NeuroAI

Beijing Rossum's intelligent orthopedic robot cleared China's NMPA for spine and long-bone work after pelvic-fracture use in 20+ hospitals.

KEENON's 'general + specialized' service-robot playbook is moving from trade-show demos into real hotel, restaurant and cleaning shifts across 70+ countries.

Hikrobot, Hikvision's mobile-robot arm, crossed 200,000 units off the line as China's AMR market scales fast.

Da'ai Robot's lower-limb rehab exoskeleton has fallen from roughly ¥1M to consumer pricing after a new NMPA-registered model.

Zhejiang Province's effective support measures aim for over ¥1 trillion in core AI revenue by 2027, plus a ¥10-billion fund cluster — a provincial blueprint, not a national law.

MIIT's December 2025 action plan pushes AI deeper into factory floors, setting hard 2028 targets for industrial-network upgrades and shared industrial datasets.

The National Data Administration's draft Data Property Rights Registration Guidelines closed its comment window in April 2026; the rules are not yet in force and will be trialed first.

Shanghai AI Laboratory director Zhou Bowen released the 'Shusheng·Duanyan' platform at WAIC 2026, betting that scientific discovery — not chat — is the real test of frontier AI.

Founder Yin Zhiyao has grown AMEC from a single plasma etcher into a broad semiconductor-equipment platform — a quiet but central figure in China's push for chip-tool self-reliance.

Under founding director Huang Tiejun, the Beijing Academy of Artificial Intelligence persisted with a unified autoregressive route and landed Emu3 in Nature — reported as a first for a Chinese-led large-model paper.

InternLM3-8B-Instruct from Shanghai AI Laboratory was trained on 4 trillion tokens and the lab says it cut training cost by over 75% versus comparable models, while giving the weights away under Apache 2.0.

OpenAI's gpt-oss-120b and gpt-oss-20b, released under Apache 2.0 in August 2025, are open-weight reasoning models that run locally and compete directly with China's open large models (大模型) on developer machines.

Baidu's ERNIE-Image is an open-weight, single-stream DiT text-to-image model with 8B parameters that the company says leads among open_weights models and runs on a 24GB consumer card.

NetEase Youdao's open-source Confucius4-R2T2 real-time speech recognition model topped Hugging Face's ASR trending chart, beating OpenAI's Whisper.

Chinese banks have begun pricing credit off token consumption and compute contracts, because AI borrowers own almost nothing a branch can pledge.

A State Council General Office opinion puts a labour test inside AI project review and tells agencies to weigh the pace of adoption in labour-intensive industries.

Trip.com's TripGenie assistant drove a roughly 400% year-on-year jump in AI-driven orders in Q2 2026, turning a chatbot into a booking agent.

Meitu's H1 2026 results show AI-driven subscription and compute-point revenue turning a consumer app maker into a profitable AI-software company.

How BYD's decision to make its 'God's Eye' advanced driver-assistance system standard across 21 models at no extra cost reshaped China's EV competition.

TSMC's N2 process is now in high-volume production and its A16 backside-power node is set to follow, reshaping how AI accelerators are built.

NVIDIA confirmed Samsung, SK hynix and Micron are all qualified and shipping HBM4 for Vera Rubin, ending months of supply-chain speculation.

MediaTek will use NVIDIA's NVLink Fusion to help hyperscalers design custom XPUs, backed by a US$3.5B NVIDIA investment.

GigaDevice (兆易创新) completed a Hong Kong IPO in January 2026 and is pitching Flash, MCU and specialty DRAM for on-device AI.

A Citi banker told Reuters Chinese AI supply-chain IPOs should stay resilient; Hong Kong hosted 118 listings in the first nine months, raising HK$388 billion, up 106% year on year.

The Hangzhou lab is wrapping up a roughly 100 billion yuan (≈US$15B) funding round at a 500 billion yuan pre-money valuation, more than doubling its original target.

The Beijing lab founded by Kai-Fu Lee is raising a pre-IPO round and unwinding its offshore structure as it targets a 2027 Hong Kong debut.

WizMind (微灵医疗) builds ultra-thin flexible electrodes that help Chinese neurosurgeons locate speech and movement areas in real time during operations.

Since early 2024 Neuralink has implanted its N1 'Link' in a small group of paralyzed volunteers; the real question is whether thought-controlled computing holds up outside the lab.

Spain's INBRAIN Neuroelectronics implanted the first graphene-based brain-computer interface in a person in 2024, betting on material science over metal electrodes.

Two Nature papers in 2023 showed brain-computer interfaces decoding attempted speech into text and voice far faster and more accurately than before — a quiet leap for assistive communication.

TAL Education's MathGPT (九章大模型), the first state-filed education large model (大模型), now powers adaptive tutoring inside Xueersi devices used by students across China.

CMB's AI Xiao Zhao assistant and the open-sourced Yi Zhao (一招) financial large model (大模型) show how a retail bank is rebuilding customer service around AI.

A medical large model (医疗大模型) deployed across more than 2,000 Zhejiang clinics turns triage, companionship and report reading into a shared public service.

The 2026 Ningbo Youth AI Film Festival closed on October 25 with entries from 30 countries and 98 universities, the latest sign that AI-native filmmaking is organizing itself into a real community.

China's largest AI image-and-video creation community closed a record $130M Series B in October 2025, betting that the winner in generative media is the ecosystem, not the model.

Huawei Cloud's MetaStudio digital-content production line packages AIGC video, multilingual dubbing, and digital-human hosting as enterprise infrastructure, with short-video production cut from days to minutes.

A Beijing Internet Court decision that an image created with AI, reflecting human intellectual input, is protected work, and the copyright belongs to the user, drew sustained international attention through 2025.

How Tsinghua-trained researcher Yang Zhilin built Moonshot AI on long context and then pivoted Kimi to open-weight models.

How a 1990-born engineer built Unitree from a one-man workshop into a humanoid-robot maker that lists on the STAR Market.

Unitree (宇树科技) closed a RMB 694m (≈US$97m) C-round in June 2025 at a ~US$1.8bn valuation, with China's tech giants piling in ahead of an IPO.

Tencent reported H1 2025 revenue of 364.5 billion yuan (≈US$50.3B) with net profit up 16%, even as AI capital expenditure more than doubled.

China's largest foundry is mass-producing 7nm-class chips with only DUV tools, becoming the bottleneck — and the backbone — of domestic AI compute.

State Grid's 2026 embodied-intelligence plan calls for about 8,500 robots worth ~¥6.8B, led by 5,000 quadruped inspection dogs for substations and power lines.

Pudu Robotics and Shenzhen CTID will open the world's first full-scenario robot-serviced hotel on the Shenzhen-Zhongshan Link, with trial operation by end of 2026.

Lingyi Precision's Beijing embodied-intelligence super-factory rolled out its first humanoid robots in April, targeting 10,000 units this year and 500,000 by 2030.

Edge Medical's integrated multi-port, single-port and remote surgical platform won NMPA approval, billed as the first unified system cleared in China.

Alibaba's Qwen3 family pairs thinking and non-thinking modes across two MoE and six dense models, all under Apache 2.0.

Alibaba's Qwen team open-sourced Qwen-Image (通义万相), a 20B image model under Apache-2.0, with standout Chinese/English text rendering and editing.

SAMR published three guidance standards (GB/Z 224, 231, 241—2026) for AI in steel — scenarios, datasets, maturity grading. They are recommended, not mandatory.

Jiangsu's five departments launched token, corpus and model vouchers to cut AI costs; a single firm can claim up to ¥5M (≈US$700k) in compute aid through mid-2027.

Beijing reopened its 2000 internet rulebook for comment, adding a first-ever 'smart information service' chapter for AI. It is a draft, not in force.

CAC Order No. 18, effective May 1, 2025, requires large personal-information handlers to audit every two years and lets regulators trigger audits after breaches.

Michael Mi's GL Ventures closed a dual-currency fund above 5 billion yuan in April 2026—the year's biggest China VC raise—backing AI names like Zhipu and Unitree.

Lin Junyang left Alibaba's Qwen team in March and, five months later, launched Pragmatik Labs at a reported $2B valuation—no product yet, only agents.

Disheng Xu's team at Greater Bay University used Shanghai AI Lab's Shusheng-S2 model to prove the 3D Ghys conjecture, a problem open since 1995.

How Huawei 'Genius Youth' Peng Zhihui built AgiBot around open hardware, a foundation model and a robot data factory.

Shanghai's NeuroXess used a 256-channel flexible implant to let an epilepsy patient play AAA games and decode Chinese speech in real time.

China's NMPA cleared Neuracle's NEO, the first invasive BCI medical device approved anywhere, letting spinal-injury patients move a glove by thought.

State Council Order No. 790, in force since Jan 1, 2025, governs network data handling and explicitly pulls generative AI training data under security duties.

The RMB 60.06bn (≈US$8.4bn) National AI Industry Investment Fund, set up in January 2025, has begun backing DeepSeek, Kimi, Kling AI and a GPU designer.