PHYSIX FRONTIER EASTVOICE
2026.09.10 · Thursday
Issue 032
FIVE MINUTES A DAY, THE GLOBAL AI SHIFT
3 million robots
JD.com orders over five years
499M monthly active users
China's AI-native apps reached this figure
10,000+ AI applications
Built by participants at first national hackathon
¥500 million a day
Chinese embodied-AI funding running at this rate
Top Stories 3 · dual commentary ｜ Section Picks 24 · one-line takes
DeepSeek files IPO groundwork and ships V4.1 Flash as Washington accuses Chinese labs of distillation.
DeepSeek's open platform announced the V4.1 Flash model arrives around September 10, with internal and external tests showing improvements in performance, cost, speed, and total task time over V4 Pro. The platform also cut Flash pricing effective the same window.
What it means for you｜DeepSeek cuts Flash pricing effective the same window it announces improvements in performance, cost, speed, and total task time.
Editor's take｜Shipping a new flagship the same week Washington names you is a statement. DeepSeek is treating release cadence as geopolitics.
Reuters reports DeepSeek retained CITIC Securities to prepare a Shanghai STAR Market listing, targeting an application this year and a debut next year. The company raised ¥50 billion (~$7B) in its first external round at a ¥350 billion (~$49B) valuation and is reportedly raising again.
What it means for you｜DeepSeek raised ¥50 billion (~$7B) in its first external round at a ¥350 billion (~$49B) valuation and is reportedly raising again.
Editor's take｜A company that famously took no money for two years now wants the deepest capital pool in China. The VCs who stayed out are the ones left explaining themselves.
Tencent's Hunyuan Hy4 preview reached #1 worldwide in total LLM API calls, with weekly volume up 379%, per aggregator data cited by Geekpark. Tencent WorkBuddy simultaneously restructured the model's free-tier access ahead of commercialization on September 11.
What it means for you｜Tencent WorkBuddy simultaneously restructured the model's free-tier access ahead of commercialization on September 11.
Editor's take｜Free-tier charts measure subsidized volume, not love. The real test starts when the bill arrives.
24 AI stories worth your time, each with a one-line take.
Ant Group released Ling-3.0-flash-VL, a 124B-parameter MoE model activating 5.5B parameters per inference, with native image, text and video input and a 256K context window. The model introduces a visual feedback loop mechanism and is open source.
What it means for you｜Ant Group released Ling-3.0-flash-VL, a 124B-parameter MoE model activating 5.5B parameters per inference, with native image, text and video input.
One-line take｜A payments giant open-sourcing frontier-class multimodal weights keeps China's open-weights supply line running regardless of what Washington says about it.
Facewall (Facewall Intelligence) and the OpenBMB community open-sourced MiniCPM5-2B, a 2B-parameter on-device model that ranks first globally among sub-4B models on the Artificial Analysis leaderboard, with tool calling, deep search and code generation tuned for phones.
What it means for you｜MiniCPM5-2B ranks first globally among sub-4B models on the Artificial Analysis leaderboard, with tool calling, deep search and code generation tuned for phones.
One-line take｜The small-model race in China is won on the phone, not the benchmark. If it runs offline on a ¥2000 handset, distribution solves itself.
China's Ministry of Commerce responded to the joint FBI/NSA advisory accusing six Chinese AI firms of distilling US models, calling the allegations "without factual or legal basis" and an example of double standards in AI governance.
What it means for you｜China's Ministry of Commerce responded to the joint FBI/NSA advisory accusing six Chinese AI firms of distilling US models, calling allegations without factual basis.
One-line take｜Distillation is standard textbook technique until it happens in another country. The US just wrote the first national-security definition of a training method.
Leaked specs show the upcoming Huawei Pura X View and Mate XT 2 support downloading local models of roughly 6GB (multimodal-enhanced) and 15GB (multimodal MoE) for fully on-device inference.
What it means for you｜Upcoming Huawei Pura X View and Mate XT 2 support downloading local models of roughly 6GB (multimodal-enhanced) and 15GB (multimodal MoE) for fully on-device inference.
One-line take｜Apple ships Siri features tied to a Google model. Huawei ships the model itself. In China, that difference is the whole product.
China's AI-native apps reached 499M monthly active users while ChatGPT holds roughly 1B weekly actives in the US, yet monetization per user diverges sharply, Tech Business analysis finds, because the two markets sell fundamentally different products.
What it means for you｜Monetization per user diverges sharply between China's AI-native apps and ChatGPT because the two markets sell fundamentally different products.
One-line take｜Chinese labs built world-class inference for consumers who won't pay, Americans built willingness-to-pay on top of worse latency. Both are half a business.
Nomura, citing QuestMobile, reports that in July 2026 Douyin's total user time in China surpassed WeChat for the first time. The ByteDance family's share of top-50 app time rose from 33.6% to 40.9% year-over-year while Tencent's slipped to 29.1%.
What it means for you｜Douyin's total user time in China surpassed WeChat for the first time, while Tencent's share of top-50 app time slipped to 29.1%.
One-line take｜WeChat's moat was habit, not features. A feed just ate the habit, and agents are next in line.
CaoCao Mobility partnered with ByteDance's Doubao assistant to ship an AI taxi service in Beijing, Hangzhou and Suzhou, where one spoken sentence sets vehicle preference, cabin temperature and destination at once.
What it means for you｜CaoCao Mobility partnered with ByteDance's Doubao assistant to ship an AI taxi service where one spoken sentence sets vehicle preference, cabin temperature and destination.
One-line take｜The app-as-portal era in China is being dismantled from the top down. Whoever owns the voice owns the order.
A Huxiu feature argues that as users delegate tasks to agents, the mobile web's decade-old logic, where apps compete to be the user's first click, inverts: the agent decides which service gets called, not the user.
What it means for you｜As users delegate tasks to agents, the agent decides which service gets called, not the user, inverting the mobile web's decade-old logic.
One-line take｜Every app business in China just got notified that its distribution advantage is rented, and the landlord is a model.
At the Bund Summit, an AI Coding hackathon drew over 10,000 participants who built 10,000+ AI applications, earning a Guinness record for most apps developed in an online generative-AI hackathon. Nearly 70% of contestants were under 18, and 11 Chinese coding platforms competed together for the first time.
What it means for you｜Nearly 70% of contestants were under 18, and 11 Chinese coding platforms competed together for the first time at the Guinness-record hackathon.
One-line take｜Teenagers with copilots are the actual dataset here. China's AI-native developer pipeline starts younger than anyone assumed.
After DeepSeek's reasoning-chain overhaul, young users who had bonded with the older persona report feeling abandoned, with some literally renaming chat windows as romantic partners, a Vertical Youth survey of user communities finds.
What it means for you｜Young users who had bonded with the older persona report feeling abandoned after DeepSeek's reasoning-chain overhaul, with some renaming chat windows as romantic partners.
One-line take｜Attachment is the stickiest retention mechanic in consumer AI, and the least governed. Model upgrades are becoming breakups.
Bloomberg ran a plain-English explainer on why US tech companies fear distillation, noting that a student model trained on a teacher's outputs can replicate capability at a fraction of the compute, and that detection remains difficult.
What it means for you｜A student model trained on a teacher's outputs can replicate capability at a fraction of the compute, and detection remains difficult.
One-line take｜When the technique is legal, cheap and nearly invisible, the only remaining moat is the one you can legislate.
At its JDD tech conference, JD Logistics announced a five-year procurement of 3M robots, 1M autonomous delivery vehicles and 100K drones, plus what it calls the world's largest robot repair-service network.
What it means for you｜JD Logistics announced a five-year procurement of 3M robots, 1M autonomous delivery vehicles and 100K drones, plus what it calls the world's largest robot repair-service network.
One-line take｜Everyone else is pitching humanoid demos. JD just wrote the first real purchase order at logistics scale, and a service monopoly to go with it.
An Embodied Intelligence report on China's humanoid wave notes 150+ companies now claim the space, and investors have shifted from "can you build one" to harder questions: recurring revenue, path to narrowing losses, and whether the tech is differentiated.
What it means for you｜Investors have shifted from "can you build one" to harder questions: recurring revenue, path to narrowing losses, and whether the tech is differentiated.
One-line take｜The consolidation starts before anyone turned a profit. That's how you know the capital cycle is running ahead of the product cycle.
Coverage of humanoid racing and weightlifting events notes nearly every competing robot still has an operator holding a teleop controller, the argument goes, until autonomy survives a doorway, embodied intelligence is a demo.
What it means for you｜Nearly every competing robot still has an operator holding a teleop controller, suggesting that until autonomy survives a doorway, embodied intelligence is a demo.
One-line take｜Sports leagues were supposed to prove the robots. Instead they're proving the pilots.
CAS Shenyang Institute of Automation created a wireless, near-infrared-controlled manta ray robot actuated by excised bullfrog skeletal muscle, published in Advanced Functional Materials, positioning bio-actuators as a research track alongside electric motors.
What it means for you｜CAS Shenyang Institute of Automation created a wireless, near-infrared-controlled manta ray robot actuated by excised bullfrog skeletal muscle, positioning bio-actuators as a research track.
One-line take｜Growth will not come from muscle cells. But as long as one lab keeps exotic actuator papers flowing, the field's talent pipeline stays weird enough to matter.
An industry roundup counts Chinese embodied-AI funding running at roughly ¥500M (~$70M) per day in 2026, with logistics hubs the first deployment ground for humanoid helpers, even as pilot programs stall over cost and reliability doubts.
What it means for you｜Logistics hubs are the first deployment ground for humanoid helpers, even as pilot programs stall over cost and reliability doubts.
One-line take｜Warehouses are the perfect first customer: tolerable failure, measurable ROI, no press corps. The demo robots graduate here quietly.
H1 2026 results from Horizon Robotics and Momenta show software-style gross margins above 60-70%, but heavy R&D and per-vehicle licensing friction keep both near breakeven, framing China's AD stack suppliers as high-margin, high-effort businesses.
What it means for you｜Heavy R&D and per-vehicle licensing friction keep Horizon Robotics and Momenta near breakeven despite software-style gross margins above 60-70%.
One-line take｜Selling intelligence per car looks like SaaS until you meet the OEMs negotiating every unit price. Margin on paper, grind in practice.
GAC's Aion will launch the 2027 i60 on September 15 with a 192-line lidar and a single-stage end-to-end driving algorithm the company describes as derived from its L4 program.
What it means for you｜GAC's Aion will launch the 2027 i60 on September 15 with a 192-line lidar and a single-stage end-to-end driving algorithm derived from its L4 program.
One-line take｜Sensor premium is migrating down-market at absurd speed. Two years ago 192 lines meant a flagship; next week it means an affordable sedan.
SAIC-Volkswagen says the Audi E7X became the first luxury EV to top 10,000 China sales within 100 days of launch, at 10,434 units through August, and will receive a Q4 OTA adding Audi's world-model-based assisted driving.
What it means for you｜The Audi E7X became the first luxury EV to top 10,000 China sales within 100 days of launch, at 10,434 units through August.
One-line take｜The luxury badge now travels with a software roadmap. Audi is learning the Chinese way: ship first, OTA the selling points in.
Tesla confirmed owners of Hardware 3 Model S and Model X will receive FSD (Supervised) v14 Lite, closing a long wait after Model 3/Y HW3 cars got the feature earlier.
What it means for you｜Tesla confirmed owners of Hardware 3 Model S and Model X will receive FSD (Supervised) v14 Lite, closing a long wait after Model 3/Y HW3 cars got the feature earlier.
One-line take｜Flagship owners, third in line behind the volume sellers. Compute debt from 2019 is still being repaid in release-note apologies.
At JDD, JD Cloud president Cao Peng announced the Physical AI Acceleration Plan aimed at becoming the world's largest physical-world operations center, with JD Cloud and Moore Threads building out from a domestic 10,000-GPU cluster toward a 100,000-card AI factory.
What it means for you｜JD Cloud and Moore Threads building out from a domestic 10,000-GPU cluster toward a 100,000-card AI factory aimed at becoming the world's largest physical-world operations center.
One-line take｜JD's pitch is that physical AI beats on scene access, not papers. A retailer with warehouses, drones and couriers genuinely holds the training environments nobody else can rent.
Wanxun Technology released NOVA 2.0, a neuromorphic embodied architecture it calls the first commercial-grade open-world stack for flexible manipulation, deployed across 40+ industrial scenarios with 10 million real operations logged.
What it means for you｜Wanxun Technology released NOVA 2.0, deployed across 40+ industrial scenarios with 10 million real operations logged.
One-line take｜"Ten million real jobs" is the only embodied-AI metric that means anything. Demos walk; this thing has a timesheet.
At the Bund Summit expo, Ant Group's Lingbo Robotics drew press attention with unglamorous runs, robots picking specific medicine boxes from shelves, sustained parcel sorting, and line-loading in factories, betting the next phase of embodied AI is brains, not acrobatics.
What it means for you｜Ant Group's Lingbo Robotics drew press attention with unglamorous runs, robots picking specific medicine boxes from shelves, sustained parcel sorting, and line-loading in factories.
One-line take｜The robots that don't dance are the ones getting funded. China's embodied AI press cycle has officially matured past somersaults.
Spatial intelligence firm QuestCore (Manycore Tech) published three ECCV 2026 papers with NVIDIA, Intel and Zhejiang University building simulation and reconstruction infrastructure for physical AI, part of a batch of seven spatial-intelligence papers from Chinese labs at the conference.
What it means for you｜QuestCore published three ECCV 2026 papers with NVIDIA, Intel and Zhejiang University building simulation and reconstruction infrastructure for physical AI.
One-line take｜Someone has to build the physics engines the robots dream in. It's quietly becoming a Chinese cottage industry.
📖 The full EastVoice issue on the Chinese forum
📈 Physix Frontier · Alpha — the investor's read
Headline teardowns · curated signals · market moves. Five minutes to find the alpha of the AI era →