Huawei launches CloudMatrix 384 AI System to rival Nvidia’s most advanced AI system
On Saturday, Huawei Technologies displayed an advanced AI computing system in China, as the Chinese technology giant seeks to capture market share in the country’s growing artificial intelligence sector. Huawei’s CloudMatrix 384 system made its first public debut at the World Artificial Intelligence Conference (WAIC), a three-day event in Shanghai where companies showcase their latest AI innovations, drawing a large crowd to the company’s booth.
The Huawei CloudMatrix 384 is a high-density AI computing system featuring 384 Huawei Ascend 910C chips, designed to rival Nvidia’s GB200 NVL72 (more below). The AI system employs a “supernode” architecture with high-speed internal chip interconnects. The system is built with optical links for low-latency, high-bandwidth communication. Huawei has also integrated the CloudMatrix 384 into its cloud platform. The system has drawn close attention from the global AI community since Huawei first announced it in April.
The CloudMatrix 384 resides on the super-node Ascend platform and uses high-speed bus interconnection capability, resulting in low latency linkage between 384 Ascend NPUs. Huawei says that “compared to traditional AI clusters that often stack servers, storage, network technology, and other resources, Huawei CloudMatrix has a super-organized setup. As a result, it also reduces the chance of facing failures at times of large-scale training.
Huawei staff at its WAIC booth declined to comment when asked to introduce the CloudMatrix 384 system. A spokesperson for Huawei did not respond to questions. However, Huawei says that “early reports revealed that the CloudMatrix 384 can offer 300 PFLOPs of dense BF16 computing. That’s double of Nvidia GB200 NVL72 AI tech system. It also excels in terms of memory capacity (3.6x) and bandwidth (2.1x).” Indeed, industry analysts view the CloudMatrix 384 as a direct competitor to Nvidia’s GB200 NVL72, the U.S. GPU chipmaker’s most advanced system-level product currently available in the market.
One industry expert has said the CloudMatrix 384 system rivals Nvidia’s most advanced offerings. Dylan Patel, founder of semiconductor research group SemiAnalysis, said in an April article that Huawei now had AI system capabilities that could beat Nvidia’s AI system. The CloudMatrix 384 incorporates 384 of Huawei’s latest 910C chips and outperforms Nvidia’s GB200 NVL72 on some metrics, which uses 72 B200 chips, according to SemiAnalysis. The performance stems from Huawei’s system design capabilities, which compensate for weaker individual chip performance through the use of more chips and system-level innovations, SemiAnalysis said.
Huawei has become widely regarded as China’s most promising domestic supplier of chips essential for AI development, even though the company faces U.S. export restrictions. Nvidia CEO Jensen Huang told Bloomberg in May that Huawei had been “moving quite fast” and named the CloudMatrix as an example.
Huawei says the system uses “supernode” architecture that allows the chips to interconnect at super-high speeds and in June, Huawei Cloud CEO Zhang Pingan said the CloudMatrix 384 system was operational on Huawei’s cloud platform.
According to Huawei, the Ascend AI chip-based CloudMatrix 384 with three important benefits:
- Ultra-large bandwidth
- Ultra-Low Latency
- Ultra-Strong Performance
These three perks can help enterprises achieve better AI training as well as stable reasoning performance for models. They could further retain long-term reliability.
References:
https://www.huaweicentral.com/huawei-launches-cloudmatrix-384-ai-chip-cluster-against-nvidia-nvl72/
https://semianalysis.com/2025/04/16/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72/
U.S. export controls on Nvidia H20 AI chips enables Huawei’s 910C GPU to be favored by AI tech giants in China
Huawei’s “FOUR NEW strategy” for carriers to be successful in AI era
FT: Nvidia invested $1bn in AI start-ups in 2024
3 thoughts on “Huawei launches CloudMatrix 384 AI System to rival Nvidia’s most advanced AI system”
Leave a Reply to Alan Weissberger Cancel Reply
Google Gen AI:
While Nvidia’s B200 chip significantly outperforms Huawei’s Ascend 910C at the individual chip level, Huawei’s strength lies in its system-level design.
The CloudMatrix 384 delivers approximately 300 PFLOPs of dense BF16 compute performance, almost double the 180 PFLOPs of Nvidia’s GB200 NVL72.
Huawei’s system boasts greater memory capacity (3.6x more) and higher memory bandwidth (2.1x more) compared to Nvidia’s GB200 NVL72.
The CloudMatrix 384 consumes significantly more power than the Nvidia GB200 NVL72 (559 kW vs. 145 kW). This translates to lower power efficiency per FLOP for Huawei’s system.
The refusal of European government administrations to evict Huawei frustrates the bosses of Ericsson and Nokia, the Nordic vendors that would most obviously benefit from it. Börje Ekholm, Ericsson’s CEO, was critical of the Swedish government’s decision to ban Huawei from its 5G network in late 2020. That was probably because he feared Chinese reprisals would hurt Ericsson, whose China revenues fell 46% in 2021. More recently, however, Ekholm has grumbled about the probability of losing market share in some parts of the world to aggressive Chinese rivals. Earlier this month, he sounded doubtful that new opportunities to replace high-risk vendors in Europe would ever materialize.
But Justin Hotard, who succeeded Pekka Lundmark as Nokia’s CEO in April, struck a more optimistic tone last week. “I think that’s significant,” he said on a call with reporters, referring to Virkkunen’s desire to clamp down on high-risk vendors. “We see that as an opportunity for us given our market position.”
Nokia did, notably, land a contract last November to replace Huawei at about 3,000 of Deutsche Telekom’s 5G sites in Germany – roughly 8% of the RAN total, according to data the operator included in its capital markets day presentations last year. The deal was a long time in the making and is curious given that Deutsche Telekom opposes efforts to evict Huawei. But the introduction of a third RAN vendor could seem prudent. German authorities, under mounting domestic, EU and US pressure, might always decide on a tougher approach, as the UK previously did.
The longer they and other countries resist, the worse things could be for Ericsson and Nokia. RAN sales flattened year-over-year in the first quarter of 2025, after steep drops in 2023 and 2024 that lowered annual revenues last year by more than a fifth, to about $35 billion, compared with the figure in 2022, according to the analyst company Omdia. This year, Omdia expects “low single-digit percentage growth.” But market share gains would put the Nordic companies in a much healthier position, and meaningful gains could only come at Huawei’s expense, unless Ericsson and Nokia were trampling over each other’s turf.
Job cuts shrank the size of the Nokia workforce from about 86,700 employees in 2023 to 75,600 last December, and the axe has fallen heavily on mobile. In a restructuring that harks back to an earlier era and could precipitate further cuts, Nokia is moving several corporate functions outside individual units and into a group-wide layer. Additional cuts within mobile would now be hard to realize without hurting critical research and development activities.
Ericsson, while in stronger RAN shape, is also fighting market stagnation made worse by US tariffs and the weakness of the US dollar. It looks unhealthily reliant on American business, generating 44% of its second quarter revenues in the US. Work in India has seemingly dried up, following a frenzied deployment of 5G networks, and China revenues fell by 37% year-over-year for the recent second quarter, to about 2.2 billion Swedish kronor ($230 million). To protect margins, Ericsson cut about 6,000 jobs last year, or 6% of its total workforce.
Telcos, which have not profited from 5G, are in no mood for early Huawei swap-outs that would gobble funds. The biggest opportunity for other vendors seemed to be a Vodafone RAN tender encompassing 170,000 base station sites across Europe and Africa, according to the operator’s annual report last year. But Vodafone’s sale of businesses in Hungary, Italy and Spain, and the exclusion from the tender of joint network ventures in the UK and the Netherlands, are likely to have dramatically lowered that number.
Whether or not Huawei deserves its EU and US blacklisting, the danger is that Europe’s continued reliance on it further weakens Nokia’s mobile networks business. Added to Europe’s other negatives, it could ultimately persuade Ericsson to follow the journey of the similarly named Leif Erikson, the Viking thought to have sailed to America more than 1,000 years ago. A relocation of headquarters from Sweden to the US was threatened by Ekholm last year. For a region increasingly short of big companies with global influence, that would surely be a disaster.
https://www.lightreading.com/5g/europe-s-love-of-huawei-is-crushing-ericsson-and-nokia
China is ramping up efforts to build a domestic artificial-intelligence ecosystem that can function without Western technology, as it steels itself for a protracted tech contest with the U.S. Many of China’s home grown AI initiatives were on display at an AI conference that ended this week in Shanghai, which Chinese authorities used as a showcase for products free of U.S. technologies.
Shanghai-based StepFun, touted a new AI model that it said required less computing power and memory than other systems, making it more compatible with Chinese-made semiconductors. Although Chinese chips are less capable than American products, Huawei Technologies and other companies have been narrowing the gap by clustering more chips together, boosting their performance.
China also released an AI global governance plan at the event, the World Artificial Intelligence Conference, which called for establishing an international open-source community through which AI models can be freely deployed and improved by users. Industry participants say it showed China’s ambition to set global standards for AI and could undermine the U.S., whose leading models aren’t open-source.
The conference followed a series of announcements and investments in China aimed at turbocharging its AI capabilities, including rapid expansions in power generation and skills training. The whole-nation effort, led by Beijing, includes billions of dollars in spending by state-owned enterprises, private companies and local governments.
China’s securities regulator has largely refrained from approving domestic initial public offerings of companies whose businesses aren’t related to “hard tech” sectors such as semiconductors and AI, so that capital can concentrate on funding technologies of strategic importance, people involved in the process said.
AI is expected to upend economies and militaries, and leadership in the sector is seen as critical to future global influence and national security.
China also clearly wants U.S. technologies. Access to advanced chips has been a priority for Chinese negotiators in trade talks, The Wall Street Journal has reported. Sales of Nvidia’s H20 AI chips to China were recently restored by Washington after it restricted them in April, a reversal seen by Beijing as a gesture of good faith in the talks.
The Trump administration is taking steps to try to preserve the U.S.’s lead. It recently announced an AI “action plan,” which aims to slash red tape and make it easier for tech companies to build data centers needed to train AI models.
Earlier this year at the White House, OpenAI and Japan’s SoftBank unveiled a $500 billion effort to build new AI data centers, though the project has faced delays.
China, though, has shown a willingness to spend whatever it takes. The rising popularity of DeepSeek, the Chinese AI startup, has buoyed Beijing’s hopes that it can become more self-sufficient. Huawei has published several papers this year detailing how its researchers used its homegrown chips to build large language models without relying on American technology.
“China is obviously making progress in hardening its AI and computing ecosystem,” said Michael Frank, founder of think tank Seldon Strategies.
China’s biggest AI challenge is overcoming its difficulty in sourcing the world’s most-advanced chips. Washington has cut China off from some of Nvidia’s most sophisticated semiconductors, as well as the advanced machine tools used to make cutting-edge chips, restrictions that many experts believe will continue to hold China back.
Huawei is helping spearhead efforts to navigate those restrictions. During a meeting with President Xi Jinping in February, Chief Executive Officer Ren Zhengfei told Xi about “Project Spare Tire,” an effort by Huawei and 2,000 other enterprises to help China’s semiconductor sector achieve a self-sufficiency rate of 70% by 2028, according to people familiar with the meeting.
Increasingly, the company has been able to bundle together the best chips it can produce to match the performance of some American computing systems. That is helping local companies reach some of the same computing goals as the U.S., such as training state-of-the-art generative AI models, though it consumes more power than U.S. chips.
U.S. researcher SemiAnalysis recently reported that one such Huawei cluster, which connects 384 of its Ascend chips, outperforms Nvidia’s flagship system with 72 graphics-processing units on some metrics.
AI is expected to upend economies and militaries, and leadership in the sector is seen as critical to future global influence and national security.
The U.S. maintains its early lead, with Silicon Valley home to the most popular AI models and the most powerful chips. Much of China’s AI spending has led to waste and overcapacity.
China also clearly wants U.S. technologies. Access to advanced chips has been a priority for Chinese negotiators in trade talks, The Wall Street Journal has reported. Sales of Nvidia’s H20 AI chips to China were recently restored by Washington after it restricted them in April, a reversal seen by Beijing as a gesture of good faith in the talks.
The Trump administration is taking steps to try to preserve the U.S.’s lead. It recently announced an AI “action plan,” which aims to slash red tape and make it easier for tech companies to build data centers needed to train AI models.
https://www.wsj.com/tech/ai/how-china-is-girding-for-an-ai-battle-with-the-u-s-5b23af51