AI Agents Become the New Interface; Edge Hardware and Embodied AI Get Dual Boost
State AI+ plan targets >70% smart-device penetration by 2027
Tech Trends
The commercialization rhythm of frontier tech: AI, LLMs, XR and compute.
Tech Trends
Robot 'Brain' Hits Inflection; General Models Act Zero-Shot
GPT-6 Astra zero-shot generalization bridges hear-to-do gap
China Captures ~90% of Enterprise Open-Source Model Spend, Menlo Finds
Menlo: 89% of corporate open-model spending flows to Chinese vendors (Kimi 61%)
MIIT Issues Agent-Device Network-Access Rules, Shifting AI Governance to Terminals
China's first systemic rules for agent-equipped devices; curbs forced invocation
Wuzhen summit spotlights physical AI: embodied intelligence nears its 'last few millimeters'
Wuzhen expo puts embodied AI in focus; mass-production inflection meets fine-manipulation bottleneck
China Mobile debuts 'Lingtao Brain' at Mobile Network Congress, pushing embodied AI toward networked intelligence
Dual-brain synergy plus embodied network moves robots from standalone to networked intelligence
Import Expo spotlights sovereign AI chips and edge compute as physical-AI base takes shape
CIIE showcases autonomous AI chips and edge compute for physical AI
Embodied AI hits volume-procurement inflection as JD plans 3 million robots in five years
Embodied intelligence moves from demos to volume buys, maturing supply chain
Near-Memory Compute Chip Validated as AI Inference Silicon Race Intensifies
Qualcomm HBC tape-out and Mojin funding spotlight inference-chip race
Embodied AI Moves Into Factories and Pharmacies as Deployments Multiply
Xiaomi 8-hour shift and Ant LingBot pharmacy show real-world robotics
Tencent Hunyuan Launches Hy4: 770B MoE with Vision and Computer Use
Domestic model jumps to 770B parameters, open-weights and agent execution
Wuzhen Summit Elevates Open Source to Theme; 20 Bodies Issue Agent Self-Regulation Pact
Verifiable identity badges in focus as openness and governance advance
China's AI Inference Data Volume Surpasses Training for First Time
Daily token calls exceed 175 trillion; model iteration shrinks to weeks
First Embodied-AI Dataset Quality Standard Takes Effect; Xiaomi Model Wins Award
Eight-dimension framework shifts focus from scale to quality
Wuzhen Summit Opens on Open Source; DeepSeek Open-Sources Inference Stack
China's open-source ecosystem and domestic compute hit simultaneous milestones
AI Agents Enter Trust Era as HKMA Launches KYA, Goldman Tightens AI Use
Autonomy drives identity governance; research agents reach high-value tasks
Apple Integrates Alibaba's Qwen into Apple Intelligence in Landmark Deal
Qwen becomes system-level AI backbone for China; beta opens Nov 4
MiniMax H3 Tops SuperCLUE Video Ranking as Domestic Models Match Overseas
H3 leads I2V benchmark; domestic video models catch up on application
DeepSeek Open-Sources Super-agent: From 'Single Session' to 'Resident System'
DeepSeek releases Super-agent prototype and code
Consumer-Grade Embodied Robots Accelerate: From Factory to Home
Embodied AI enters homes as volume nears inflection
AgiBot Innovation Day: Embodied AI Enters a 'Measured' Delivery Phase
Worken metric and GO-3 Preview model unveiled
Domestic AI Compute Chips Accelerate: Moore Threads Turns Profit, Huixi Files IPO
Moore Threads Q3 non-GAAP profit turns positive; embodied-chip firm eyes IPO
AGIBOT unveils GO-3 embodied model; Lingqu OS open-sourced
AGIBOT releases GO-3 foundation model and open-sources OS
Google launches Gemini 4 Carbon; safety becomes battleground
Gemini 4 Carbon debuts as model safety race intensifies
NVIDIA Upgrades Isaac Robotics Platform as Physical AI Sees Dense Catalysts
GTC Berlin launches Isaac Lab 3.0 and open runtime; Agile ONE SN humanoid debuts.
Arm China's Gitee Community Unveils Jishu Embodied-AI Kits with Open-Source AIOS
Gitee launches JishuFriends robot dog and JishuBuddy tool; on-device AI accelerates.
NVIDIA Jetson Thor Ships in Volume: A Compute Base for Embodied AI
Jetson Thor ships to robot developers; 128 GB memory runs 70B models on-device
Moonshot Releases Kimi K3.1 Pro: 3T-Parameter Open Model Ranks Global Top Four
Kimi K3.1 Pro is the largest open model by parameters; ranks 4th globally
iFLYTEK Unveils Spark X2.6, a Fully Domestic-Compute End-Cloud Model
293B-A30B MoE on domestic 10k-GPU cluster; 4–4.5x efficiency gains
Moonshot Releases Kimi K3.1 Pro, the World's Largest Open-Weight Model
2.8T-parameter open weights intensify the price-performance race
iFlytek Launches Spark X2.6, Advancing Full-Stack Domestic Compute
Spark X2.6 trained on domestic compute; 1M-token context
China's Open-Source AI Models Top 10 Billion Global Downloads
NDRC: open-source models 10bn+ downloads; humanoids 90% of global shipments
Domestic Compute Seeks Breakthrough as Open-Source RISC-V V100 Debuts
6th RISC-V China Summit; domestic server-grade V100 chip unveiled
Domestic LLMs Refresh as Fully-Local Compute and Device-Cloud Synergy Lead
iFLYTEK Spark X2.6 and MiniMax M3.1 Flash open-source; local compute accelerates
iFlytek Unveils Spark X2.6 on Fully Domestic Compute with Edge-Cloud Split
iFlytek debuts Spark X2.6; edge-cloud on domestic compute takes center stage
Embodied Intelligence Moves to Standardization as EIBench and Genie Studio 2.0 Debut
Embodied-AI benchmark and full-lifecycle platform launched, speeding industrialization
RISC-V China Summit: Domestic Server Chip V100 Signals Open-Source Compute Push
Shenjilian and CAS unveil server-spec RISC-V chip V100
World Physical AI Conference: Embodied Intelligence's ChatGPT Moment Seen Within a Year
Unitree's Wang sees one-to-three-year window for physical AI
Huang's GTC Berlin Keynote: The Era of Personal AI Agents and AI Factories
RTX Spark and Vera Rubin redefine computing
Embodied-AI World Models Break Through: Anyverse Open-Sources MWA-Phoenix
Open base model tops sim-to-real benchmarks
D-Robotics' Sunrise S600 Packs 560 TOPS to Anchor Embodied-AI 'Brain'
Flagship robot chip lands orders with UBTECH, sold to 60+ countries
DeepSeek Eyes Winverse AI Lead as Capital Pours Along Robotics+LLM Chain
Liang Wengfeng bets on robot 'brain'; domestic AI chips scale up
Domestic AI Chips Debut En Masse: RISC-V and Full Local Supply Chains
Kunlunxin M100, Xuantie E908A, Everest 200 highlight autonomy push
Embodied Intelligence Moves From Demo to Deployment
MIIT sees 100k humanoids this year; OpenAI builds the 'brain'
DeepSeek V4.1-Flash-1017 Ships, Marking China's Task-Aware Compression Phase
DeepSeek opens search API; cost-per-task drops 40–50x
Edge AI and Open Models Race Ahead, Loosening the Compute Power Structure
Meta's 816B open model runs locally; Nvidia and Apple enter edge AI
Domestic AI Chips Debut En Masse at CIIF: Edge, Training and Dense GPU Clusters
AX8860, DF1000 and Tiangai 300IXL144 highlight China's compute stack
Global AI Governance Tightens as Microsoft and Gates Urge Stronger Oversight
Microsoft, Gates call for AI regulation; UK mulls mandatory safety reports
2026 AI Industry Conference Opens in Jinan as Embodied Intelligence Scales Up
National AI summit leaves Beijing; humanoids and embodied AI take stage.
Industrial Fair Watch: Manufacturing's 'Autonomous Agent' Era as Compute Chips and Embodied Robots Race
Industrial AI moves from pilots to scale; edge-compute bottleneck shows.
Google Gemini 4 Argon Hits 1M-Token Context; Microsoft's Maia 300 Aims at Inference Crown
Gemini 4 Argon's 1M context and Maia 300 chip reshape models and inference.
Embodied AI Starts 'Working the Line'; Edge LLM Chips Near Mass Production
Humanoids deliver factory KPIs; edge LLM chips ready; embodied AI turns commercial.
Microsoft Unveils Maia 300 AI Chip; Hyperscaler Silicon Reshapes Inference Supply
Maia 300 targets inference, 40% more tokens per megawatt.
On-Device LLM Chips Head to Mass Production; Embodied AI Moves From Demo to Shipment
On-device LLM chips mass-produce; embodied AI ships at volume.
OpenAI DevDay Wrap: GPT-6.1 and Decision Models Reshape Agent Cost Structure
DevDay ships GPT-6.1 and Decisions API; layered Agent inference cuts cost.
Open-Source and Domestic Models Advance: Meta Muse Wonder, Step 5 Break Through
Meta open-sources Muse Wonder; Step 5 joins ecosystems — open & domestic accelerate.
Microsoft Unveils Maia 300: In-House AI Silicon Scales to Gigawatts
Microsoft's third-gen chip claims 40% more tokens per megawatt than GPU systems
AI Agents Reshape Compute Demand: Supply Gap Widens to Structural Shortage
Agent scale-up lifts compute demand; global satisfaction rate to fall to 71% by 2027
National-Day AI Watch: Agents Get to Work, Robots Enter Real Scenes
Domestic models and compute catch up; embodied AI validated
'AI+' Action Fully Launched; World's First Open Trillion-Param Model Released
CPC-State Council opinion lands; 70+ embodied training sites
NVIDIA Launches Isaac ROS 5.0, Bringing AI Agents to 1.3M ROS Developers
Open physical-AI libraries expand; full Jetson edge support
OpenAI Open-Sources Codex Terminal Agent, Reshaping the AI Coding Paradigm
Rust-rewritten lightweight agent hits 104K+ GitHub stars
Edge Embodied Robots Work Offline as AI Compute Moves Onboard
On-device embodied AI breaks through; robots act offline under 10ms
Embodied AI's Commercial Test: 20,000th Unit, 300 Robots at Chimelong
Embodied AI staffs parks and factories; scale delivery meets real crowds
Beijing Yizhuang's Embodied-AI Chain Takes Shape
6,000 sqm data base trains 40 robot types to act autonomously
China Generative AI Users Top 700 Million
Over 50% penetration; domestic 100k-GPU cluster online
Embodied AI Moves from Demo to Deployment
CIFTIS and WAIC show robots at work; precision to 0.2 micron
Domestic Compute Base Accelerates: Ascend 950 and Neuromorphic Chips
Atlas 950 hits 1 EFLOPS FP8; neuromorphic chips fill gaps
Domestic LLMs Go Global: Kimi K3 Lands on Amazon Bedrock via Revenue Share
Kimi K3 joins Amazon Bedrock; first revenue-share model export to global clouds
Embodied Intelligence Industrializes: Edge Compute Chips and Robot Foundation Models Land
Huawei launches Ascend 960 supernode; Unitree open-sources robot foundation model
September's flagship model wave: AI shifts from answering questions to completing work
Frontier labs ship new flagships; the race turns agentic.
Humanoids enter their mass-production year as embodied-AI supply chains scale up
H1 funding tops ¥90B as humanoid manufacturing scales.
NVIDIA's Vera Rubin compute platform enters production ramp as AI factories go rack-scale
Vera Rubin ramps up; token cost drops ~35× vs Blackwell.
AI giants rarely align on slowing down; model training enters the 100k-GPU era
Compute arms race shifts to efficiency and cost.
Huawei launches world's first 3D data-center architecture for 100k-card Ascend nodes
Electro-mechanical delivery cut to 3 months.
DeepSeek releases V4.1-Flash, outperforming GPT-5.6 Sol
Cache-hit price drops to 0.006 dollars per million tokens.
AI compute race heats up; Oracle cloud revenue jumps 62%
RPO orders reach 664 billion dollars.
Enflame debuts on STAR Market, opens up 188%
Tencent paper gain exceeds 30 billion yuan.
Citic Securities: humanoid-robot catalysts keep coming; AIDC, semis and lithium equipment resonate
Thematic-investment heat continues.
Brain-computer interface clears clinical trial; invasive approach hits key milestone
Human–machine interaction may reshape rehab and interaction scenarios.
Domestic EDA breaks through at advanced nodes; foundry toolchain gains autonomy
Upstream semiconductor gap filled; supply-chain resilience improves.
Note: Articles under "Industry" and "Finance" are informational and technical commentary only, and do not constitute investment advice or an offer.