microsoft builds a decision model on Alibaba's Qwen

# AI Solidot 11h

Microsoft's Decision-1 is reportedly post-trained from Alibaba's Qwen3.5-9B, showing Chinese open weights becoming a base layer for compact enterprise decision systems.

Also: Leiphone

More
  • Decision models output predictions, probabilities and confidence rather than long natural-language answers, making them suited to repeated classification and operational choices.
  • Microsoft says the model processed more than 10,000 Xbox player-feedback records and is being tested for incident response, quality control and scientific discovery.
  • The company claims better latency and accuracy than several comparison systems and a cost below one-twentieth of a large proprietary model on text classification.
  • The strategic signal is that open Chinese weights can travel through the AI stack even when the originating model is not the end product; data, post-training and deployment may matter more than brand.
  • Microsoft says it may later replace Qwen with its own or OpenAI models, so this is evidence of practical utility, not a lasting dependency.
Open at Solidot Ask Claude ChatGPT

China's hidden AI workers may be masking productivity gains

# Culture Initium 18h

Initium examines workers who use AI quietly while concealing it from employers, a social signal about trust, evaluation and the gap between formal output and actual work.

More
  • The framing suggests that AI adoption is often invisible: employees may use models to accelerate tasks while preserving the appearance of unaided effort.
  • That behavior makes conventional productivity measurement unreliable and can create a conflict between managerial surveillance and worker self-protection.
  • The supplied excerpt is incomplete, so it does not establish prevalence or sector-level effects, but it identifies a meaningful workplace norm emerging around AI disclosure.
  • In China, where hierarchy and performance signaling strongly shape employment, hidden use may be a rational response to fear that visible efficiency will become a reason to cut staffing.
Open at Initium Ask Claude ChatGPT

Beijing sanctions Paraguayan and Belizean ministers over Taiwan

# Policy SCMP 11h

China imposed asset freezes and related sanctions on ministers from Paraguay and Belize after their governments deepened ties with Taiwan, showing Beijing's willingness to personalize diplomatic pressure.

More
  • The targets are Paraguay's information and communications minister and Belize's foreign minister, among other portfolios.
  • The measures are a direct response to official dealings with Taipei rather than a broad economic sanction package against either country.
  • The action reinforces a familiar Chinese diplomatic pattern: Taiwan recognition disputes are made concrete through named individuals, not left at the level of statements.
Open at SCMP Ask Claude ChatGPT

Anta puts agents into global ordering and returns workflows

# AI Leiphone 7h

Anta says agents cut overseas ordering decisions from half a day to 10 to 20 minutes and automate more than 1,000 monthly return orders, a concrete Chinese enterprise-AI deployment.

More
  • The ordering agent converts four overlapping spreadsheets into reusable skills for item codes, category selection, translation, formulas and formatting, reducing manual entry by about 90 percent.
  • A TikTok cross-border returns agent pulls platform data into Anta's ERP and creates more than 1,000 return documents per month for the Anta brand.
  • A video agent chains trend collection, content analysis, generation and multi-site publishing, showing that the value comes from workflow integration rather than a standalone chatbot.
  • Anta's own test for production maturity is whether a process depends on the agent, whether the agent is part of the default workflow and whether it can be copied across countries.
  • The figures are company-reported and presented through a vendor partnership, so independent measurement of error rates and labor displacement is still missing.
Open at Leiphone Ask Claude ChatGPT

China's auto regulators target shortcut-heavy vehicle design

# Policy Leiphone 11h

A Chinese regulatory draft calls for stronger testing of novel controls, handles and seats, while a Great Wall executive says luxury begins with braking rather than cabin furniture.

More
  • The draft from four government departments focuses on verification of rapid vehicle-design changes, including virtual controls, door handles and new human-machine interfaces.
  • The industry's attention to brake-pedal materials suggests the policy push may extend from styling into safety-critical component validation.
  • The same roundup also carries unconfirmed reports of a Huawei midrange sub-brand and domestic personal-agent projects, but those claims are not official and should not be treated as facts.
  • The regulatory signal is more durable: China is trying to slow feature races in connected vehicles by requiring evidence that novel interfaces and hardware work safely at scale.
Open at Leiphone Ask Claude ChatGPT

Gasoline cars fall below half of global new-vehicle sales

# Signal Solidot 10h

Gasoline-only vehicles represented 49 percent of global new-car sales in the first half, with China's gasoline sales down 26 percent as electric and hybrid adoption accelerates.

More
  • The data says battery-electric sales rose 12 percent to 6.87 million and conventional hybrids rose 10 percent to 7.27 million, while the overall auto market contracted about 5 percent.
  • China recorded the largest decline in gasoline-car sales among major regions, while European battery-electric sales rose 32 percent to 1.81 million.
  • The shift is not simply an EV story: hybrids account for a large share of the transition, especially where charging infrastructure and long-distance use remain constraints.
  • The report attributes some pressure to an oil shock, so the durability of the gasoline decline depends on fuel prices as well as technology and regulation.
Open at Solidot Ask Claude ChatGPT

OnePlus 16 starts at roughly $803 before China's subsidy

# Signal Zhidx 3h

OnePlus says its new flagship combines native 185fps gaming, a 165Hz global refresh mode, a 9,000mAh battery and a 200-megapixel camera to broaden its gaming-led appeal.

Also: Leiphone

More
  • The phone uses a Snapdragon flagship plus dedicated network, display and touch chips, making the gaming claim a full-stack scheduling and I/O effort.
  • OnePlus reports 520Hz multi-touch sampling, 100W wired charging and 50W wireless charging, alongside a custom high-refresh panel developed with BOE.
  • The entry model costs 5,699 yuan, roughly $803, or 5,199 yuan, roughly $732, after subsidy; higher configurations rise to 8,199 yuan, roughly $1,155.
  • The supplied report says OnePlus sales were up 18 percent year over year, though that is a company claim without independent market-share evidence.
  • This is the same launch as item 7; its value here is the pricing and sales-positioning detail.
Open at Zhidx Ask Claude ChatGPT

FAST confirms a rare pulsar triple system

# Signal Solidot 11h

China's FAST telescope helped confirm the first known evolving triple system containing a native pulsar, a strong domestic science result built on five years of monitoring.

More
  • The pulsar spins every 3.2 milliseconds and orbits a white dwarf every eight days, while a roughly solar-mass star circles the inner pair every 73.5 years.
  • Researchers used 192 FAST measurements plus optical, infrared and gamma-ray archives to identify the distant companion and explain the pulsar's unusual spin change.
  • The system offers a natural laboratory for stellar evolution and multi-body gravity, including the effect of an outer companion on a compact binary.
Open at Solidot Ask Claude ChatGPT

China builds a molten-salt and supercritical-CO2 power plant

# Signal SCMP 9h

A Shandong project combines supercritical carbon-dioxide generation with large-scale molten-salt storage to reduce waste and stabilize power from a coal-site platform.

More
  • The Ruitan project is intended to demonstrate how high-temperature storage and a supercritical CO2 cycle can improve conversion efficiency and grid flexibility.
  • Its significance is architectural: storage and generation are being designed as one thermal system rather than adding batteries after a conventional plant.
  • The project remains under construction, so its commercial efficiency and reliability are not yet established.
Open at SCMP Ask Claude ChatGPT

Orbbec drops lawn-robot ODM after only thousands of units shipped

# Signal Leiphone 10h

Chinese 3D-vision company Orbbec is abandoning lawn-mower robot manufacturing after weak volume and tighter US rules, returning to sensor modules where its capabilities fit better.

More
  • The ODM program shipped only several thousand units, below the scale needed to support a complete hardware business and dependent on one customer's distribution strength.
  • The report cites new US FCC controls on advanced foreign-made robots, a crowded lawn-robot market and weak customer volume as the combined reasons for the retreat.
  • Orbbec will continue supplying perception modules for outdoor robots, preserving the higher-leverage part of its position in the stack.
  • The move illustrates the gap between a component supplier's technical competence and the customer, channel, compliance and service capabilities required to sell finished robots.
Open at Leiphone Ask Claude ChatGPT

Chinese desalination plant cuts energy use by 60 percent

# Signal SCMP 13h

A Bohai Bay desalination plant claims to cut electricity use by more than half while turning waste brine into industrial salt, linking water infrastructure to chemical recovery.

More
  • The Caofeidian facility entered trial production with a process designed to recover value from millions of tonnes of concentrated brine rather than treating it only as waste.
  • The reported energy reduction addresses the main operating constraint of desalination, while salt recovery could improve project economics and reduce discharge impacts.
  • The claims come from state media coverage of a trial-stage facility; long-term output quality, maintenance and full-system energy accounting remain unverified.
Open at SCMP Ask Claude ChatGPT

China's AI rollout is managed by a web of enforcers

# Policy The Wire China 20h

A Wire China profile maps the Chinese government bodies overseeing AI deployment, a useful governance angle even though the supplied description lacks institutional detail.

More
  • The item promises coverage of the agencies responsible for managing China's AI rollout rather than treating regulation as a single ministry function.
  • That fragmented governance matters because model safety, data, cybersecurity, industry policy and ideological controls can impose different requirements on one product.
  • The supplied text is only a description, so specific agency roles and enforcement examples remain missing.
Open at The Wire China Ask Claude ChatGPT

Wang Yi visits four countries to reinforce China's trade corridors

# Policy SCMP 5h

China's foreign minister will visit Azerbaijan and three Balkan states while attending a regional security summit, underscoring Beijing's effort to protect trade routes and political influence.

More
  • The itinerary includes Azerbaijan, Croatia, Montenegro and Bulgaria, linking Central Asian connectivity with Balkan commercial relationships.
  • The ministry says Wang will attend the Conference on Interaction and Confidence-Building Measures in Asia as President Xi's special representative.
  • The visit is diplomatic positioning rather than a single deal, but it shows how Beijing uses senior-level travel to keep corridor politics connected to trade.
Open at SCMP Ask Claude ChatGPT

Uyghurs from Xinjiang face statelessness after fighting in Syria

# Culture The Wire China 20h

A Wire China profile follows Uyghurs from Xinjiang who fought in Syria and now face uncertain legal status, connecting China's minority policies to transnational conflict and citizenship.

More
  • The supplied description is brief and does not establish how many people are affected or which governments recognize their status.
  • Its value is the cross-border human consequence: displacement and conflict can leave people politically unrepresented even after the battlefield ends.
Open at The Wire China Ask Claude ChatGPT

Huaren Pharma sues over a $450 million bad-debt exposure

# Policy TMTPost 4h

Huaren Pharma is suing over more than 320 million yuan, roughly $45 million, after a 1.418 billion yuan, roughly $200 million, receivable became a governance and recovery crisis.

More
  • The lawsuit seeks 308 million yuan in principal plus interest, about $43 million before interest, from a nominal debtor and a jointly liable pharmaceutical company.
  • The claim is roughly 11 times Huaren Pharma's first-half net profit, making recovery materially important even though the company fully impaired the larger receivable in late 2024.
  • The proposed debt restructuring through equity and assets failed to execute, pushing the company from negotiation toward litigation and possible insolvency or recovery proceedings.
  • Regulators cited weak controls over trade business, contract monitoring and disclosure, linking the bad debt to institutional governance rather than a single missed payment.
  • The case illustrates how Chinese listed companies can carry large trade receivables outside their core operating model while local courts, regulators and minority investors pressure management to recover them.
Open at TMTPost Ask Claude ChatGPT

Huawei's eight-year foldable strategy meets Apple's first attempt

# Culture Leiphone 7h

Chinese coverage argues that Huawei's foldable lead comes from eight years of hinge, form-factor and supply-chain iteration, not one launch-cycle feature comparison.

More
  • The article says Huawei covers outward-folding, inward-folding, clamshell, wide and dual-fold designs, using successive generations to reduce weight and improve durability.
  • It presents the hinge and large-screen multitasking as the real product moat, while Apple's first foldable has not yet completed a full market-validation cycle.
  • Huawei's Mate X7 starts at 11,999 yuan, roughly $1,690, while the older Mate X6 has fallen to 8,999 yuan, roughly $1,270, illustrating how accumulated manufacturing experience can lower entry prices.
  • The supplied sales claim that Huawei held 60 percent of China's foldable market in the first quarter comes from IDC as cited by the article, but the piece itself is strongly promotional.
  • The broader China signal is that premium hardware competition is increasingly about accumulated reliability data and ecosystem adaptation rather than novelty at launch.
Open at Leiphone Ask Claude ChatGPT

Chinese investors rush into orbital AI data centers

# Analysis 36Kr 14h

Chinese space-computing startups are attracting capital after SpaceX and other foreign players made orbital AI infrastructure credible, but the sector remains heavy on narrative and light on proven economics.

Also: The Wire China

More
  • The thesis combines persistent solar power and radiative cooling with orbital compute, potentially reducing terrestrial electricity and thermal constraints.
  • Chinese startups have reportedly raised more than 100 million yuan, roughly $14 million, in individual rounds, while several new companies appeared in the first half of 2026.
  • The stronger near-term use case may be processing satellite data in orbit for oceans, agriculture, emissions and disaster monitoring, rather than sending general-purpose data-center workloads into space.
  • Investors describe a progression from smart satellites to cooperative constellations and eventually orbital data centers, but launch, radiation, interconnect and maintenance costs remain unresolved.
  • Chinese enthusiasm is partly an example of policy and capital following a foreign technology narrative: the article explicitly describes investors changing their view after Elon Musk's announcements.
  • This is the same broader China orbital-computing theme as item 101, but the account adds the financing and speculative-market angle.
Open at 36Kr Ask Claude ChatGPT

The US is pursuing cheaper hypersonic missiles aimed at Chinese defenses

# Analysis SCMP 5h

The US Navy is pursuing mass-produced Blackbeard hypersonic missiles for carrier aircraft, a China-relevant shift from exquisite weapons toward salvos that could stress coastal and naval defenses.

More
  • Castelion's missile is designed to exceed Mach 5 and strike ships or land targets while emphasizing lower unit cost than existing hypersonic systems.
  • The supplied report frames Chinese warships and coastal installations as the relevant defensive problem, making affordability and volume as important as speed.
  • The strategic question is whether production scale, targeting networks and survivability can be achieved together; the excerpt does not establish a deployment date or quantity.
Open at SCMP Ask Claude ChatGPT

Office AI vendors are pricing against labor budgets, not just seats

# Culture TMTPost 8h

Chinese analysis argues that agents are changing enterprise software from per-seat productivity tools toward per-action services, while employers may freeze hiring before the technology is fully reliable.

More
  • Tencent's WorkBuddy still uses a 198-yuan monthly seat model, while Salesforce charges per agent action and Intercom charges for some successful outcomes, signaling a shift in software economics.
  • The new pricing lets vendors compete directly with customer-service, operations and analyst labor costs, not merely with other software subscriptions.
  • The article cites a 2026 McKinsey survey in which 39 percent expected workforce reductions from AI within a year, versus 14 percent reporting that AI had already reduced headcount in the prior year.
  • The gap between expectation and realized cuts suggests that management is pricing in future capability before agents have proven stable performance in complex work.
  • For China, the key social implication is that canceled hiring plans may create labor-market pressure without the visible announcement of layoffs.
Open at TMTPost Ask Claude ChatGPT

Chinese storage firms booked 93% of disclosed Q3 overseas orders

# Signal TMTPost 4h

Chinese storage companies disclosed about 74 GWh of system orders in the third quarter, with 67.8 GWh overseas, showing how domestic suppliers are exporting the energy-storage buildout.

More
  • The 93 percent overseas share comes from an incomplete sample of disclosed system contracts, not the entire industry; the report also counts about 246.6 GWh of cell orders.
  • Europe accounted for roughly 31.2 GWh, or 42 percent of disclosed system contracts, while the Middle East represented about 21.3 GWh, or 28.7 percent.
  • The geographic mix shows Chinese vendors competing in both mature European markets and faster-moving Middle Eastern and Asia-Pacific projects.
  • The same roundup says TikTok Shop Mexico's daily gross merchandise value rose 133 times in a year, but the storage figure is the more durable industrial signal.
  • For US technical readers, the key implication is that Chinese battery and system firms are increasingly dependent on foreign project pipelines even as domestic demand remains a foundation.
Open at TMTPost Ask Claude ChatGPT

A 55-year-old graduate student brings AI into China's tea industry

# Culture TMTPost 8h

A 55-year-old former programmer used AI to prepare for a tea-processing master's degree and now builds sensor-driven jasmine-tea equipment, an unusually human view of retraining and late-career insecurity.

More
  • Xiao Chaohu used models to classify ten years of exam questions, count topic frequency and generate practice tests against scanned reference books.
  • He later connected sensors, machine vision, data collection and software to control tea scenting, combining an old computing background with agricultural engineering.
  • His return to education followed shrinking media work, weaker foreign technology budgets in China and a family tea farm whose raw product captured little value.
  • The story exposes a labor-market problem that technology narratives often miss: even highly adaptable older workers face a narrowing path back into formal employment.
  • His case also shows AI as an equalizer for self-directed study, but not as a substitute for institutional credentials, field knowledge or a viable local tea business.
Open at TMTPost Ask Claude ChatGPT

Taiwan considers keeping fighter pilots until age 58

# Analysis SCMP 7h

Taiwan plans to extend military pilot service as 66 US-built F-16Vs arrive, an adjacent China-security story about the human bottleneck behind advanced hardware.

More
  • The proposed retention age would rise from about 48 for lieutenant colonels to 58, addressing the risk of acquiring aircraft faster than experienced crews can be trained.
  • The supplied report frames the issue as having fighters but not enough pilots, a common constraint when complex systems expand faster than skilled labor pools.
  • The China relevance is strategic rather than commercial: pilot retention affects Taiwan's readiness and the operational value of its US-supplied fleet.
Open at SCMP Ask Claude ChatGPT

Anhui becomes the Yangtze Delta's innovation engine

# Analysis TMTPost 8h

Chinese analysis casts Anhui as the Yangtze Delta's industrial core, using memory chips, vehicles and fusion research to show how inland provinces can build new technology clusters rather than merely receive relocated factories.

More
  • The article contrasts the traditional Shanghai-Jiangsu-Zhejiang hierarchy with Anhui's growing role in long-cycle R&D and advanced manufacturing.
  • It highlights the EAST fusion device's 1,066-second high-temperature plasma run and Anhui's first-place national rankings in vehicle production, new-energy vehicles and auto exports.
  • The argument is institutional: long-term university, government and industrial coordination can turn a region with weaker initial geography into an innovation node.
  • The reported rise in projects signed with neighboring provinces suggests that capital and supply chains are increasingly flowing toward technical capability, not simply toward the highest-GDP cities.
  • The analysis is rhetorical and celebratory, but it offers a useful map of how regional policy frames Anhui's role in China's next manufacturing cycle.
Open at TMTPost Ask Claude ChatGPT
Source ledger
ChinAI 403 0/0 HTTPError: 403 Client Error: Forbidden for url: https://chinai.substack.com/feed
ChinaTalk 200 0/20
Pekingnology 200 1/20
Ginger River Review 200 0/20
Sinocism 200 0/20
QbitAI 200 2/10
36Kr 200 4/30
Solidot 200 10/20
Leiphone 200 12/20
GeekPark 200 1/30
IT Home 200 12/60 +35 over cap
SCMP 200 10/50 +4 over cap
Initium 200 4/15
Jiemian 200 8/30 +15 over cap
China Digital Times 200 0/15
The Wire China 200 5/50
Techmeme 200 1/15
Hacker News 200 0/20
V2EX 200 8/50 +39 over cap
Zhidx 200 4/20
InfoQ China 200 7/20
New Wisdom 200 8/15 +2 over cap
TMTPost 200 8/16 +8 over cap