Companies
Alibaba
Ant Group
Baidu
ByteDance
Cambricon
DeepSeek
Huawei
iFlyTek
MiniMax
Moonshot AI
Pony.ai
StepFun
Tencent
Newsfeed · Live
Sep 8, 2026
Compare
2022
2027
DeepSeek
$7.4B
Moonshot AI
$8.2B
Zhipu AI (Z.ai)
$5.8B
CUMULATIVE RAISED · FEATURED
Products
Doubao (豆包)
344.9M MAU
Xiaohongshu (小红书)
300M+ MAU
Qwen (千问)
165.6M MAU
DeepSeek
127.1M MAU
Tencent Yuanbao (元宝)
57.4M MAU
NEWSFEED · LIVE
SYNCED 03:15 AM
SEP 8, 2026
1 story
In pictures: from factory floors to AI – the evolution of innovation in Hong Kong
SEP 7, 2026
14 stories
China’s Xiaomi unveils folding phone with home-grown chip as it takes on Apple, Huawei
China vows deeper cooperation with Qatar in energy, AI and Middle East de-escalation
ByteDance Founder Joins AI Elite in Race to Perfect World Models
Why Tencent is swapping Bilibili equity for debt, and what AI has to do with it
China hits critical Japanese chipmaking chemical with steep anti-dumping measures
CXMT Highlights Smartphone, AI Advances as Market Share Climbs
BYD’s Profit Rebound Has a Price
IQE CEO Warns of Emerging China Supply Risk for Chip Industry
China’s Moonshot and Z.ai bring AI model subscription race to Tmall’s retail shelves
Chinese shoppers can now buy AI subscriptions from DeepSeek and Z.ai on Tmall
Malaysia Eyes Huawei Chips for AI Project Despite US Warning
China won’t permit AI dominance; South Korea’s deepfake crisis: 5 weekend reads you missed
Breakingviews - COMMENTARY: China's AI dragons breathe fire on Nvidia's moat
How generative AI helps SenseTime turn a profit even as Chinese peers struggle
SEP 6, 2026
5 stories
🗞️ByteDance's $30B Loan, Moonshot's IPO Filing, and China's Hottest AI Chip Debut Yet
Punching above their weight: how China’s AI giants stretch each dollar in compute race
Why China’s AI giants spend far less and still shrink compute gap with US rivals
Why China is being urged to study Ukraine’s wartime PR blitz in the AI age
China will never allow any country in this world to achieve AI dominance: Victor Gao
This Week
Items77
Companies15
Rounds18
Last sync03:15 AM
Most Active · This Week
ByteDance
6
Moonshot AI
5
MiniMax
5
Enflame Technology
5
Huawei
Huawei is a Shenzhen telecom-and-electronics giant now a full-stack AI infrastructure player — its Ascend chips, Pangu foundation models, Huawei Cloud (CloudMatrix), and HarmonyOS AI form a domestic alternative to Nvidia/Western AI stacks. Wholly owned by its employees and not listed on any exchange, it is a central player in China's domestic AI computing buildout.
FY2025 Revenue
¥880.9B
R&D Investment
¥192.3B
openPangu-2.0-Pro
Text
Open
505B-param MoE (18B active), open-sourced with weights, inference code & technical report; trained 100% on Huawei Ascend 910B NPUs
AIME 202695.4/100
GPQADiamond87.9/100
BrowseComp65.7/100
Context512K
Input
Output
View Docs ↗
Pangu Ultra MoE (5.5)
Text
Closed
718B-parameter MoE model (256 experts), trained on the full-stack hardware and software of Huawei Cloud's AI Cloud Service (CloudMatrix 384 supernodes). Part of the Pangu Models 5.5 family, which per Huawei Cloud's own release delivers upgrades across five capabilities: natural language processing, computer vision, multimodal, prediction, and scientific computing.
Context—
Input
Output
View Docs ↗
FY2025 Revenue
¥880.9B
net profit ¥68B
R&D Investment
¥192.3B
21.8% of FY2025 revenue
早报｜华为Mate XT 2首发「韬定律」麒麟芯片/20.99万起，小米澎程上市/「豆包手机」定档下周三
China’s Xiaomi unveils folding phone with home-grown chip as it takes on Apple, Huawei
Huawei Hongmeng 7 conference is the most comprehensive summary: three folds, wide and straight boards are all here, and the information density is dazzling.
Malaysia Eyes Huawei Chips for AI Project Despite US Warning
Yu Minhong: I have been particularly uneasy in recent years and have made a reputation. It is said that DeepSeek will purchase 160,000 Huawei AI chips. Tesla driverless taxis will operate 24 hours a day, 24 hours a day, Lei Feng Morning Post.
DeepSeek Plans Big Huawei AI Chip Order to Power New Data Center
Yushu robot dog exposed production chain: 421Wh battery fake label changed to 97Wh, "smuggled" on the plane; rare! Xiaomi Huawei released the time slot, netizens: Immortal fight; the dark side of the moon launches the IPO of Hong Kong stocks
The first day of Apple's new CEO! Exposed the new iPhone to resume giving away charging heads, the official response; Yushu responded that "over 100 yuan reimbursement needs Wang Xingxing's approval" and other rumors; Huawei, Xiaomi, Honor mobile phone collective price increase
Morning News | Dyson Releases $499 AI Toothbrush with Camera/Huawei, Xiaomi, Honor Collective Price Increase/Strongest Model Fable 5.1 Released
T Morning Report | Nvidia invests $3.5 billion in MediaTek; Fluorogen Science and Technology Innovation Board is priced at 142.18 yuan/share raised 6.1 billion yuan; Huawei's revenue in the first half of the year was 467.8 billion yuan, 1/4 of which was invested in R&D.
Ruijin Hospital and Huawei Cloud released RuiPath 2.0, a large model of Ruizhi pathology
Paradigm and Huawei reached a heavyweight computing power strategic cooperation, becoming the first AI companies to embrace the domestic high-end computing power base.
From Data to Agent: Huawei Cloud AI-Ready Data Infrastructure Full Link Capability Debuts at Digital Expo
Huawei’s Chip, Data Center Bid Takes US-China AI Rivalry to Egypt
I went to China to see a different AI future. It looked familiar
Rest of World's publisher recounts a delegation trip through Beijing and Shanghai with media, tech, policy and finance leaders, finding China's AI strategy centered on embodying AI into robotics, EVs and manufacturing rather than chasing frontier models alone, with Huawei, Xiaomi and DeepSeek cited as evidence of a deployment-first approach shaped partly by US chip export controls.
Huawei Pitches Egypt With AI Chips, Data Center Buildout
Huawei Courts Egypt With AI Chips in Test of US Tech Diplomacy
Morning Post | iPhone 18 Pro is coming, Foxconn offers 8,800 yuan recruitment bonus/WeChat gray test AI photo retouching/Huawei launches wide straight phone
Ray7 is the first in the world to integrate Huawei DriveONE chip-level drive, leading the industry in pure electric chassis technology
Alipay released the AHA protocol system for cross-terminal interconnection of smartphones, co-constructed with more than 20 companies including Qianwen, Huawei, OPPO, vivo, and Honor.
Wenjie’s “baby stroller” is on the market, co-designed by Huawei; DeepSeek’s price increase strategy is implemented today; a college student was sentenced for using AI face video to steal 50,000 yuan
Zhongjian Technology received a regulatory fine for violating Huawei's embodied smart hotspot letter | Quick read announcement
⚔️Out of the Trough: How Huawei's Ascend Chips Climbed Back
Huawei IJCAI 2026 Paper Inventory: From "Scale Density" to "Design Density"
DeepSeek is rumored to restart financing; Huawei Yu Chengdong: mobile phones may have to increase prices on a large scale in the future; the first criminal case for destroying an AI model was pronounced, and the programmer was sentenced for "deleting the library and running away"
China's A.I. Is Surging Across Africa
African developers are increasingly choosing cheap, open-source Chinese AI models over U.S. systems, citing better performance on local languages, lower cost, and free customization; Alibaba's Qwen and Huawei-backed DeepSeek cloud packages are leading adoption across tech hubs from Uganda to Nigeria.
AI no longer forgets after use: Huawei Noah open source MindMemOS, memory and skills evolve together
A U.S. committee delegation came to China to inspect Huawei, DeepSeek, etc.: the results were rejected; LV responded to the Jasmine Milk White trademark lawsuit: intellectual property is an absolute core asset; Apple’s market value exceeded US$5 trillion
Starting from 169,800 yuan, Huawei's Qiankun six-piece set is available on the car, and the Yipai M8 squeezes into the crowded six-seater market
Huawei Pangu started a business at No. 2, and its valuation increased 10 times in three months
Exclusive丨Huawei’s 22-year veteran pushes back on LPU from the end of AI, Yuanchuan Micro completes a new round of hundreds of millions of yuan in financing
Hard Krypton Debut | Former Huawei Intelligent Driving Keywords Enter AI Shipbuilding, Already Received an Order of 100 Million Yuan, Specializing in Leisure Boats
After installing millions of vehicles, how does Huawei Qiankun move towards the next generation of intelligent driving?
Morning Post | Huawei’s mobile phone share rose to 23% in the second quarter/Guangzhou Automobile Honda and Honda renewed their joint venture agreement until 2038/First trailer for "Avengers 5"
漫威影业昨天发布《复仇者联盟 5：毁灭之日》首支正式预告，影片计划于 2026 年 12 月全球上映。预告片时长约 2 分 25 秒，聚焦即将到来的「毁灭之日」以及来自不同阵营的英雄集结。
影片将围绕复仇者联盟对抗这一反派展开，故事涉及多元宇宙危机。预告中还出现了神奇四侠相关角色，以及 X 战警、洛基等角色的相关画面或线索。
法院要求派拉蒙暂停 1100 亿美元收购华纳兄弟
据路透社报道，美国联邦法官周一裁定，派拉蒙天舞对华纳兄弟探索的 1100 亿美元收购必须暂停至 8 月 3 日。加利福尼亚州与纽约、科罗拉多等 11 个州于 7 月 13 日起诉，认为合并会削弱电影和电视市场竞争。
法官认为，各州已经有力证明交易可能违法；如果先完成合并，裁员、分享商业敏感信息等改变很难事后撤销。各州称，合并后公司可能占广泛发行电影市场约 27%。
法院将于 8 月 3 日再审理是否在整个诉讼期间延续暂停。按交易协议，如果交割延迟到 9 月 30 日之后，派拉蒙 CEO 大卫·埃里森每个日历日需向华纳兄弟股东支付每股 0.25 美元，合计约 700 万美元。
Netflix 为 200 小时儿童内容增加美国手语演绎
据 IndieWire 报道，Netflix 与无障碍技术公司 SignUp Media 达成合作，为约 200 小时的儿童与家庭内容制作美国手语（ASL）演绎。
首批覆盖 41 个系列或 IP 的 73 部作品，包括《吉尔莫·德尔·托罗之匹诺曹》《海兽猎人》《宝可梦：地平线》《新神奇校车》《可可瓜儿歌街》和《智能大反攻》，相关内容已于昨日开始上线。
这些手语视频通过 SignUp 的 Chrome 扩展与 Netflix 正片同步播放。手语演绎窗口可以移动、调整大小、分屏或弹出，用户无需另找一个独立的「手语版」条目。SignUp 还计划把功能扩展到电视和移动设备，并加入预告片手语演绎。
#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。
]]>Huawei Qiankun is expected to invest more than 18 billion yuan in R&D in 2026
华为昇腾950超节点首次真机亮相 如何布局AI算力“系统时代”？丨行业风口
Krypton | China Unicom and Huawei launch the world’s largest 5G-A 100M uplink network
Domestic hashrate "Big Mac"! Huawei Ascend 950 Super Node Real Machine First Public
Huawei's Ascend 384 super node has over 750 commercial units, and the 950 super node "takes over"
36 Krypton First Release | Former core members of Weilai and Huawei Intelligent Driving teamed up to start an embodied world model and completed hundreds of millions of yuan in financing within three months
Morning Post | Realme suspends product updates/Huawei withdraws trademark application for "World"/Nubia's world's first AI smartphone equipped with Doubao mobile assistant
Huawei Jin Yuzhi: Large-scale commercial use of L3 and L4 is expected to be realized in 2027 and 2028
Shenzhen Huaqiang: Actively assisting Huawei in building a computing power ecosystem, and related work is being implemented as soon as possible
Huawei’s Ascend 950 super node will debut as a real machine for the first time, and the high prosperity of the domestic computing power sector is expected to continue
Samsung’s “Wide Fold” is coming, can it challenge Apple and Huawei?
Paul Triolo on Chinese labs makings chips, SMIC, Huawei and the importance of AI governance
Qijing GT7 officially starts nationwide delivery, Huawei Qiankun ADS 5 is delivered with the vehicle
Former Huawei "genius boy" interviewed DeepSeek and went berserk on the spot when he was questioned about copying code_Wonderful video presented to you_36氪
Microsoft's layoff compensation plan exposed: you can get up to 10 months of basic salary as severance pay; Huawei's Automotive BU President talks about the "smart driving bubble": overheated publicity, real ability gap is huge; Byte performance rules adjustment
Morning Post | Nubia confirms the launch of the world's first AI smartphone/Ferrari executive: Chinese car companies are engaged in FMCG/Huawei's young genius responds to DeepSeek interview controversy
Behind the conflict between Huawei’s young genius and DeepSeek, what is Liang Wenfeng’s view on talent?
Huawei Hubble withdraws from Zhongke Jinghe
8:01 Krypton丨The former Huawei genius boy complained about the DeepSeek interview, saying that the interviewer questioned him about plagiarism; Apple folding screen is already in mass production; Trump called for the purchase of Dell computers, causing Dell to skyrocket by tens of billions of dollars
DeepSeek recruitment was publicly complained by "Huawei's genius boy", "I met the most unprofessional"
Huawei Atlas 950 real machine will be unveiled at WAIC soon. Will domestic computing power usher in the super node era?
Silicon-based liquidity with a valuation of 7.74 billion yuan hits the “first share of AI Token factory” Alibaba + Huawei + Meituan + Zhipu Betting
Former Huawei genius boy responds to DeepSeek interview controversy
The World Artificial Intelligence Conference opens next week! Huawei’s largest super-node real phone, the world’s first AI smartphone, and other debuts
Poison! It was revealed that Meta actually hired outsourcers to trap competitors on a large scale; Tencent significantly reduced its holdings in Kuaishou stock and is no longer a major shareholder; the former Huawei "genius boy" complained about the DeepSeek interview process
Morning Post｜Revealed that Huawei Mate90 will be equipped with "Tao's Law" new Kirin chip/Intel raises CPU price/Enterprise WeChat releases AI recording hardware
Former Huawei "genius boy" publicly complained about the DeepSeek interview process
Huawei Mate90 series is expected to be equipped with new Kirin chips based on Tao's Law
Morning Post | It is revealed that the release of the foldable screen iPhone will be delayed/Huawei He Tingbo releases the V2 version of "Tao's Law" paper/7-Eleven sues Nike for suspected plagiarism in the color matching of new shoes
Meituan Open-Sources LongCat-2.0, a 1.6T-Parameter Model Trained on Chinese Chips
Meituan open-sourced LongCat-2.0, a 1.6-trillion-parameter mixture-of-experts model with a native 1-million-token context window, under an MIT license. Meituan said the near-frontier agentic-coding model was trained and served entirely on a cluster of domestic AI chips — from vendors including Huawei, Moore Threads, and MetaX — with no Nvidia hardware.
Huawei updates Tao’s Law paper!
[Morning Post] Huawei releases V2 version of "Tao's Law" paper; Doubao and Qianwen: The intelligent agent function will be offline
Huawei's He Tingbo releases V2 version of "Tao's Law" paper, supplementing project details and measured data
First published on 36Kr | Cambridge associate professor started a silicon photonic chip, has cooperated with L'Oreal and Huawei, and received 100 million investment
Yijing joins hands with Huawei for global actual testing of Qiankun! CCTV's "Extraordinary Step" witnessed the "three major leaps" of Chinese automobiles
Huawei Technologies Co. increases capital to approximately 41.14 billion
Morning Post | It is revealed that Apple plans to produce 10 million folding iPhones/Nintendo responds to rumors of 10% salary increase for all employees/Huawei Hongmeng 6 terminal equipment exceeds 70 million
Heavy! Another AI "shovel seller" hits the IPO, Alibaba, Huawei, Hubble, etc. are betting on it
In Huawei’s “Moments”, Qijing is not the “background board”
Huawei Cloud: Plans to officially stop selling application platform-AppStage and other services on December 30, 2026
Huawei: openPangu-2.0-Flash model is officially open source and online
The chassis has also begun to "calculate": What exactly is Huawei's Touring platform solving?
Huawei Turing platform has completed three rounds of upgrades, covering the five sectors of Hongmeng Smart Travel
DeepSeek-V4 is here! Huawei’s Ascendant blessing is still the “price butcher”
Huawei helps global operators use the network as the foundation and intelligence as the wings to innovate collaboratively across industry, network and computing, and embark on a new journey of token management.
Huawei and Cambricon tighten grip on China’s AI chip market as Nvidia slips
Huawei and Cambricon lead China's AI server chip surge, piling pressure on Nvidia
Domestic chip suppliers led by Huawei and Cambricon are on track to raise their share of China's AI server chip market to 56% in 2026, up from 46% in 2025.
HKU MaRS Lab won the IEEE TRO Fu Jingsun Memorial Best Paper Award for the work of Huawei’s young genius
Huawei and others set up new technology service company in Shenzhen
Huawei Qiankun Smart Driving High-end Package Will Adjust Prices Starting from July
Huawei and Google are both grabbing it! Why are mobile phone widgets a must-have for AI? ｜AI Artifact History
nova 16 Ultra experience: Young people’s first flagship Huawei?
Huawei announces Wi-Fi 7 patent licensing rate, US$0.5 per device
Lakeside Dialogue 2026: Huawei’s “Eight-Year Long March” for Barrier-free
Huawei Xiaoyi: Hide the utensils in your body and wait for the time to move | AI Utensils Chronicle
In the Agent era, Huawei Cloud begins to rebuild the foundation
For a limited time, starting from 154,900 yuan, Huawei Qiankun Zhijia ADS Pro is available in the new dark blue S07 extended range version
ByteDance in Talks to Buy 50,000 AI Inference Chips From China's Iluvatar CoreX
ByteDance is negotiating to purchase at least 50,000 inference GPUs from Shanghai startup Iluvatar CoreX — potentially its third domestic GPU supplier after Huawei and Cambricon — as it scales its Doubao chatbot and reduces dependence on Nvidia amid US export controls.
Morning Post | SpaceX surged on its first day of listing/Apple executives: Siri does not want to be users’ emotional companions/Huawei’s Yu Chengdong: He wants to take the Pangu model from China’s No. 1 to the world’s No. 1
Huawei releases Hongmeng 7: In the AI era, chasers are overtaking
Huawei AI Glasses Experience: It’s easy to wear and love, let’s talk about AI
Huawei is considering deploying Ascend AI chips in Latin America, cloud chief says
Huawei's cloud chief said the company is considering deploying its Ascend AI chips in Latin America as part of Beijing's push for AI self-reliance.
Kweichow Moutai: Elect Chairman Chen Hua and Appoint Multiple Executives
Zhijie RX is officially announced as Huawei’s first supercar SUV
The difference in the proportion of loaded vehicles is only 0.08%! Guoxuan and Zhongchuang compete for the giant whale battery platform, who can reap Huawei’s ecological dividends?丨Observation of the Greater Bay Area Auto Show
Cyrus established a new brand, Saidou Technology, and introduced Doubao AI to start de-Huawei self-rescue
Why does Huawei want to build Hongmeng? Probably the most popular explanation at present
China Plans $295 Billion Investment to Build Nationwide AI Data Centers
China is preparing to spend roughly 2 trillion yuan ($295 billion) over five years on a nationwide network of interconnected computing hubs, relying on domestic suppliers like Huawei for at least 80% of AI chips.
Exclusive丨Huawei’s young genius Wang Yuxin started his own business and completed tens of millions of first-round financing in the first month
Looking at Huawei Cloud from the INSPIRE Creators Conference: It is finding a clearer position for itself in the AI era
In the Token War, Huawei Cloud chose the third path｜Forefront
Huawei Cloud releases new Agentic AI series products to create a "silicon-based black land" in the intelligent era
Huawei Cloud releases four major areas of "Industry AI Dream Factory"
Huawei Cloud releases new generation ModelArts Next model training and promotion platform
A former Huawei employee started a wire-controlled chassis, participated in the compilation of national standards, and received investment from Songhe and Su Gaoxin丨36 Krypton first release
WeChat opens up to smartphone AI assistants after ByteDance phone backlash
Tencent grants Huawei, Honor, Xiaomi, Oppo, and Vivo built-in voice assistants access to WeChat for natural-language messaging and calls, pushing into the AI phone sector while also exploring AI-enabled payments via its WorkBuddy agent.
Exclusive｜Huawei Frame Leap Technology completed angel round financing of 10 million U.S. dollars and will release video products Leadde
Krypton Evening News | OpenAI CFO talks about the company's AI equipment: it will be officially released before the end of this year; Tencent customer service responded to cooperation with Huawei, Xiaomi, etc.; Hong Kong launched the first productivity-level super intelligence
Tencent customer service responds to cooperation with Huawei, Xiaomi, etc.
The Tau Law: Huawei Wants to Change the Rule of the Chip Race
It's more like a methodology for surviving the near term, for keeping Huawei’s silicon competitive for as long as EUV stays out of reach
Huawei's AI talent entrepreneurship is heating up, and the former Noah's Ark Laboratory Wang Yunhe's company is valued at US$100 million
Exclusive | Former Huawei Terminal Cloud Large Model Technical Director and talented young man Li joined Jiyi Intelligent
Huawei Hubble invests in Wuxi Hongyin Technology Company
Huawei releases nova 16 series: 200 million main cameras, red maple images, and ten years of answers
Using technology and Kipchoge to jointly empower runners, Huawei WATCH GT Runner 2 track legendary model is released
Qijing GT7 opens for pre-sale: starting from 219,900 yuan, the "luxury content" is off the charts, and it smells of Huawei from the inside out
Huawei and Nanjing Univ. build 2D parallel chip Mengqi-1000, published in Nature Electronics
Researchers from Nanjing University and Huawei unveiled a world-first parallel processor using molybdenum disulfide 2D semiconductors, claiming record transistor density and a path beyond silicon Moore's Law limits
Another DeepSeek Moment'? Huawei Milestone Alters China Trajectory in Chip Race: Analysts
Analysts say Huawei's new Tau Scaling Law and LogicFolding architecture — targeting 1.4nm-equivalent chip density by 2031 — could reshape the global semiconductor race, with Nvidia cited as most exposed.
Peking University Unveils 3D Design Tool to Power Huawei's Chip Ambitions
A new three-dimensional chip design tool from Peking University is positioned to complement Huawei's LogicFolding architecture, giving China's domestic semiconductor roadmap an academic research leg up.
Huawei's New Chip Scaling Law Aims to Sidestep ASML Chokepoint, but Hurdles Remain: Analysts
Huawei's time-scaling approach offers a path around EUV lithography restrictions, but analysts warn that translating the architectural concept into volume production faces significant near-term obstacles.
Huawei unveils new scaling law and tech that narrows gap with TSMC, Samsung
Huawei introduced the Tau (τ) Scaling Law at ISCAS in Shanghai, claiming chips designed under it will achieve 1.4nm-equivalent transistor density by 2031; a LogicFolding architecture is already used in 381 mass-produced chips.
Token factory lands in Wuxi Relying on Huawei Ascend 384 cluster, computing power enters the era of "intelligent unit billing"
DeepSeek-V4: Huawei Ascend Adapter, Cost-effective King, Latest Underlying Technology
“中国DeepSeek与华为的崛起正威胁美国在技术、政治与军事领域的主导地位”
华为昇腾产业链解析_新浪新闻
DeepSeek V4适配华为昇腾_新浪新闻
NEWSLETTER: DeepSeek bets on Huawei as China pushes to end reliance on Nvidia
DeepSeek unveils new AI model tailored for Huawei chips as China pushes for tech autonomy
华为昇腾超节点支持DeepSeek V4模型，寒武纪也适配完成
8大国产AI芯片火速适配DeepSeek-V4！华为百度阿里全吻上来了
DeepSeek牵手华为，谁在背后悄悄数钱？
DeepSeek-V4 is late but here! Millions of contexts have become standard, and both Huawei Ascend and Nvidia have been included in its technical reports
DeepSeek-V4 released! Entering the era of millions of contexts, fully supported by Huawei Ascend Super Node
DeepSeek-V4 clearly supports Huawei Ascendant 950 chip
“训练为王”终结 ：AI推理新浪潮，英伟达谷歌华为切换赛道
告别英伟达！DeepSeek V4全面换装华为昇腾，国产AI芯片迎来真正的逆袭！
华为昇腾芯片将为DeepSeek-V4推理，通往国产算力自由
DeepSeek's V4 model will run on Huawei chips, The Information reports
EXCLUSIVE: Huawei's new AI chip finds favour with ByteDance, Alibaba which plan to place orders, sources say
Qianwen 3.5 detonated the global AI industry chain, and Nvidia, Huawei Ascend, AMD, Apple, etc. were the first to adapt.
华为云CloudMatrix AI Infra入选人工智能TOP案例
Zhipu AI breaks US chip reliance with first major model trained on Huawei stack
Why Nvidia chip is unlikely to derail ambitions of China’s Huawei, Moore Threads
当寒武纪、华为昇腾被同时适配，商汤科技在下一盘什么棋？
建数智赋能工程，创智慧油田未来-华为企业业务
Meet Tesla FSD's Fiercest Competitors in China
Huawei, Momenta, Li Auto, and XPeng are challenging Tesla's self-driving dominance through proprietary end-to-end systems and aggressive OEM partnerships with Chinese automakers.
DeepSeek新模型与国产芯片的“零日之舞”，寒武纪与华为昇腾“即时适配”彰显国产AI生态协同加速度
Huawei Ascend, Cambrian, and Haiguang information adapt to DeepSeek’s latest model
CITIC Securities: Huawei's Ascendant product group unveiled, accelerating the breakthrough of leading domestic computing power
从超节点到集群，华为亮出AI算力全家桶
基于华为昇腾训练 性能不输英伟达！自家模型爆料：DeepSeek-R2本月要发布
Treasure of the town hall is popular for punching cards! Huawei Ascend 384 Super Node Debut | 2025WAIC
How Huawei Trains DeepSeek-R1-Class LLMs Using Its Own Ascend Chips
Pangu Ultra MoE, a 718B-parameter model trained on 6,000+ Huawei Ascend NPUs, proves that frontier AI models can now be built in China entirely without Nvidia GPUs.
一文看懂华为昇腾芯片
US fine-grained ban on Huawei Ascend rule violators can be punished with up to 20 years in prison
T Morning Post | United States bans global use of Huawei Ascend chips; five departments interview takeaway platform companies; JD's revenue growth in the first quarter hit a new high in nearly three years
China’s iFlytek touts reasoning model trained entirely with Huawei’s AI chips
Similar to H100! Silicon-based mobile x Huawei Cloud launched the first DeepSeek-R1 based on the Ascend CloudMatrix supernode
Led by Huawei's talented teenagers, Zhiyuan Robot Company releases the world's first large model of universal embodied base, GO-1
[Theme for Tomorrow] Huawei Ascend Comprehensively Accelerates DeepSeek Large Model Practice in Shengjing Hospital
DeepSeek AI models debut on platform powered by Huawei
First release! Silicon Mobile x Huawei Cloud jointly launches DeepSeek R1&V3 inference service based on Ascend Cloud!
Chinese AI firm iFlyTek says its LLMs are trained completely on Huawei platform
拆解华为盘古大模型：与ChatGPT有何不同？
计算机行业专题报告：华为昇腾，国产AI算力“扛旗者”
华为盘古大模型 3.0 发布，友商们该急起来了？
just! Huawei Pangu large model updated information, but you may be disappointed
The "Android Era" of large models begins! Huawei’s new version of Pangu is about to debut. B-end application capabilities may exceed GPT-3
Huawei’s “Pangu” is expected to become the most powerful ChatGPT application in China! The first multi-modal 100 billion AI large model will be launched soon. List of listed companies in the industry chain
ChatGPT detonates computing power demand! Huawei Ascend leads AI supercomputing, these listed companies are deploying related businesses
华为昇腾加速落地：今年启动建设超20个计算中心 助力人工智能发展
The World Artificial Intelligence Conference is coming: Huawei's "Ascend" stocks are soaring, and you can understand the potential investment panorama in one picture
The other side of Huawei’s rise
How does Cambrian compete with Nvidia and Huawei HiSilicon?
全球最强！华为昇腾 910 AI 芯片正式商用，全场景 AI 框架 MindSpore 问世
Ren Zhengfei
任正非
Founder
AI Chips
View full chip database ↗
🔒 Full AI Chips Specs — Paid Subscribers Only
See the free preview on the full chip database ↗, or enter the email on your paid Recode China AI subscription to unlock everything here.
Reference Table
All Models
Compare models across all 43 Chinese AI companies
Company ↕
Model ↕
Type ↕
AA Score ↕
Released ▼
Context ↕
Input ↕
Output ↕
Tencent
Hy4-preview
Open
770B total/49B active MoE flagship with Gated DeepSeek Sparse Attention; beat GLM-5.3 and Kimi K3 in Tencent's internal blind eval; pitched as open-source coding/agent frontier
Text
—
2026-08-27
1M
$0.83/1M tokens
$2.5/1M tokens
Alibaba
Qwen-3.8-Next
Open
180B-parameter/ 6B parameters activated, A GDN + QSA hybrid architecture, muon optimizer adopted, early preview of the architecture used in Qwen4
Multimodal
—
2026-08-26
262K
$0.16/1M tokens
$0.47/1M tokens
Zhipu AI (Z.ai)
GLM-5.3-Flash
Open
Zhipu's first natively multimodal GLM-5 model, 320B total/18B active, KDA+MLA hybrid attention, mHC adopted, tested anonymously as ox-alpha on OpenCode and OpenRouter
Multimodal
57/100
2026-08-26
1M
$0.075/1M tokens
$0.25/1M tokens
DeepSeek
DeepSeek-V4-Flash-Vision-Exp
Closed
Experimental multimodal (vision) variant of V4-Flash; positioned near Claude Opus 4.8 on agent benchmarks; API-only
Multimodal
—
2026-08-21
1M
$0.44/1M tokens
$1.32/1M tokens
SenseTimeOpen-sourced 8B unified any-to-any model with native 4K image generation and improved editing
Multimodal
—
2026-08-21
32K
—
—
Alibaba
Qwen3.8-27B
Open
27.78B dense model, Apache 2.0, text/image/video input, agent/coding gains over Qwen3.6-27B
Multimodal
52/100
2026-08-14
262K
$0.40/1M tokens
$3/1M tokens
Zhipu AI (Z.ai)
GLM-5.3
Open
Post-trained on the same 744B GLM-5.2 base; built for agentic coding & cyber defense; open weights/API pricing to follow after safety review
Text
60/100
2026-08-14
1M
$1.40/1M tokens
$4.40/1M tokens
MiniMax
MiniMax-Music3
Open
Open-weight text-to-music model; generates full songs with sung vocals (up to 5 min) from lyrics + structured caption
Multimodal
—
2026-08-14
—
$0.15 per track
—
DeepSeek
DeepSeek-V4-Pro-0813
Open
GA release of flagship (was in preview since April); 1.6T total/49B active MoE, agentic gains
Text
53/100
2026-08-13
1M
$0.435/1M tokens
$0.87/1M tokens
Alibaba
Wan2.2-Animate-2-14B
Open
Faster/distilled 14B character-animation & replacement model from Tongyi Lab, Apache 2.0
Video Gen
2026-08-07
—
—
—
Alibaba
Wan 3.0
Closed
Public beta all-in-one video gen/edit model; single continuous shot up to 30s, director-level camera moves, character/scene consistency, and doc/xls/ppt/pdf/md-to-video capability; no open weights (unlike Wan 2.2)
Video Gen
2026-08-06
—
—
—
ByteDance
SeedRealtime
Closed
Native audio-visual full-duplex LLM fusing audio/video/text streams for real-time watch, listen, speak interaction; replaces separate ASR/vision/TTS pipeline; live in Doubao app
Multimodal
—
2026-08-05
—
—
—
Alibaba
Qwen-Image-3.0-Pro
Closed
Third-gen image model with ~4.5K-token ultra-long prompts, ~10px fine text rendering, native rendering in 12 languages, dense infographic/document layouts, reference-preserving edits; $0.04/image, API-only
Image Gen
2026-08-05
—
$0.003/image
$0.04/image
XiaomiVLA robot foundation model, Qwen3-VL backbone + Diffusion-Transformer (Mixture-of-Transformers); trained on 100,000+ hrs UMI data plus 7,200+ hrs real-robot data; demoed autonomous laundry folding, washer loading, suitcase packing
Embodied
2026-08-05
—
—
—
Alibaba
Qwen3.8-Max
Open
2.4T MoE, ~95B active/token; text+image+video input; explicit reasoning (low/high/xhigh, xhigh default); flat pricing across full 1M ctx, max output 131K; open weights promised ~mid-Aug 2026 (HF/ModelScope)
Multimodal
58/100
2026-08-03
1M
$2.00/1M tokens
$6.0/1M tokens
RedNote (Xiaohongshu)
dots3-Note Preview
Open
Lightest model in RedNote HiLab's dots3 series — 280B total / 16B active params (MoE), 512K context, natively understands text, images, and audio. An internal harness built on the model scored a certified perfect 42/42 (IMO gold) at IMO 2026.
Text
—
2026-08
512K
—
—
ByteDance
Seedance 2.5
Closed
Native single-shot 30s video generation up to 4K, accepts up to 50 multimodal references; launched head-to-head vs MiniMax H3
Video Gen
2026-07-31
—
—
—
MiniMax
H3 (Hailuo H3)
Closed
Omni-modal video generator (text/image/video/audio input), up to 15s native 2K video with native stereo audio; open weights planned within days
Video Gen
2026-07-31
—
—
—
DeepSeekPublic-beta API release; retrained V4-Flash beats DeepSeek's own larger V4-Pro-Preview on all 9 published agent/coding benchmarks
Text
52/100
2026-07-31
1M
$0.14/1M tokens
$0.28/1M tokens
SenseTimeLightweight 8B-MoT open unified multimodal model (NEO-Unify architecture); adds native 4K image generation and stronger instruction-following edits vs prior U1
Image Gen
2026-07-31
—
—
—
Huawei
openPangu-2.0-Pro
Open
505B-param MoE (18B active), open-sourced with weights, inference code & technical report; trained 100% on Huawei Ascend 910B NPUs
Text
—
2026-07-31
512K
—
—
Alibaba
Qwen3.7-Flash
Closed
Native vision-language Flash upgrade with stronger object recognition, spatial intelligence, and multimodal agent execution
Multimodal
—
2026-07-25
1M
$0.03/1M tokens
$0.13/1M tokens
Ant Group
Ling-3.0-Flash
Open
124B-param MoE, only 5.1B active/token; hybrid-reasoning combines Ling speed + Ring reasoning; outperforms Ant's own 1T Ling-2.6 on 11/12 benchmarks; API-only (no open weights at launch), free on OpenRouter through 2026-08-03
Text
38/100
2026-07-23
256K
$0.021/1M tokens
$0.063/1M tokens
Alibaba
ABot suite
Closed
Amap's full-stack embodied AI upgrade: five models for navigation, manipulation, reasoning, memory, and motion control; claims SOTA on 17 benchmarks
Embodied
2026-07-22
—
—
—
Tencent
Hy-Embodied-VLA-0.5
Open
Vision-language-action model trained on 10,000+ hours of data, deployable across different robot platforms
Embodied
2026-07-19
—
—
—
Tencent
Hy-Embodied-VLM-1.0
Open
Perception model for robots; best-in-class on 19 of 38 benchmarks at 1/10th the compute of Tencent's prior flagship
Embodied
2026-07-19
—
—
—
TencentReasoning engine for embodied cognition: joint language-visual reasoning, world-state prediction, subgoal planning; ships with open RxBrain-Bench
Embodied
2026-07-19
—
—
—
SenseTime
SenseNova-U1 Pro
Closed
Unifies seeing, generating, acting via NEO-unify architecture; native 8K output; debuted at WAIC 2026
Multimodal
—
2026-07-19
—
—
—
Kunlun Tech
Mureka v9.5
Closed
AI music model with O3 reflective-reasoning + MuCo creation agent; pitched as least AI-sounding music generator, unveiled at WAIC
Multimodal
—
2026-07-19
—
—
—
Kunlun Tech
Matrix-Game 3.5
Closed
Interactive world-model upgrade shown at WAIC; 5B model, real-time 720p-class generation at 20FPS with 1-minute memory
Video Gen
2026-07-19
—
—
—
ModelBest
MiniCPM-RobotTrack
Open
0.9B on-device VLA model for person-tracking, unveiled at WAIC 2026; SOTA on 3 tracking benchmarks, runs 5+ FPS on Unitree Go2 onboard compute
Embodied
2026-07-17
—
—
—
ModelBest
MiniCPM-RobotManip
Open
1.5B on-device VLA model for robot manipulation, unveiled at WAIC 2026; outperforms larger models like π0.5 and Qwen-VLA per ModelBest
Embodied
2026-07-17
—
—
—
WeRide
WITT
Closed
Physical AI foundation model built on Atomic Physical Facts; claims 98% lower token cost and 200x data efficiency vs general models (self-reported, unverified)
Embodied
2026-07-17
—
—
—
Unisound
U2-Med
Closed
First tri-medical LLM spanning healthcare, insurance and pharma; built on U2 foundation model
Multimodal
—
2026-07-17
—
—
—
Moonshot AI
Kimi K3
Open
Flagship model, 2.8T total parameters, strong coding and agent capabilities, open weights by July 27
Multimodal
60/100
2026-07-16
1M
$3.00/1M tokens
$15.00/1M tokens
Ant Group
LingBot-VA 2.0
Open
Industry-first embodied-native causal video-action model; dual-stream mixture-of-transformers architecture, unifies frame prediction and policy execution from scratch
Embodied
2026-07-11
—
—
—
Ant Group
LingBot-World 2.0
Open
Open-source world model generating hour-long, no-decay 720p/60fps interactive experiences with built-in dynamic agents; up from minutes-level in v1.0
Embodied
2026-07-09
—
—
—
Alibaba
RynnWorld-4D
Open
DAMO Academy release; generates robot's predicted future as joint RGB + depth + optical-flow stream for manipulation planning, not just 2D video
Embodied
2026-07-08
—
—
—
Ant Group
LingBot-VLA 2.0
Open
6B open-source cross-embodiment model; trained on 60K hrs real-world data across 20 robot morphologies from 17 manufacturers
Embodied
2026-07-08
—
—
—
Ant Group
LingBot-Vision
Open
1.1B-param ViT-g/16 self-supervised vision model using masked boundary modelling, trained on ~161M curated images; 4 sizes released under Apache-2.0
Embodied
2026-07-07
—
—
—
Ant Group
LingBot-Depth 2.0
Closed
Next-gen monocular depth model paired with LingBot-Vision; claims to roughly halve robotic depth-estimation error in challenging scenarios
Embodied
2026-07-07
—
—
—
Tencent
Hy3 (Hunyuan 3.0)
Open
295B total / 21B active + 3.8B MTP layer, 192 routed experts (top-8) + 1 shared; official release upgrades Apr 2026 preview, deployed across WeChat, Yuanbao, WorkBuddy
Text
—
2026-07-07
256K
$0.18/1M tokens
$0.60/1M tokens
Shengshu Technology
Vidu S1
Closed
Real-time interactive video model with voice-controlled avatars; 540p up to 42fps; runs on consumer-grade GPUs instead of server clusters
Video Gen
2026-07-03
—
—
—
Zhipu AI (Z.ai)
GLM-5.2
Open
753B params; Rolled out to all GLM Coding Plan tiers (Lite/Pro/Max/Team); standalone API + MIT open weights
Text
51/100
2026-06-01
1M
$1.40/1M tokens
$4.40/1M tokens
MiniMax
MiniMax M3
Open
428B params (23B active); 1-million-token context window; novel attention mechanism called MSA; native multimodality
Multimodal
44/100
2026-06
1M
$0.30/1M tokens
$1.20/1M tokens
Moonshot AI
Kimi-K2.7-Code
Open
1T total params / 32B active, 384 experts, ~30% lower reasoning-token usage vs K2.6
Code
—
2026-06
256K
$0.95/1M tokens
$4.00/1M tokens
StepFun
Step 3.7 Flash
Open
198B total / 11B active params; 1.8B ViT vision encoder; 3 reasoning levels (low/med/high); ~400 TPS; Apache 2.0; supports NVIDIA NIM on-prem deployment
Multimodal
—
2026-06
256K
$0.2/1M tokens
$1.15/1M tokens
Meituan
LongCat-2.0
Open
1.6T para + MoE; trained on 50,000 domestic Chinese chips; next-gen flagship
Text
—
2026-06
1M
$0.75/1M tokens
$2.95/1M tokens
XPeng
VLA 2.0
Closed
Second-generation vision-language-action world model — deliberative reasoning, controllable generation, and long-horizon forecasting. Runs on 4 self-developed Turing AI chips (~3,000 TOPS effective) in the Robotaxi platform.
Embodied
2026-06
—
—
—
Alibaba
Qwen3.7-Max
Closed
Flagship closed-source reasoning model. Ranked #1 on Artificial Analysis Intelligence Index (57/100) out of 218 models at release. 1M token context window. Text-only (no multimodal). Architecture reportedly dual-72B. Extended thinking / chain-of-thought mode. Released May 19, 2026.
Text
57/100
2026-05
1M
$2.50/1M tokens
$7.50/1M tokens
Baidu
ERNIE 5.1
Closed
MoE successor to ERNIE 5.0. ~⅓ total params, ~½ active params, 6% of pre-training cost. AIME 2026: 99.6% with tool use (#2 globally). Arena Search: 1,223 (#4 global, #1 China). Released May 9, 2026.
Text
—
2026-05
128K
$0.59/1M tokens
$2.65/1M tokens
Baichuan AI
Baichuan-M4
Closed
Medical AI flagship (May 2026). World #1 on HealthBench, HealthBench Hard & HealthBench Professional. Hallucination rate 3.3% — industry low via factuality-aware RL. Surpasses GPT-5.5, Opus 4.7, DeepSeek-V4-Pro. Powers 百小医 AI family doctor.
Text
—
2026-05
—
—
—
ModelBest
MiniCPM-V 4.6
Open
1.3B multimodal model (SigLip2 + Qwen3.5-0.8B). 260K token context. Runs on consumer phones (iOS/Android/HarmonyOS). Launched May 2026.
Multimodal
—
2026-05
260K
—
—
ModelBest
MiniCPM5-1B
Open
1B language model with 128K token context. SOTA on-device LLM at its size class. Apache 2.0.
Text
—
2026-05
128K
—
—
Zhipu AI (Z.ai)
GLM-5.1
Open
754B flagship — top scores on AIME 2026 (95.3) and GPQA-Diamond (86.2); strong agentic and coding ability
Text
—
2026-04
200K
$0.98/1M tokens
$3.08/1M tokens
MiniMax
MiniMax-M2.7
Open
229B MoE; self-evolving agent — 30% perf gain over 100+ rounds; agent teams, complex skills, 24/7 background agents
Agent
2026-04
1M tokens
$0.28/1M tokens
$1.20/1M tokens
Moonshot AI
Kimi K2.6
Open
1T MoE (32B active), 256K ctx; native multimodal agentic — top SWE-Bench (80.2) and AIME 2026 (96.4)
Multimodal
—
2026-04
256K
$0.60/1M tokens
$2.50/1M tokens
DeepSeek
DeepSeek-V4-Pro
Open
Flagship MoE: 1.6T total / 49B active params, 1M context, FP4+FP8 precision, built-in Think/Non-Think reasoning modes. Codeforces rating 3206.
Text
—
2026-04
1M
$0.435/1M tokens
$0.87/1M tokens
DeepSeek
DeepSeek-V4-Flash
Open
Efficient MoE: 284B total / 13B active params, 1M context. Same architecture as V4-Pro at far lower compute cost. Codeforces 2816.
Text
—
2026-04
1M
$0.14/1M tokens
$0.28/1M tokens
Baidu
ERNIE-Image
Open
8B single-stream Diffusion Transformer (DiT). Best text rendering in open source (LongTextBench: 0.9733). Up to 4K output. Apache 2.0. Released Apr 15, 2026.
Image Gen
2026-04
—
$0.03/image
—
Ant Group
Ring-2.6-1T
Open
1T total params, ~63B active per token; adaptive reasoning via high / xhigh modes; up to 66K output tokens; open weights (MIT); released May 8 2026; free tier available on OpenRouter
Text
—
2026-04
262K
$0.075/1M tokens
$0.625/1M tokens
Ant Group
Ling-2.6-1T
Open
1T total params, ~63B active per token; fast thinking approach cuts token cost to ~1/4 of comparable models; open weights (MIT); released Apr 23 2026; free tier available on OpenRouter via Novita AI
Text
34/100
2026-04
262K
$0.075/1M tokens
$0.625/1M tokens
Ant Group
Ling-2.6-flash
Open
Efficient sparse MoE: 104B total / 7.4B active, hybrid linear attention. 340 tokens/s on 4× H20. Strong agent and tool-use performance.
Text
—
2026-04
262K
$0.10/1M tokens
$0.30/1M tokens
Tencent
Hy3-preview
Closed
295B MoE (21B active); reasoning, coding, and agentic workloads. Via Tencent Cloud TokenHub. Released Apr 23, 2026.
Text
—
2026-04
256K
¥1.2/1M tokens
¥4.0/1M tokens
Xiaomi
MiMo-V2.5-Pro
Open
1.02T MoE, 42B active; KV-cache reduced ~7×; latest generation
Text
—
2026-04
1M
$0.0036/1M tokens
$0.087/1M tokens
AGIBot
Genie Operator-2 (GO-2)
Closed
AGIBot's new-generation embodied foundation model, powering autonomous manipulation and locomotion across its AGIBOT A2 and X2 robot lines.
Embodied
2026-04
—
—
—
Xiaomi
MiMo-V2-Pro
Closed
1T+ MoE, 42B active; hybrid attention; 1M token context
Text
—
2026-03
1M
$0.0036/1M tokens
$0.087/1M tokens
Unitree Robotics
UnifoLM-VLA-0
Open
Vision-Language-Action model enabling the G1 humanoid to autonomously perform household tasks from natural language commands. Runs onboard the robot. Open-sourced March 2026 — Unitree's first AI model release.
Embodied
2026-03
N/A
—
—
Kunlun Tech
SkyReels V4
Closed
Audio-visual creation model (Mar 2026). Dual-stream architecture; #1 globally on Text-to-Video (With Audio) and Image-to-Video (With Audio) tracks at release. Generates clips up to 3 minutes.
Video Gen
2026-03
—
—
—
Kunlun Tech
Mureka V9
Closed
Music generation model (Mar 2026). Paragraph-level text control; enhanced mixing quality, vocal expression, and style richness.
Audio
2026-03
—
—
—
Kunlun Tech
Matrix-Game 3.0
Closed
Physics-simulation interactive world model (Mar 2026). 5B params; 720P @ 40FPS; covers 1,000+ scenarios with Unreal Engine data. Industrial-grade real-time interactivity.
Text
—
2026-03
—
—
—
ByteDance
Seed2.0 Pro
Closed
ByteDance's flagship multimodal model. Understands text, image, and video. Ranks #3 globally on LMSYS Vision Arena and #6 on overall text arena. AIME 2025: 98.3, SWE-bench: 76.5%.
Multimodal
—
2026-02
272K
$0.47/1M tokens
$2.37/1M tokens
ByteDance
Seedream 5.0 Lite
Closed
Unified multimodal image generation with chain-of-thought visual reasoning, real-time web search, and native editing. Supports up to 14 reference images. Up to 4K output at 2048×2048.
Image Gen
2026-02
—
$0.026/image
—
ByteDance
Seedance 2.0
Closed
Unified audio-video joint generation model. Accepts text, image, video, and audio inputs; outputs native 2K video (up to 15s) with synchronized audio. 30% faster than Seedance 1.5 Pro.
Video Gen
2026-02
—
—
—
Alibaba
Qwen3.5
Open
Unified vision-language MoE family. Flagship: 35B-A3B (35B total / 3B active) and 397B-A17B. Thinking mode on by default. Supports image, video, text input. 201 languages. Apache 2.0.
Multimodal
—
2026-02
262K (1M w/ YaRN)
—
—
Alibaba
Qwen3-Coder-Next
Open
80B total / 3B active MoE coding agent. Excels at long-horizon agentic tasks, tool use, and IDE integration (Claude Code, Cline, Qwen Code). No thinking mode.
Code
—
2026-02
256K
—
—
Zhipu AI (Z.ai)
GLM-5
Open
754B MoE (40B active); 28.5T token pre-training; top-tier SWE-bench score of 77.8
Text
—
2026-02
200K
$1.40/1M tokens
$4.40/1M tokens
MiniMax
MiniMax-M2.5
Open
229B; SOTA SWE-Bench (80.2); 80% of MiniMax's own code generated by this model; M2.5-Lightning at 100 tok/s
Agent
2026-02
1M tokens
$0.30–$1.00/hr
$0.30–$1.00/hr
Baidu
ERNIE 5.0
Closed
2.4 trillion parameter unified multimodal (text + image + video + audio) in a single autoregressive framework. LMArena Text: 1,460 (#1 China, #8 global); Vision: 1,226 (#1 China, #8 global). Released Feb 6, 2026.
Multimodal
—
2026-02
128K
—
—
Ant Group
LingBot-VLA
Open
Open-source VLA foundation model trained on ~20,000 hours of real-world dual-arm robot data across 9 embodiments
Embodied
2026-02
—
—
—
Ant Group
Ring-2.5-1T
Open
World's first open-source 1T-parameter thinking model. Gold medal level at IMO 2025 (35/42 pts) and CMO 2025 (105/126). 3× throughput for sequences >32K.
Text
—
2026-02
256K
—
—
Ant Group
Ming-flash-omni-2.0
Open
Any-to-any omni model: accepts image, text, video, and audio; outputs image, text, and audio. 100B total / 6B active MoE. Supports zero-shot voice cloning and image generation/editing.
Multimodal
—
2026-02
—
—
—
Ant Group
LLaDA-2.1-flash
Open
Novel diffusion-based language model — not autoregressive. 103B params. Generates text by iterative token editing rather than left-to-right decoding. 102K monthly downloads.
Multimodal
—
2026-02
—
—
—
Baichuan AI
Baichuan-M3
Open
235B medical model (Feb 2026) built on Qwen3 architecture. Former world #1 on HealthBench. Hallucination rate 3.5%. Outperforms human doctors in diagnostic accuracy. 48GB VRAM with W4 quantization.
Text
—
2026-02
—
—
—
JD.com
JoyAI-LLM-Flash
Open
Open-sourced lightweight model — 4.8B total / 3B active params, pretrained on 20T text tokens. Released on HuggingFace as part of JD's open-source push.
Text
—
2026-02
—
—
—
Moonshot AI
Kimi K2.5
Open
1T MoE (32B active); trained on 15T vision+text tokens; thinking & instant modes; agent swarm support
Multimodal
—
2026-01
256K
$0.40/1M tokens
$1.90/1M tokens
DeepSeek
DeepSeek-OCR-2
Open
3B visual OCR model with document-to-markdown conversion, layout-aware grounding, and dynamic resolution. 1.66M monthly downloads.
Multimodal
—
2026-01
—
—
—
Tencent
HunyuanImage 3.0
Open
80B MoE (13B active); text-to-image and image-to-image with reasoning. Open weight on HuggingFace. Released Jan 26, 2026.
Image Gen
2026-01
—
—
—
Meituan
LongCat-Flash-Lite
Open
68.5B MoE, 2.9–4.5B active; fast and efficient
Text
—
2026-01
256K
—
—
AIsphere
PixVerse R1
Closed
Real-time world model; 1080p, <15s latency, physics-aware, infinite temporal continuity
Video Gen
2026-01
N/A
—
—
SenseTime
SenseNova V6.5 Omni
Closed
Real-time multimodal streaming model. Powers SenseChat's live audio/video interaction. Successor to V6 Pro.
Multimodal
—
2026-01
—
—
—
LimX Dynamics
FluxVLA Engine
Closed
LimX's Vision-Language-Action embodied-AI engine powering autonomous manipulation and locomotion on its humanoid and legged robots.
Embodied
2026
—
—
—
DeepSeek
DeepSeek-V3.2
Open
685B MoE with DeepSeek Sparse Attention (DSA). V3.2-Speciale variant won gold at 2025 IMO and IOI. 4.16M monthly downloads on HuggingFace.
Text
—
2025-12
128K
$0.252/1M tokens
$0.378/1M tokens
Kuaishou
Kling I2V 2.0
Closed
Image-to-video with industry-leading subject preservation
Multimodal
—
2025-12
N/A
$0.16/video (5s)
—
Xiaomi
MiMo-V2-Flash
Open
309B MoE, 15B active; trained on 27T tokens; open-source release
Text
—
2025-12
N/A
$0.01/1M tokens
$0.30/1M tokens
AIsphere
PixVerse V5.5
Closed
Text/image-to-video; HD output, multiple aspect ratios, character consistency
Video Gen
2025-12
N/A
—
—
Tencent
Hunyuan3D 3.0
Closed
3D asset generation from text, image, or sketch input
3D Gen
2025-11
—
—
—
iFlyTek
Spark X1.5
Closed
MoE reasoning model (29.3B total / 3B active parameters). Supports 130+ languages. Runs on a single Huawei Ascend server. Launched Nov 2025.
Text
—
2025-11
—
—
—
MiniMax
MiniMax-M2
Open
230B MoE (10B active); interleaved thinking; open-source under Modified MIT; free API available
Agent
2025-10
1M tokens
—
—
Infinigence AI
Megrez2-3x7B-A3B
Open
MoE edge model: 3×7B experts, 3B active parameters. Open-weight, Apache 2.0.
Text
—
2025-09
—
—
—
Alibaba
Qwen-Image
Open
Image generation and editing foundation model. Exceptional Chinese + English text rendering. Supports style transfer, object insertion/removal, layered editing, and depth/edge estimation. Released Aug 2025.
Image Gen
2025-08
—
—
—
ModelBest
MiniCPM-V 4.5
Open
8B multimodal model (Qwen3-8B + SigLIP2). Apache 2.0.
Multimodal
—
2025-08
—
—
—
Moonshot AI
Kimi k2
Open
Agentic model with tool use, web browsing, and code execution
Text
—
2025-07
128K
$0.57/1M tokens
$2.30/1M tokens
MiniMax
MiniMax-M1
Open
First open-weight large-scale hybrid-attention reasoning model; test-time compute scaling
Text
—
2025-06
1M tokens
$0.80/1M tokens
$2.20/1M tokens
MiniMax
Hailuo-02
Open
Physics-aware video generation; top-3 globally on VBench
Video Gen
2025-06
N/A
$0.06/video (5s)
—
Shengshu Technology
Vidu Q3
Closed
World's first storytelling-focused video model. Up to 16 sec, 1080p 24fps, native audio sync. Multilingual. Pro and Turbo variants. MCP integration.
Video Gen
2025-06
—
~$0.07/sec
—
Huawei
Pangu Ultra MoE (5.5)
Closed
718B-parameter MoE model (256 experts), trained on the full-stack hardware and software of Huawei Cloud's AI Cloud Service (CloudMatrix 384 supernodes). Part of the Pangu Models 5.5 family, which per Huawei Cloud's own release delivers upgrades across five capabilities: natural language processing, computer vision, multimodal, prediction, and scientific computing.
Text
—
2025-06
—
—
—
RedNote (Xiaohongshu)
dots.llm1
Open
RedNote's first open-source LLM — 142B total / 14B active params (MoE), pretrained on 11.2T high-quality tokens with no synthetic data. Performance comparable to Alibaba's Qwen2.5-72B.
Text
—
2025-06
32K
—
—
DeepSeek
DeepSeek-R1-0528
Open
Latest R1 update: 685B, adds system prompt support, deeper reasoning (23K avg tokens on AIME vs 12K prior). Distilled 8B version achieves 86% on AIME 2024.
Text
—
2025-05
128K
$0.50/1M tokens
$2.15/1M tokens
Kuaishou
Kling 2.0
Closed
4K-capable video generation; advanced camera controls and scene coherence
Video Gen
2025-04
N/A
$0.16/video (5s)
—
Kuaishou
Kolors 2.0
Open
Upgraded photorealistic image model; open-source available
Image Gen
2025-04
N/A
$0.004/image
—
SenseTime
SenseNova V6 Pro
Closed
620B MoE hybrid. Real-time audio/video streaming. Ranked #1 in China in multimodal reasoning at launch. Lowest reasoning cost in industry at launch (Apr 2025).
Multimodal
—
2025-04
256K
¥2.8/1M tokens
¥8.4/1M tokens
Kunlun Tech
Skywork-OR1-32B
Open
Open-source math and coding reasoning model (Apr 2025). 32B params; rivals DeepSeek-R1 on competitive programming benchmarks. 7B variant also available.
Text
—
2025-04
—
—
—
Qihoo 360
360Zhinao3-7B
Open
Latest open-weight 360Zhinao generation (Base / Instruct / O1.5 long chain-of-thought reasoning variant). Free for commercial use.
Text
—
2025-04
32K
—
—
Baidu
ERNIE X1
Closed
Baidu's first dedicated reasoning model with extended chain-of-thought
Text
—
2025-03
128K
$0.28/1M tokens
$1.10/1M tokens
Baidu
ERNIE 4.5
Closed
Improved multimodal and reasoning; best ERNIE for Chinese enterprise tasks. The flagship ERNIE 4.5 model is closed-source, though several variants in the ERNIE 4.5 family are openly available on HuggingFace.
Multimodal
—
2025-03
128K
$0.55/1M tokens
$2.20/1M tokens
StepFun
Step-3
Closed
StepFun's 2025 flagship; 1M token context and native reasoning mode
Text
—
2025-01
1M
$0.57/1M tokens
$1.42/1M tokens
Baichuan AI
Baichuan-M2
Open
32B medical reasoning model (2025) built on Qwen2.5-32B with innovative Large Verifier System for real-world clinical reasoning.
Text
—
2025-01
—
—
—
ModelBest
MiniCPM-o 2.6
Open
8B any-to-any model (text + vision + speech). Surpasses GPT-4o and Gemini 1.5 Pro on single-image understanding benchmarks. Launched Jan 2025.
Multimodal
—
2025-01
—
—
—
Kuaishou
Kling 1.6 Pro
Closed
Previous flagship; supports 1080p, 5-10s clips, camera controls
Video Gen
2024-12
N/A
$0.14/video (5s)
—
Infinigence AI
Megrez-3B-Omni
Open
3B omni model processing text, vision, and audio. Outperforms LLaVA-NeXT-Yi-34B on vision benchmarks. Optimised for on-device and edge deployment.
Multimodal
—
2024-12
—
—
—
Kunlun Tech
Skywork-o1
Open
Reasoning model (Nov 2024) — among China's first o1-style models with chain-of-thought Chinese logical reasoning. Open-source 8B variant (Llama 3.1 base) plus proprietary advanced version.
Text
—
2024-11
—
—
—
Qihoo 360
360Zhinao2-7B
Open
Second-generation 360Zhinao with Base and Chat variants at 4K / 32K / 360K context lengths.
Text
—
2024-11
360K
—
—
Baichuan AI
Baichuan4-Turbo
Closed
Enterprise flagship general LLM (late 2024). 10%+ usability gain vs prior generation; priced at ~80% of GPT-4o. Supports multimodal input.
Text
—
2024-10
192K
—
—
Baichuan AI
Baichuan4-Air
Closed
MoE variant (PRI architecture) of Baichuan4. High performance at low cost for API deployments.
Text
—
2024-10
192K
—
—
01.AI
Yi-Lightning
Closed
MoE flagship API model (Oct 2024). Ranked #6 on Chatbot Arena at launch — joint 3rd among LLM companies. 40% faster inference than prior Yi models. Final model before 01.AI halted pre-training in early 2025.
Text
—
2024-10
16K
$0.14/1M tokens
$0.14/1M tokens
01.AI
Yi-Coder
Open
Code model (Sep 2024); 1.5B and 9B variants; supports 52 programming languages; 128K context window.
Code
—
2024-09
128K
—
—
iFlyTek
Spark 4.0
Closed
Flagship LLM (Aug 2024). Claims comparable performance to GPT-4 Turbo on Chinese language benchmarks.
Text
—
2024-08
—
—
—
StepFun
Step-2
Closed
Very long context window; strong document understanding
Text
—
2024-07
256K
¥0.038/1K tokens
¥0.12/1K tokens
StepFun
Step-1X-Image
Closed
High-resolution image generation model
Image Gen
2024-07
N/A
¥0.04/image
—
01.AI
Yi-1.5
Open
Open-source series (May 2024); 6B–34B variants. Improved coding, math, reasoning, and instruction-following over original Yi.
Text
—
2024-05
—
—
—
Qihoo 360
360Zhinao-Search
Open
Retrieval + 1.8B reranking models that ranked #1 in the Retrieval and Reranking tasks of the C-MTEB leaderboard.
Embedding
2024-05
—
—
—
01.AI
Yi-VL-34B
Open
34B vision-language model (early 2024). Open-weight multimodal extension of Yi-34B.
Multimodal
—
2024-01
—
—
—
01.AI
Yi-34B
Open
Founding open-source model (Nov 2023). Topped HuggingFace Open LLM Leaderboard and C-Eval at launch. 200K-token context variant (Yi-34B-200K) also available.
Text
—
2023-11
200K
—
—
Kunlun Tech
Skywork-13B
Open
Founding open-source bilingual LLM (Oct 2023). 13B params; pretrained on 3.2T tokens. Led same-scale models on CEVAL, CMMLU, and MMLU at launch.
Text
—
2023-10
—
—
—
Baichuan AI
Baichuan2
Open
Open-source bilingual LLM series (Sep 2023); 7B and 13B variants; trained on 2.6T tokens. Available on Hugging Face under permissive license.
Text
—
2023-09
4K
—
—
JD.com
ChatRhino (言犀)
Closed
JD's 100B-parameter industry-focused LLM, blending 70% general data with 30% native supply-chain intelligence data for retail, logistics, finance, healthcare, and city-services applications.
Text
—
2023-07
32K
—
—
Meituan
LongCat-Image
Open
Image generation model, ~6B params; data-quality focused; released Dec 2025
Image Gen
—
N/A
—
—
MeituanChain-of-thought reasoning variant of LongCat-Flash
Text
—
—
256K
—
—
Meituan
LongCat-Flash-Chat
Open
560B MoE, 27B active; open-source; 500K free tokens/day
Text
—
—
128K
—
—
WeRide
WeRide GENESIS
Closed
Generative Engineered Neural Environment for Simulated Intelligence in Self-driving. WeRide's proprietary general-purpose simulation platform combining physical AI with generative AI. Rapidly builds photorealistic simulated cities in minutes, generates diverse edge-case scenarios from billions of km of real-world data, and models realistic pedestrian and driver behavior — all at centimeter-level fidelity. Supports L2++ through L4 AV development and validation via four modules: AI Scenarios, AI Agents, AI Metrics, and AI Diagnosis.
Embodied
—
—
—
—
Pony.ai
PonyWorld 2.0
Closed
Second-generation world model underpinning the Virtual Driver L4 autonomous driving platform. Introduces an Intention layer — a structured representation of decision-making that enables the system to evaluate its own driving decisions, identify accuracy gaps across scenarios, and direct targeted data collection rather than broad undirected improvement. Direct sensor-to-action architecture with no language models in the inference pipeline; runs on 1016 TOPS across three NVIDIA DRIVE Orin-X SoCs with redundant failover.
Embodied
—
—
—
—
Unisound
UniGPT (Shanhai 山海)
Closed
60B+ parameter general large model (v5.0); underpins all Unisound vertical products. Medical, enterprise, and consumer deployments.
Text
—
—
—
—
—
Unisound
U2-ASR 2.5
Closed
First LLM-based semantic ASR model for Chinese. Covers 100+ dialects across 7 dialect systems. >90% accuracy. Available via Token Hub API.
Audio
—
—
—
—
Unisound
U2-TTS / U2-TTS-Clone
Closed
Text-to-speech and voice cloning with full-duplex millisecond response. Available via Token Hub API.
Audio
—
—
—
—
Unisound
U1-OCR
Closed
Industrial-grade document intelligence model for OCR and document understanding. Launched February 2026.
Multimodal
—
—
—
—
—
Disclosed, 2026 YTD
Funding & Markets
Private funding rounds and public market data for China's leading AI companies
43
Companies tracked
22 public · 21 private
22 public · 21 private
~$29.9B
AI startup funding
12 AI-native cos
12 AI-native cos
Compare companies →
Valuation, total funding & AI ARR, side-by-side
Most Active Investors
1
Tencent
16
2
HongShan Capital
12
3
Alibaba
12
4
IDG Capital
9
5
Shunwei Capital
7
Derived live from 167 rounds
ByteDance
PRIVATE
Valuation$600B
Total Raised~$12.61B+
AI ARR$4B
Alibaba
BABA / 9988.HK
Market Cap—
Cloud Revenue (FY25)¥158B
AI ARR¥10B
Zhipu AI (Z.ai)
2513.HK
Market Cap—
Total Raised$5.79B
AI ARR$1.6B
MiniMax
0100.HK
Market Cap—
Total Raised$2.17B
AI ARR$800M
Moonshot AI
PRIVATE
Valuation$35B
Total Raised~$8.2B+
AI ARR$300M
DeepSeek
PRIVATE
Valuation$50B
Total Raised$7.35B
AI ARR$500M
StepFun
PRIVATE
Valuation$10B
Total Raised$3.23B+
Last Round2026
Baidu
BIDU / 9888.HK
Market Cap—
ExchangeNASDAQ: BIDU · HKEX: 9888
Founded2000
Ant Group
PRIVATE
Valuation$79B
Total Raised$20.53B+
Last Round2020
Tencent
0700.HK
Market Cap—
Revenue (FY25)¥751.8B
Founded1998
Kuaishou
1024.HK
Market Cap—
ExchangeHKEX: 1024
Kling ARR$500M
Xiaomi
1810.HK
Market Cap—
Revenue (FY25)¥457.3B
Founded2010
Meituan
3690.HK
Market Cap—
Revenue (FY25)¥364.9B
Founded2010
AIsphere
PRIVATE
Valuation$2B
Total Raised~$669M+
Last Round2026
Unitree Robotics
688836.SS
Market Cap—
FY2025 Revenue¥1.71B
Founded2016
SenseTime
0020.HK
Market Cap—
Revenue (FY25)¥5.01B
Founded2014
Baichuan AI
PRIVATE
Valuation$2.7B
Total Raised$1.04B
Last Round2024
01.AI
PRIVATE
Valuation$1B
Total Raised$300M+
Last Round2024
Kunlun Tech
300418.SZ
Market Cap—
Revenue (FY25)¥8.2B
Founded2008
Evoken
PRIVATE
Valuation$2B
Total Raised$430M+
Last Round2026
iFlyTek
002230.SZ
Market Cap~¥190B
FY2025 Revenue¥27.1B
Founded1999
Infinigence AI
PRIVATE
ValuationUndisclosed
Total Raised~$251M+
Last Round2026
Enflame Technology
PRIVATE
Valuation—
Total Raised—
Last Round—
SiliconFlow
PRIVATE
Valuation¥7.7B
Total Raised$287M
Revenue¥55.3M
Cambricon
688256.SS
Market Cap—
FY2025 Revenue¥6.5B
Founded2016
ModelBest
PRIVATE
Valuation¥20B
Total RaisedUndisclosed
Last Round2026
Moore Threads
— (SSE)
Market Cap—
FY2025 Revenue¥1.505B
Founded2020
WeRide
WRD / 0800.HK
Market Cap—
FY2025 Revenue~$93.6M
Founded2017
Pony.ai
PONY / 2026.HK
Market Cap—
FY2025 Revenue$90.0M
Founded2016
Shengshu Technology
PRIVATE
ValuationUndisclosed
Total Raised~$440M
Last Round2026
DiDi Autonomous Driving
PRIVATE
ValuationUndisclosed
Total Raised$1.39B+
Last Round2025
Unisound
9678.HK
Market Cap—
FY2025 Revenue¥1.21B
Founded2012
Momenta
06880.HK
Market Cap—
FY2025 Revenue¥2.4B
Founded2016
Qihoo 360
601360.SS
Market Cap~$10.1B
Market Cap~$10.1B
Founded2005
CXMT
688825.SS
Market Cap—
IPO Proceeds~$8.6B
Founded2016
Manus (Butterfly Effect)
PRIVATE
Valuation~$500M
Total Raised$75M
Last Round2025
LimX Dynamics
PRIVATE
Valuation¥15B
Total Raised~$400M+
Last Round2026
XPeng
XPEV / 9868.HK
Market Cap—
ExchangeNYSE: XPEV · HKEX: 9868
Founded2014
AGIBot
PRIVATE
Valuation~$1.5B+
Total Raised$84M+
Last Round2025
Huawei
PRIVATE
FY2025 Revenue¥880.9B
R&D Investment¥192.3B
Founded1987
Pragmatik Labs
PRIVATE
Valuation$2B
Total Raised$220M
Last Round2026
RedNote (Xiaohongshu)
PRIVATE
Valuation~$70B
Total Raised~$917M+
Last Round2026
JD.com
JD / 9618.HK
Market Cap—
FY2025 Revenue¥1,309.1B
Founded2004
Data methodology: Funding data on this page is compiled from publicly available sources — company press releases, regulatory filings, and media reports — and fact-checked by Tony Peng with the assistance of AI. Where a round appears in both a company disclosure and press report, company disclosure takes precedence. All figures may be approximate.
Shipped Products
Applications
AI products from China's leading companies, by category
Est. 2026
The essential tracker for China's AI ecosystem
One place to follow every major model release, funding round, product launch, and breakthrough from China's AI ecosystem.
Tony Peng
AI observer · Journalist · Communications professional
China AI Index is a research project built and maintained by Tony Peng — a long-time observer of the AI industry with a background in journalism and communications — with the assistance of AI (Claude Code, Qoder, and Z Code). Tony is also the writer behind Recode China AI, a newsletter tracking the latest advances and news stories in intelligent technologies.
It exists to give English-speaking researchers, investors, journalists, and developers a structured reference for China's AI ecosystem — most authoritative information is published in Chinese, and this site translates and synthesizes it.
What we track
Models
Every major LLM, reasoning model, vision model, and generative AI model from Chinese labs — with benchmark scores, context windows, and live API pricing so you can compare apples to apples.
Financing
Full funding histories for private companies (ByteDance, DeepSeek, MiniMax, Moonshot…) and market data for public ones (Alibaba, Baidu, Tencent, Kuaishou). Valuations, investors, round sizes.
Newsfeed
News covering model launches, research breakthroughs, product announcements, and strategic moves — presented in a clean timeline with company tags and category filters.
People
Key founders, CEOs, CTOs, and researchers behind China's AI labs — with titles, Chinese names, and links to their social profiles so you know the faces shaping the industry.
Members
Regulation
A timeline of China's AI rules — generative AI measures, algorithm filing requirements, model approval lists, data law touchpoints, and provincial subsidy programs. Click any entry for the original text.
Members
Chips
Domestic AI accelerators, training clusters, and memory chips — specs, interconnect bandwidth, process nodes, and deployment status for China's emerging chip ecosystem.
Data Disclaimer
All pricing, benchmark scores, funding amounts, and valuations are sourced from public company announcements, tech media reporting, and developer documentation. They are provided for informational purposes only and may be approximate or subject to change.
Stock prices shown are approximate historical ranges — not real-time quotes. Always verify with a live financial data source before making investment decisions. Benchmark scores reflect results at the time of model release and may not account for subsequent updates.
This database is independently maintained and has no affiliation with any of the companies listed. Information is updated manually on a best-effort basis.
China AI Regulation
A timeline of generative AI rules, algorithm filing requirements, model approval lists, and data law touchpoints — with links to original texts
2026
Jul
Sets a lifecycle security framework for AI agents across five stages — assessment, preparation, deployment, use, decommissioning. Requires least-privilege access, restricted network exposure, full audit logging, secondary confirmation on high-risk operations (payments, deletions, permission changes), sandboxing, and secure data wipe on decommission. Reflects a shift from treating agents as lightweight LLM wrappers to integrated systems with memory/tools/autonomy that need dedicated governance.
View Official Text ↗
2025
Mar
Mandates two-layer labeling for all AI-generated text, images, audio, video, and virtual scenes: an explicit label visible to users (text, voice, or graphic indicator) and an implicit label embedded in file metadata as a digital watermark carrying generation information and producer details. Download, copy, and export functions must preserve labels. Content distribution platforms must verify metadata and add indicators when redistributing AI content. Users publishing AI-generated content must declare its origin. Prohibits malicious deletion, tampering, forgery, or concealment of required labels, and bans tools that enable such conduct. Issued March 7, 2025; effective September 1, 2025.
View Official Text ↗🔒 8 more regulations — Paid Subscribers Only
You're viewing the first 2 entries free. The full regulation timeline — 10 laws, measures, and strategy documents back to 2017 — is included with a paid Recode China AI subscription.
China's Chip Ecosystem
Domestic AI accelerators, large-scale training clusters, and memory chips — tracking the hardware layer behind China's AI buildout.
AI Accelerators
Clusters & Supernodes
Memory & HBM
Role
Precision
| Company | Chip | Role | Peak Compute | Memory Tech | Capacity | Bandwidth | Interconnect Bandwidth |
TDP | Process | Status | Highlights |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Huawei ↗ | Ascend 910B |
Training |
256 TFLOPS
BF16
|
HBM2e | 64 GB | 2.0 TB/s |
392 GB/s
HCCS 3.0
|
400 W | SMIC N+2 | Production | Most-deployed domestic training chip. Used at Alibaba, ByteDance, Baidu, and Tencent. |
| Huawei ↗ | Ascend 910C |
Training |
800 TFLOPS
FP16
|
HBM2e | 96 GB | 4.0 TB/s |
800 GB/s
HCCS
|
900 W | SMIC N+2 | Production | Dual-die upgrade to 910B. Powers the Atlas 900 A3 SuperPoD (300 PFLOPS). |
| 🔒 18 more accelerators — paid subscribers only | |||||||||||
Specs marked ~ are estimates from third-party analysis. Click any column header to sort. Interconnect Bandwidth = chip-to-chip fabric bandwidth per chip (bidirectional).
| Cluster / Supernode | Operator | Chip | Chip Count | Total Compute | Interconnect Fabric | Notes |
|---|---|---|---|---|---|---|
Atlas 900 A3 SuperPoD (CloudMatrix 384) |
Huawei Cloud | Ascend 910C | 384 |
300 PFLOPS
FP16
|
All-optical (OXC) | Huawei's current-generation supernode. Uses optical circuit switching (OXC) for low-latency inter-chip communication. Announced April 2025 as a full-stack domestic alternative to Nvidia GB200 NVL72. |
Atlas 950 SuperPoD |
Huawei Cloud | Ascend 950 | 8,192 |
8 EFLOPS
FP8
|
16.3 PB/s aggregate | Next-generation supernode targeting 2026 delivery. 1,152 TB total memory. Full liquid cooling. Footprint larger than two basketball courts. Would represent a major leap in domestic training capacity if delivered at spec. |
| 🔒 3 more clusters — paid subscribers only | ||||||
Cluster sizes and compute figures are from official announcements or credible press reports.
| Company | Product | Type | Capacity | Bandwidth | Process | Status | Notes |
|---|---|---|---|---|---|---|---|
| CXMT ↗ | DDR5 |
DDR | 16–32 Gb dies | ~6400 MT/s | ~16nm (G4) | Production | CXMT's latest-generation DDR5, unveiled Nov 2025 — its push into the high-end DRAM market against Samsung, SK Hynix, and Micron. |
| CXMT ↗ | LPDDR5X |
LPDDR | Mobile | ~8533 MT/s | ~16nm | Production | Low-power mobile DRAM for smartphones and edge/AI devices. |
| 🔒 2 more memory chips — paid subscribers only | |||||||
DRAM, HBM, and NAND from Chinese memory makers. Specs marked ~ are estimates from third-party analysis or press reports.