Huawei unveils Ascend 960 supernode with NPO, accelerating AI chip roadmap to 2027
At Huawei Connect 2026 in Shanghai, Huawei unveiled the Ascend 960 supernode, the industry's first using near-packaging optics (NPO) technology, and announced its Ascend 960 chip is ready ahead of schedule. The Ascend 960DT will launch in Q1 2027, three quarters early, and the 960PR in Q3 2027. The supernode supports up to 4,096 cards, delivering 8 EFLOPS FP8 compute and 1 PB HBM capacity, targeting trillion-parameter model training. Huawei also reported over 1,000 Ascend supernode deployments globally.
Reference imageEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Common ground
- Huawei has delivered chips like the 910B and 920C on schedule despite severe US export controls, showing real capability under pressure.
- The NPO supernode architecture is a genuine engineering achievement, improving system-level efficiency for large AI models.
- China is building an independent AI computing ecosystem, including chips, interconnects, and software, to reduce reliance on US supply chains.
- The software ecosystem gap between CUDA and CANN is real, with CUDA having a decade of maturity and optimization tools.
Points of contention
- Whether the Ascend 960's 'doubling performance' claim is meaningful without a clear baseline or real benchmarks.
- If Huawei's progress is a 'paradigm shift' and 'leapfrog' or just catching up under duress, given the 920C is roughly A100-class from 2020.
- Whether the 2027 roadmap is a credible technical plan or a strategic narrative to reassure investors and partners.
- If the market rally in A-share stocks reflects genuine structural change or just sentiment based on announcements.
Blind spots
- Neither side provided independent benchmarks for cost per token or energy per inference at scale for the NPO supernode.
- The debate lacked a detailed comparison of how CANN's open-source ecosystem will evolve to match CUDA's toolchain maturity.
- There was no discussion of potential supply chain bottlenecks for Huawei's future chips, like advanced lithography or memory.
WorldAttention’s read
The debate shows a clear split: the Eastern Agent sees Huawei's accelerated roadmap and system-level innovations as a strategic defeat for US export controls, building sovereign AI infrastructure under extreme pressure. The Neutral Agent respects Huawei's delivery record but insists that without silicon samples and real benchmarks, claims of a 'paradigm shift' are premature—calling it a well-executed narrative rather than a proven breakthrough. Both agree on the geopolitical significance and the software ecosystem gap, but disagree on whether this is leapfrogging or just catching up. The blind spots include a lack of independent performance data and a deeper look at future supply chain risks.
Reporting timeline
Huawei launches world's first NPO-based supernode for AI compute, Chinese firms accelerate ecosystem
Huawei has unveiled the Ascend 960 supernode, the world's first to use near-packaging optics (NPO) technology, at its annual Connect conference. The supernode is designed to address communication bottlenecks in large-scale AI model training by enabling thousands of chips to function as a single logical computer. Huawei executive Wang Tao stated that in a 100,000-card cluster, traditional 8-card server architectures can waste nearly 40% of time on inter-chip communication, while the new supernode architecture can boost model floating-point utilization (MFU) by 2.75 times. The article notes that Chinese AI chip firms including BirenTech, Moore Threads, and MetaX have also released supernode products, signaling industry consensus. Huawei reports over 1,000 Ascend supernode deployments globally. However, challenges remain: domestic GPU utilization is below 30%, and a structural imbalance exists between high-end demand and mid-to-low-end supply. Chinese Academy of Engineering academician Zheng Weimin called for better software to utilize existing mid-to-low-end compute resources. The article positions these developments as part of a broader Chinese strategy to overcome hardware restrictions through system-level innovation and open-source software unification.
Read sourceHuawei's Wang Tao Unveils Ascend 960 Supernode with NPO Engine for Trillion-Parameter AI Models
At the Huawei Connect 2026 conference in Shanghai, Huawei Deputy Chairman and Rotating Chairman Wang Tao delivered a keynote speech outlining the company's AI strategy, centered on hardware monetization and building a robust computing base via 'supernodes and clusters.' He announced the Ascend 960 supernode, the industry's first to use a Near-Packaged Optics (NPO) optical engine (Hi-ONE), which integrates 36 200G channels for a 7.2T capacity per engine. The supernode supports up to 4,096 cards, provides 8E FP8 compute power and 1PB HBM capacity, and is scheduled for release in mid-2027. Wang Tao argued that NPO offers better reliability and manufacturability than CPO. He stated that a 10,000-card cluster of 4K-scale supernodes could achieve 2.75x the Model FLOPS Utilization (MFU) of traditional 8-card server clusters, crucial for training trillion-parameter models. The speech also highlighted the Ascend ecosystem's growth, with CANN now open-source and Ascend becoming a natively installable platform on PyTorch. Huawei positions itself as a provider of a 'silicon black soil' computing base, offering an alternative in the global AI infrastructure market.
Read sourceChina Merchants Securities: Huawei Ascend 960 Chip Development Accelerated, NPO Technology Exceeds Expectations
A research report from China Merchants Securities, published via Zhitong Finance, provides an in-depth analysis of China's domestic AI computing chip supply chain, focusing on Huawei's latest developments. Key findings include that the Ascend 960 chip's design tape-out (DT) has been accelerated from Q4 2027 to Q1 2027, with the production release (PR) also moved forward to Q3 2027. The report highlights the new 'Lingqu' series of 11 chip families covering computing, connectivity, storage, and management. Notably, the Ascend 960 supernode will be the first product to adopt NPO (Narrowband Pluggable Optics) technology, using 5,500 Hi-ONE optical engines to replace 48,000 traditional 800G optical modules, significantly reducing power consumption. The report also details plans for a 4096-card supernode and a million-card cluster, and forecasts future chips: Ascend 970 (2028, 3.6 PFLOPS FP8) and Ascend 980 (2029, 7.2 PFLOPS FP8). The report recommends investment opportunities across the AI computing chain, including foundry, GPU/ASIC, networking chips, optical chips, packaging, testing, equipment, components, and storage. Risks cited include weaker-than-expected end demand, macroeconomic impacts, and slower semiconductor localization progress.
Show 17 older updatesHide older updates
Huawei Unveils New Super Node, Shifting AI Compute Competition to System Level
According to a research report from China Merchants Securities, Huawei has released the industry's first super node using NPO (Near-Packaged Optics) technology, the Ascend 960, and upgraded its Kunpeng super node at the Huawei Connect 2026 conference in Shanghai. The report states that Huawei's AI strategy centers on computing power, aiming to accelerate training and inference for large-scale models. The Ascend 960 super node, built on Huawei's Lingqu interconnect and the Hi-ONE NPO optical engine, shifts the compute competition from single cards to a system level and has entered large-scale commercial deployment. The report highlights that the super node uses 5,500 Hi-ONE units to replace 48,000 800G optical modules, reducing power consumption by over 550 kilowatts and achieving 99.8% system availability. The report recommends focusing on supply chain stocks and seizing opportunities from the volume production of domestic computing clusters, noting that the technology drives growth in high-speed interconnect, optical engines, and liquid cooling sectors.
Read sourceHuawei Unveils Ascend 960 Supernode with Optical Interconnect, Outlines AI Computing Blueprint
At the Huawei Connect 2026 conference in Shanghai, Huawei disclosed its strategy for AI infrastructure, centered on 'supernode + optical interconnect' as the computing foundation for an intelligent world. Huawei Deputy Chairman and Rotating Chairman Wang Tao announced the Ascend 960 supernode, the first to use a Near-Packaged Optics (NPO) optical engine, integrating 5,500 Hi-ONE optical interconnect products to replace 48,000 800G optical modules, reducing power consumption by over 550 kW and doubling system uptime to 99.8% availability. The Ascend 960 supports up to 4,096 cards via the Lingqu interconnect protocol, targeting models with tens of trillions of parameters. Huawei also introduced the 'AI landing seven-step method' and the DIMAK engineering system to help enterprises integrate AI into core production processes. The company has deployed over 1,000 sets of the Ascend 384 supernode across industries including internet, telecom, finance, and healthcare, and has partnered with over 3,000 industry partners to develop 7,000+ solutions.
Read sourceHuawei Announces Ascend 960 Supernode at Connect 2026; CITIC Securities Bullish on NPO Supply Chain
At the Huawei Connect conference held in Shanghai from September 17-19, 2026, Huawei rotating chairman Wang Tao stated that supernodes are the mainstream computing infrastructure for the Agentic AI era. Huawei announced the Ascend 960 chip is ready ahead of schedule with doubled performance: the 960DT (4 PFLOPS FP4) launches in Q1 2027, and the 960PR (8 PFLOPS FP4) in Q3 2027. The company unveiled the industry's first NPO-based supernode, the Ascend 960 Supernode (Atlas 960 Liquid-Cooled), planned for Q3 2027, offering 8 EFLOPS FP8 and 16 EFLOPS FP4 compute power with 2-microsecond latency. It uses 5,500 Hi-ONE NPO units replacing 48,000 800G optical modules, reducing power consumption by over 550 kilowatts. CITIC Securities recommends paying attention to investment opportunities in Huawei's optical interconnect supply chain, citing NPO as a solution to copper interconnect limitations in AI clusters. Risk factors include slower-than-expected AI development, industry competition, and geopolitical risks.
Read sourceHuawei Unveils Ascend 960 Supernode with NPO, Accelerates Chip Roadmap
At the 2026 Huawei Connect conference in Shanghai (September 17-19), Huawei announced that supernodes have become the mainstream computing infrastructure for the Agentic AI era, with the Ascend chip series as the foundation. The Ascend 950 supernode is already in large-scale commercial use. Huawei unveiled the industry's first supernode using NPO (Near-Packaged Optics), the 'Ascend 960 Supernode,' scheduled for Q3 2027. The Ascend 960DT chip will launch in Q1 2027 (three quarters ahead of schedule) with 4 PFLOPS (FP4), and the Ascend 960PR in Q3 2027 (one quarter ahead) with 8 PFLOPS (FP4). Future plans include the Ascend 970 (14 PFLOPS FP4) in 2028 and Ascend 980 (28 PFLOPS FP4) in 2029. The 960 supernode uses 5,500 Hi-ONE NPO products, replacing 48,000 800G optical modules, reducing power consumption by over 550 kW and doubling system uptime. CITIC Securities views NPO as a key solution to copper interconnect limitations and recommends focusing on Huawei's optical interconnect supply chain investment opportunities.
Read sourceHuawei unveils Ascend 960 SuperNode with NPO, targets AI computing efficiency leap
At Huawei Connect 2026, Huawei announced the Ascend 960 SuperNode, a next-generation AI computing infrastructure supporting up to 4,096 Ascend cards. It introduces Near-Packaged Optics (NPO) technology for the first time, positioning optoelectronic conversion closer to chips to reduce transmission loss and power consumption. The Ascend 960 chip roadmap includes the 960DT in Q1 2027 and 960PR in Q3 2027, followed by Ascend 970 and 980. Huawei claims a hundred-thousand-card cluster using 4K SuperNodes can improve model computing resource utilization by 2.75 times versus traditional 8-card server solutions. The article analyzes implications for domestic tech sectors: a clearer path for domestic computing power via system-level innovation, new technical demands on optical communication chains (with 5,500 NPO engines replacing 48,000 800G modules), and evolution of AI infrastructure toward complete computing systems. Key segments to watch include AI chips, servers, optical communications, high-speed PCBs, connectors, liquid cooling, and data center infrastructure.
Read sourceJinyuan Securities: Huawei Ascend 960 AI Chip Doubles Performance, NPO Super Node Deployed
Jinyuan Securities released an electronics industry commentary report on Huawei Connect 2026. The report notes that Huawei launched the next-generation Ascend 960 AI chip, which delivers a 100% increase in AI computing power and a 40% improvement in energy efficiency over the previous generation. Additionally, Huawei achieved the first large-scale commercial deployment of Near-Packaged Optics (NPO) super node technology, marking the entry of optical interconnects into the industrialization phase within the AI server sector. The report suggests that breakthroughs in the Ascend 960 and NPO technologies will accelerate the development of domestic AI chips and high-speed interconnect ecosystems, benefiting companies involved in China's AI computing infrastructure, optical modules, and advanced packaging supply chains.
Huawei Unveils Next-Gen AI Computing Base with NPO Technology for Large Model Training
At the Huawei Connect conference in Shanghai on September 17, Huawei released its next-generation AI computing infrastructure, the Ascend 960 supernode. The system supports up to 4,096 computing cards, up from 1,024 in the previous generation, and debuts NPO (near-package optics) technology to address communication bottlenecks in large-scale clusters. The Ascend 960 is designed to meet training and inference demands for models with up to 10 trillion parameters. Huawei Executive Director Wang Tao stated that the Ascend 960 chip development exceeded expectations, with air-cooled and liquid-cooled versions launching in Q2 and Q3 2025 respectively. Huawei's Hi-ONE optical interconnect product, the industry's first mass-produced NPO product with 7.2T capacity, replaces 48,000 800G optical modules with 5,500 Hi-ONE modules, reducing power consumption by over 550 kW and improving system availability to 99.8%. Analyst Zhu Keli noted the infrastructure provides a domestic base for large model training and upgrades upstream and downstream industries, while CTO Miao Fuyou highlighted China's full industrial chain advantage in AI infrastructure, particularly in connection technology.
Huawei unveils Ascend 960 AI computing cluster with NPO optical interconnect technology
At the Huawei Connect conference in Shanghai on September 17, Huawei released its next-generation AI computing infrastructure, the Ascend 960 supernode. The system supports up to 4,096 compute cards, up from 1,024 in the previous generation, and debuts NPO (near-package optics) technology, a new optical interconnect product called Hi-ONE. Huawei claims the Hi-ONE module, the industry's first mass-produced NPO product with 7.2T capacity and built-in light source, replaces 48,000 800G optical modules with 5,500 Hi-ONE modules, reducing power consumption by over 550 kW and doubling system uptime to 99.8% availability. The Ascend 960 is designed to match training and inference needs for 10-trillion-parameter models. Huawei rotating chairman Wang Tao stated the company's AI strategy centers on computing power, hardware monetization, and open-source ecosystem building. The Ascend 910C supernode has been deployed in over 1,000 units, and the Ascend 950 is commercially available. Experts quoted include Zhu Keli, who noted the system provides a domestic foundation for large model training and drives upstream industries like optical communications and liquid cooling, and Miao Fuyou of the Global Computing Alliance, who highlighted China's full industrial chain advantage in AI infrastructure, particularly in connectivity technology.
Huawei Unveils Ascend 960 AI Supernode With NPO Technology for 10-Trillion-Parameter Models
At the Huawei Connect conference in Shanghai on September 17, Huawei unveiled its next-generation AI computing infrastructure, the Ascend 960 supernode. The system supports up to 4,096 compute cards, up from 1,024 in the previous generation, and debuts a key technology called NPO (near-package optics) to match the training and inference needs of 10-trillion-parameter models. Huawei's rotating chairman Wang Tao stated that the Ascend 960 chip development exceeded expectations, with air-cooled and liquid-cooled versions launching in Q2 and Q3 2025 respectively. The system uses 5,500 Hi-ONE optical interconnect modules, replacing 48,000 800G optical modules, reducing power consumption by over 550 kW and doubling system uptime to 99.8% availability. Experts quoted include Zhu Keli, founding dean of the National Research Institute of New Economy, who said the infrastructure provides a domestic base for large model training and helps reduce fragmentation in China's computing ecosystem, and Miao Fuyou, CTO of the Global Computing Consortium, who highlighted China's advantage in full-industry-chain capabilities and connection technology.
Read sourceHuawei Unveils Next-Gen AI Computing Base with NPO Technology for 10-Trillion-Parameter Models
At the Huawei Connect conference in Shanghai on September 17, Huawei released its next-generation AI computing infrastructure, the Ascend 960 supernode. The system supports up to 4,096 compute cards, up from 1,024 in the previous generation, and debuts a key technology called NPO (Near-Package Optics) to meet training and inference demands for 10-trillion-parameter models. Huawei's rotating chairman Wang Tao stated the company's AI strategy focuses on computing power, hardware monetization, and open-source ecosystem development. The Ascend 960 air-cooled and liquid-cooled versions are scheduled for release in Q2 and Q3 2025 respectively. Huawei's Hi-ONE NPO product, the industry's first mass-produced NPO, uses 5,500 modules to replace 48,000 800G optical modules, reducing power consumption by over 550 kW and achieving 99.8% system availability. Analyst Zhu Keli noted the announcement provides a domestic base for large model training and drives upgrades in optical communications and liquid cooling industries. Global Computing Alliance CTO Miao Fuyou highlighted China's full industrial chain advantage in AI infrastructure, particularly in connectivity technology.
Huawei unveils NPO-based supernode; Nvidia CEO flags supply bottleneck; AI chip stocks rally
On September 18, A-share AI chip stocks rallied, driven by Huawei's announcement of the Ascend 960 supernode, the industry's first using near-package optics (NPO) technology, at the Huawei Connect 2026 conference. The supernode supports up to 4,096 cards, 8 EFLOPS FP8 and 16 EFLOPS FP4 compute, and 1 PB HBM capacity for trillion-parameter model training. Huawei's rotating chairman Wang Tao said the Ascend 960DT chip will launch in Q1 2027, three quarters ahead of schedule, with annual upgrades planned through 2029. Separately, Nvidia CEO Jensen Huang stated the current bottleneck is production capacity, not demand, and Nvidia forecast ~70% revenue growth for fiscal year ending January 2028. Policy support from China's MIIT and NDRC also boosted sentiment. The article lists 39 A-share AI chip concept stocks with a combined market cap of ~2.76 trillion yuan, noting that 14 stocks saw net margin buying in the past week. Analysts forecast net profit for Fudan Microelectronics to surge 4.3x and Cambricon to grow over 1.7x in 2025.
Read sourceHuawei Unveils Ascend 960 Supernode; AI Chip Stocks Surge on Policy and Demand Optimism
On September 18, A-share indices rose, reigniting the computing hardware sector, with AI chip stocks like Fudan Microelectronics and Hygon Information gaining attention. At the Huawei Connect 2026 conference on September 17, Huawei released the Ascend 960 supernode, the world's first to adopt Near-Packaged Optics (NPO) technology, featuring FP8 computing power of 8 EFLOPS and HBM capacity up to 1 PB. Huawei Vice Chairman Wang Tao stated the Ascend 960DT will launch in Q1 2027, three quarters early, with annual releases planned through 2029. Separately, NVIDIA CEO Jensen Huang noted the bottleneck is NVIDIA's chip production capacity, not demand; NVIDIA projected ~70% revenue growth for fiscal year ending January 2028, reaching about $673 billion. China's MIIT and NDRC issued guidelines to accelerate AI chip and cloud training equipment development. Bank of America Securities raised its 2030 global semiconductor forecast. According to East Money, 39 A-share AI chip concept stocks have a combined market cap of ~2.76 trillion yuan. Institutions forecast Fudan Microelectronics' net profit may surge 4.3x year-on-year and Cambricon's by over 1.7x this year.
Read sourceHuawei Announces Ascend 960 AI Chip Launching Early With Doubled Performance
At Huawei Connect 2026 on September 17, Huawei's rotating chairman Wang Tao outlined the company's AI strategy, centered on computing power through hardware monetization and an open-source ecosystem. He officially launched the Ascend 960 super node, described as the world's first to apply NPO technology, capable of supporting models with up to 10 trillion parameters. Wang Tao stated that over 1,000 Ascend super nodes have been deployed and the Ascend 950 super node is in commercial deployment. Regarding AI chips, he said the Ascend 960 chip is ready ahead of schedule with doubled performance. The Ascend 960DT is scheduled for release in Q1 2027, three quarters earlier than planned, and the Ascend 960PR in Q3 2027, one quarter earlier. Huawei plans to release a new generation annually, with the Ascend 970 in 2028 and Ascend 980 in 2029.
Read sourceHuawei's Wang Tao Says Ascend 960 Chip to Launch in Q1 2027, Ahead of Schedule
At Huawei Connect 2026, Huawei's rotating chairman Wang Tao announced that the Ascend 960 chip is ready ahead of schedule, delivering a doubling in performance. The Ascend 960DT will be released in Q1 2027, three quarters earlier than originally planned, while the Ascend 960PR will follow in Q3 2027, one quarter ahead of schedule. Wang Tao also outlined Huawei's future roadmap, stating the company will maintain an annual release cadence, with the Ascend 970 scheduled for 2028 and the Ascend 980 for 2029. The announcement was reported by Jin10 Data on September 17.
Read sourceHuawei's Wang Tao Announces Ascend 960 Chip Launch Moved Up to Early 2027
At Huawei Connect 2026, Huawei Rotating Chairman Wang Tao announced that the Ascend 960 chip is ready ahead of schedule with a doubling in performance. The Ascend 960DT will be released in Q1 2027, three quarters earlier than originally planned, and the Ascend 960PR will follow in Q3 2027, one quarter earlier than planned. Wang Tao stated that Huawei will maintain an annual evolution cadence, with the Ascend 970 scheduled for 2028 and the Ascend 980 for 2029. The report was published by the STAR Market Daily on the 17th.
Read sourceHuawei Rotating Chairman Wang Tao Announces 960DT Product and Ascend 960PR Chip Launch Timelines
Huawei Rotating Chairman Wang Tao has announced that the company will launch the 960DT product in the first quarter of 2027, followed by the Ascend 960PR chip in the third quarter of 2027. The statement, reported by financial data platform Jin10, provides a clear product roadmap for Huawei's hardware offerings. The Ascend 960PR chip is part of Huawei's Ascend series, which is central to the company's AI computing strategy. The announcement outlines specific launch timelines for these products, indicating Huawei's planned advancements in its technology portfolio over the next several years. No further details on the specifications or applications of the 960DT product or the Ascend 960PR chip were provided in the source item.
Huawei Unveils Ascend 960 Supernode, Advances AI Chip Launch Schedule at Connect 2026
At the Huawei Connect 2026 conference in Shanghai, Huawei unveiled the industry's first Ascend 960 supernode using Near-Packaged Optics (NPO) technology, upgraded its Kunpeng supernodes, and launched the OceanStor M900 AI memory storage system. Wang Tao, Huawei's Vice Chairman and Rotating Chairman, stated that computing power is central to Huawei's AI strategy and that the pace of AI transformation exceeds any previous technological revolution. He noted that large model parameters are approaching 10 trillion and are expected to exceed 100 trillion by 2030. Huawei announced that the Ascend 960 chip's R&D progress has exceeded expectations, achieving doubled performance. The Ascend 960DT will launch in Q1 2027, three quarters ahead of schedule, and the Ascend 960PR in Q3 2027, one quarter ahead. Huawei plans to maintain a pace of one generation per year, with the Ascend 970 in 2028 and Ascend 980 in 2029. The Ascend 960 supernode features a 4,096-card scale, delivering up to 8 EFLOPS of FP8 computing power and 1 PB of HBM capacity. Huawei also introduced Agentic supernode clusters capable of supporting up to one million cards using the Lingqu UnifiedBus for unified interconnectivity.
Read source