huawei

According to a report by Huang Xinyi from the Science and Technology Innovation Board Daily, at the HarmonyOS Office Industry Summit, Huawei launched its first HarmonyOS desktop computer, the Huawei Qingyun HM650 series, and simultaneously released the HarmonyOS Computer Enterprise Edition, enhancing government and enterprise terminal deployment, device management, and data security capabilities.According to on-site reports from IT Home on November 5, the 2026 HarmonyOS Office Industry Summit was held in Wuhan. Zhu Dongdong, President of Huawei's Terminal Tablet and PC Product Line, announced that the Qingyun HM650 series has achieved 100% localization certification, is equipped with the Kirin X90 chip, and has improved overall performance by 15% through software and hardware collaboration. It also debuts with the HarmonyOS Computer Enterprise Edition, supporting features such as HarmonyOS AI, Xiao Yi Work, Xiao Yi Deep Research, and Xiao Yi Smart Note. Huawei's Executive Director Yu Chengdong stated that the cumulative shipment of HarmonyOS computers has surpassed one million units.

According to Gelonghui, citing the China Academy of Information and Communications Technology, on November 4, a joint release of the "Self-Regulatory Convention for the Development of Intelligent Agents" was made by 20 entities including the China Academy of Information and Communications Technology, Huawei, Alibaba, Tencent, and Xiaomi.The convention mentions the construction of interconnected infrastructure for intelligent agents, co-building interoperable infrastructure, developing technologies and standards for intelligent agent registration and resolution, capability description and discovery, intent understanding and transformation, network routing and connection, resource scheduling and measurement, protocol compatibility and mutual recognition, continuously reducing the cost of cross-domain discovery and connection for intelligent agents, and supporting the interconnected collaboration of intelligent agents.The convention also mentions accelerating the empowerment of intelligent agents across various industries, adhering to serving the real economy, developing low-threshold tools for intelligent agent development and deployment, collaborating on research and development of intelligent agents aimed at real operational scenarios that create actual value, promoting the embedding of intelligent agents into existing production systems and business processes, creating benchmark applications, continuously reducing the cost of implementing intelligent agents, and assisting industries in improving quality and efficiency.

The Information investigated the smuggling chain of high-end storage chips in South Korea and interviewed a dealer in Shenzhen in his 20s. He originally worked at a chip company and transitioned to the chip black market business by leveraging his network. Having been in the industry for only three years, he is now dressed in luxury brands and busy connecting buyers and sellers every day. The United States will begin to restrict exports of advanced high-bandwidth memory (HBM) to China starting at the end of 2024, but the rules have loopholes: HBM exports are restricted when sold separately, but if packaged together with processors, it may be judged based on the processing power of the processor to determine if it crosses the regulatory line.A smuggler in his 40s controls a chip design company that specializes in designing cheap and lower-performance server processors. These chips are rarely purchased but are used as a pretext for procuring high-end HBM from South Korea. Packaging factories place HBM from Samsung and SK Hynix on the same substrate as these processors and declare exports under the category of low-performance processors. After the goods enter mainland China via Hong Kong, workshops reheat and remove the HBM, while the cheap processors are discarded, and any mishandling during disassembly could damage the memory.A stack of HBM3E in the Chinese black market was priced at only a few hundred dollars a year ago, but it has now risen to about 4,000 dollars. An AI chip typically requires 6 to 8 stacks, costing between 24,000 to 32,000 dollars just for the memory. Some of the memory ultimately flows to Huawei customers. Huawei's next-generation Ascend 960DT is scheduled to launch in the first quarter of 2027, but there is a shortage of domestic HBM supply, and Huawei has asked some customers to procure memory on their own. With the increase in imitators, the aforementioned smuggler responsible for procuring chips from South Korea plans to sell his controlling stake in the company next year, fearing that the spread of the route will attract strict scrutiny from regulatory authorities in South Korea and the United States.

According to a report by the Science and Technology Innovation Board Daily on November 3, reporters learned from multiple industry chain insiders that the artificial intelligence company DeepSeek's next-generation large model, DeepSeek V5, plans to be trained based on Huawei's Ascend chips. Previously, the Ascend chips were mainly used for inference tasks of the DeepSeek large model.The report stated that at the end of September this year, DeepSeek open-sourced infrastructure components for the Huawei Ascend computing power platform, covering the TileLang high-level language compilation tool, computing libraries, and distributed communication libraries, which correspond one-to-one with the previously open-sourced components for the NVIDIA platform. The report was written by Huang Xinyi, a reporter for the Science and Technology Innovation Board Daily.

At the Decoding China 2026 event, Huawei announced the introduction of the time compression concept of the "Tao (τ) Law" into base station product design for the first time, summarizing the challenges faced by traditional basebands within a single TTI as "endless calculations and incomplete transmissions."At the same time, Huawei released the OSCAR intelligent channel processing solution and the new generation UBBPi baseband series, providing a collaborative solution direction for the constraints of computation and transmission within a single TTI.

The semiconductor research organization SemiAnalysis has released the actual test results of Huawei's Ascend 950PR. The organization selected 1,055 sets of commonly used matrix calculations for large models. Huawei's own underlying program has allowed the 950PR to achieve 86% of its theoretical computing power, higher than NVIDIA's H100 at 75.5%; however, the actual computation speed is still only 43% of the H100, and only 21% compared to the relatively newer B300. This set of results has not yet utilized the newly ported operators from DeepSeek.DeepSeek is working on reducing data movement and waiting from the software side. In attention computation, after adjusting storage with FlashMLA, the two sets of data that originally needed to be read separately can now be retrieved at once; TileLang allows the chip to compute while preloading the next batch of data. SemiAnalysis measured that automatic scheduling has reduced the time for some small computing tasks by up to about 40%. DeepSeek has also ported DeepEP, which is responsible for MoE expert communication, but communication and computation will still compete for memory bandwidth, and Huawei plans to continue improving the firmware.Last year, DeepSeek attempted to train R2 using Huawei's Ascend chips, but even with Huawei engineers on-site for assistance, it could not be completed due to issues related to chip stability, inter-card communication, and software support. This year, both parties continue to collaborate, and by the end of September, DeepSeek focused on publicly releasing six sets of Ascend adaptation code. In October, it was reported that Huawei is adjusting the next generation of chip designs according to DeepSeek's model requirements. The 950PR tested this time is mainly aimed at inference input processing, while DeepSeek is preparing to use the 950DT for large-scale pre-training as early as the end of the year. Previously, both parties had conducted tests at the thousand-card level, but to enable tens of thousands of chips to complete a round of large model training continuously, more complex coordination and stability issues still need to be resolved.

According to Bloomberg's report on October 22, Huawei Technologies Co., Ltd. has secured a loan of 15 billion yuan (approximately 2.25 billion USD), amid a significant increase in chip spending for artificial intelligence-related projects. A knowledgeable source stated that this seven-year offshore yuan loan was provided by six Chinese lending institutions, including Bank of China, China Construction Bank, and Industrial and Commercial Bank of China, and the funds will be used for general corporate purposes. The source requested anonymity as the discussion pertains to non-public matters.According to Moneycontrol, which reprinted the content attributed to Bloomberg, this transaction illustrates the role of domestic lending institutions in Huawei's offshore financing since the United States imposed restrictions on Huawei in 2019. Before finalizing, Huawei had exploratory contacts with international banks. Huawei, Bank of China, China Construction Bank, and Industrial and Commercial Bank of China did not immediately respond to requests for comment.The report states that Huawei's research and development expenditure increased to 121 billion yuan in the first half of this year, while stockpiling due to soaring memory costs led to a negative cash flow of 39.9 billion yuan, eroding current profits. According to compiled data from Bloomberg, this is Huawei's first offshore syndicated loan since raising 11.8 billion Hong Kong dollars (approximately 1.5 billion USD) in 2020.

DeepSeek announced the full open-source of its internal inference framework DeepInfer, and simultaneously open-sourced the production-level token encoding and decoding component deepseek-recipe. DeepInfer integrates high-performance operators and a deeply optimized inference engine, supporting distributed inference, speculative decoding, SWA Bounded Replay, and structured output, compatible with NVIDIA and Huawei Ascend platforms.DeepSeek stated that it has implemented high throughput and low latency large-scale inference services for DeepSeek V4.1 Flash based on DeepInfer, and future inference code and optimizations for new models will continue to be updated here. deepseek-recipe is fully aligned with the processing logic of the DeepSeek official API, aiming to ensure that the community receives an experience consistent with the official deployment. The open-source repository and technical report have been released simultaneously.

According to The Information, DeepSeek is collaborating closely with Huawei to jointly design the next generation of AI chips, aiming to build models that do not rely on NVIDIA's computing power. Insiders revealed that Huawei engineers have entered DeepSeek's office to understand its model architecture and algorithm requirements, and to test how the chips handle data transmission and compression, with the first compatible chip targeting Ascend 950DT.The report pointed out that this move comes as China pushes to reduce its dependence on American chips. Limited by TSMC's foundry services and U.S. export controls, Huawei's Ascend 950DT is planned for release in 2027, but mass production may be delayed until 2028. To address the chip shortage, DeepSeek is using the TileLang programming language to bypass CUDA restrictions, while still attempting to acquire the banned NVIDIA Blackwell chips through third-party data centers. In addition, DeepSeek is forming a chip team and has begun testing clusters with over 1,000 processors.

On October 12, Huawei held a Kirin chip technology exchange meeting in Shanghai, marking the first chip-level technical communication meeting in six years, and systematically introduced the mass production progress of logic folding. Xu Songyu, head of the semiconductor business wireless terminal chip product management department, introduced that the Kirin chip has entered the testing phase in 2027, and the test results meet expectations; since 2023, the cumulative shipment of terminal products based on the Kirin chip has exceeded 100 million.The Kirin 9050 series is the first Kirin chip built on logic folding technology. Xu Songyu explained that logic folding is not simply stacking multiple chips, but rather a systematic design from a three-dimensional perspective at the beginning of the design, treating multiple dies as a complete chip for collaborative design. This series features over 50 million hybrid bonds, with effective signal interconnections exceeding 5 million, and a transistor density increase of 53.5%. The CPU large core efficiency of the Kirin 9050 Pro has improved by 41%, and the NPU computing power has increased by 140%; the current hybrid bonding pitch is 1.5 microns, and Huawei plans to further reduce it to 1 micron by 2027.The architecture of the Kirin 9050 Pro announced at the meeting includes the new generation Lingxi CPU, Maliang GPU, and Da Vinci architecture NPU, among others. This series is aimed at Mobile Agent, adopting a collaborative architecture of a large brain and a small brain on the edge, supporting the operation of a 30B MoE large model on the edge. The new generation Lingxi CPU features a 9-core 16-thread design, with a 23% improvement in multi-core performance, and the new generation Maliang GPU has a 40% performance increase.

Phoenix Technology cited a report from IDC seen by Reuters, stating that in the first half of 2026, NVIDIA's market share in China's data center AI chip market dropped to 34%. IDC data shows that from January to June this year, NVIDIA's AI accelerator shipments in the Chinese market were 678,000 units, a year-on-year decrease of 44%. The report attributed the change in market share to U.S. export control restrictions limiting product supply, as well as domestic competitors in China expanding their market share.Among the 11 Chinese chip manufacturers tracked by IDC, 9 experienced year-on-year growth in shipments. Huawei's AI accelerator shipments reached 572,000 units in the first half of the year, up from over 300,000 units in the same period last year, accounting for 28% of the overall market. Alibaba's chip design division, T-head, holds a market share of 7%, Cambricon has 5%, Baidu's Kunlun has 4%, and Haiguang Information has 3%.NVIDIA's CEO Jensen Huang has repeatedly criticized the U.S. government's export restrictions, stating that the policy forces NVIDIA to cede the $50 billion AI chip market in China to Huawei. IDC did not specify which products make up the 34% share, nor did it clarify whether it includes chips such as B200 and B300. As of the time of publication, NVIDIA had not responded to requests for comment.

Mr. Li Yi, a Douyin user, posted a video on social media stating that on June 10 of this year, he downloaded a fake wallet from the official Huawei store, which led to the exposure of his mnemonic phrase and the theft of 177,473 USDT.According to his description, the hacker changed the permissions of his wallet address to multi-signature, and then implemented a secondary scam under the pretext of unlocking it for spending money. After the victim transferred funds, all USDT in the address was transferred away.Mr. Li Yi claimed that after he released the video exposing the app, the download volume of the app in the official store plummeted. The hacker repeatedly gathered internet trolls to report his video, causing it to be forcibly taken down, and privately attempted to negotiate a conditional agreement to delete the video.He also stated that after conducting a private investigation, he learned that the scam app had five servers set up in Hong Kong, paying to have the app listed in the Huawei official store and to boost its ranking, causing a large number of victims to download it first in searches. Due to relevant bans in mainland China, related wallet apps cannot be normally launched in the store, so any app that cannot be launched but does exist in the store may be a counterfeit scam app.

Huawei announced that Huawei and Qualcomm have reached a multi-year, broad patent licensing agreement, which includes cross-licensing of patent portfolios in multiple fields such as 5G, computing, artificial intelligence, and networking, as well as Qualcomm's acquisition of some of Huawei's U.S. patents.The transaction will be completed upon obtaining the necessary regulatory approvals. The agreement reflects the two companies' shared commitment to intellectual property protection and adherence to licensing practices that are fair, reasonable, and non-discriminatory.

According to the South China Morning Post, former U.S. officials stated that Chinese semiconductor factories have stockpiled hundreds of immersion lithography machines worth billions of dollars in recent years and urged Washington and its allies to stop further exports. The Centre for Technology & Statecraft, established in August 2026, released a report in late September estimating that by early 2026, Chinese wafer fabs had acquired approximately 343 immersion deep ultraviolet lithography systems. The report was co-authored by former U.S. export control official Nicholas Brown and former White House and Department of Commerce technology policy advisor Saif Khan.The report estimates that about 270 of these are ASML Twinscan NXT:1980i. The authors claim that although these devices lag behind extreme ultraviolet lithography machines, they can be modified for the production of 7-nanometer logic chips and advanced memory required for artificial intelligence processors (including Huawei's Ascend series). Chinese wafer fabs spent over $13 billion in 2024 to purchase about 90 NXT:1980i units, and acquired another 89 units in 2025. ASML disclosed that, by destination, China accounted for 41% of its net system sales in 2024, 33% in 2025, dropping to 19% in the first quarter of 2026 and 14% in the second quarter.Since September 2023, the Netherlands has required export licenses for the NXT:2000i and more advanced immersion deep ultraviolet equipment, effectively banning these models from being shipped to China; a year later, the licensing requirements were extended to the NXT:1980i, but this model can still be approved for shipment on a case-by-case basis.

The artificial intelligence company DeepSeek has officially open-sourced infrastructure components for the Huawei Ascend computing platform, covering the TileLang high-level language compilation tool, computing libraries, and distributed communication libraries, corresponding to the previously open-sourced components for the NVIDIA platform. TileLang aims to provide a general-purpose, simpler programming language that can achieve the hardware performance limits, improving development efficiency and simplifying logic compared to CUDA, while its programming model can also leverage chip features.The TileLang route was first validated on the NVIDIA platform and has already supported the implementation of most operators in the training of the DeepSeek V4 series models. The open-sourced Ascend version encapsulates the underlying instructions of the Ascend C, providing a high-level programming approach without sacrificing hardware performance. Currently, every TileLang operator used in DeepSeek's training has a corresponding high-performance implementation on Ascend.The components open-sourced at the same time also include DeepGEMM, DeepEP, TileKernels, FlashMLA, and DeepSelect, which are used for general matrix operations, large-scale cross-device communication, conventional vector calculations and memory access, long-context sparse attention, and data filtering, respectively. DeepSeek claims that in multiple key tests, the related computing and communication performance has approached hardware limits; during the R&D process, the Huawei team provided support, and both parties collaborated to advance the 128-card supernode solution based on Ascend 950, with deep optimizations made for computing and communication.

Crypto KOL Kuai Dong (@_FORAB) posted that the DeepSeek investor meeting has once again experienced information leakage. Since the leak of internal speech recordings in May, the company has required attendees to fly to the Beijing or Hangzhou office, where they must hand over their phones, electronic devices, and backpacks in order to hear CEO Liang Wenfeng speak.Information leaked again just one day after last Sunday’s new meeting. Sources obtained by the information report reveal that DeepSeek's current important priority is to use more domestic chips and mentioned the earliest delivery time for Huawei chips.

According to The Information, citing informed sources, DeepSeek CEO Liang Wenfeng listed training large models with domestic chips such as those from Huawei as one of the company's biggest bets at the latest closed-door investor meeting, stating that this must succeed. DeepSeek expects to receive a new batch of Huawei chips for training as early as the fourth quarter of this year, although the specific models and quantities have not been disclosed.Liang Wenfeng believes that Huawei chips are expected to catch up with Nvidia in just a few years and hopes that DeepSeek will adapt to hardware beyond Nvidia in advance. DeepSeek is still using Nvidia chips to train models. In May of this year, Liang Wenfeng estimated that training the top-tier large models would require about 50,000 Nvidia GB300 chips, while switching to Huawei Ascend 950 would require about 200,000 chips, with the biggest issue at that time being insufficient supply from Huawei.

The Bank of England has maintained the policy interest rate at 3.75% for the sixth consecutive meeting and announced the cancellation of the long-term government bond sale plan.Contemporary Amperex Technology Co., Ltd. plans to invest 1.4 billion yuan to build a new lithium carbonate project in Ya'an with an annual production capacity of 40,000 tons, while Huawei simultaneously released a new computing architecture.

Huawei's rotating chairman Wang Tao stated at the Huawei Connect Conference 2026 that the Ascend 960 chip is ready ahead of schedule, achieving a doubling of performance. Among them, the Ascend 960DT will be released in Q1 2027, three quarters earlier than originally planned; the Ascend 960PR will be released in Q3 2027, one quarter earlier than originally planned.Wang Tao mentioned that they will adhere to an annual evolution pace, with the Ascend 970 planned for release in 2028 and the Ascend 980 planned for release in 2029.

Security researcher Shou Chaofan stated that he just purchased about 6TB of Fable model invocation data from a leading Chinese model relay station, which included sensitive credentials such as SSH keys, VPN configurations, Alibaba Cloud keys, and GitLab tokens. He claimed that the keys in this batch of data were sufficient for him to access the servers or internal systems of 19 leading Chinese companies, as well as 7 Chinese and CIS government-related agencies, including Huawei, Xiaomi, NIO, and MiniMax.The model relay station is positioned between users and models like Claude, so requests and responses pass through it first, allowing it to see the complete plaintext. Once developers insert SSH Keys, API Keys, VPN configurations, etc., into the Agent context, if the relay station saves or even sells these records, the company's system keys will also leak out. This is not the first time Shou Chaofan has warned about the risks of relay stations. A paper he participated in tested 428 LLM relay stations in April and found that 9 actively injected malicious code, 17 actually used AWS test keys deliberately placed by researchers to call AWS, and 1 directly transferred ETH from the test wallet.Shou Chaofan is a co-founder of the blockchain security company Fuzzland and has long been engaged in vulnerability and supply chain security research. At the end of March, he also first discovered that the source map in the Claude Code 2.1.88 release package accidentally exposed about 500,000 lines of TypeScript source code.