Summary
- Huawei
showcased its CloudMatrix 384 AI system at the WAIC in Shanghai on July
26, 2025. - Analysts
view it as a direct competitor to Nvidia’s GB200 NVL72 system, challenging
Nvidia’s dominance. - The
system uses 384 Ascend 910C AI chips in a supernode design, delivering 300
PFLOPS computing power and 48TB high-bandwidth memory. - It
aims to fill China’s AI hardware gap caused by U.S. export restrictions on
Nvidia chips. - Nvidia
CEO Jensen Huang acknowledged Huawei’s fast progress and called
CloudMatrix 384 a strong rival. - Dylan
Patel of SemiAnalysis said Huawei’s system might exceed Nvidia’s
performance in some tasks. - Huawei
shipped many Ascend 910C chips domestically, offering a local alternative
to Nvidia products. - CloudMatrix
384 has been running on Huawei’s cloud since June 2025 and drew big crowds
at WAIC. - Huawei
declined detailed comments about the system at the event. - The
development highlights China’s goal of AI self-sufficiency amid U.S.-China
tech tensions.
China’s Huawei Technologies sparked global attention on
Saturday by unveiling the CloudMatrix 384—an artificial intelligence (AI)
computing system that industry experts say directly challenges Nvidia’s
leadership in advanced AI hardware. Debuted at Shanghai’s World Artificial
Intelligence Conference (WAIC), the high-profile launch pans out against a
backdrop of persistent U.S.-China tech tensions and export bans affecting
Nvidia’s products in China.
What Did Huawei Announce at the World Artificial
Intelligence Conference?
At the WAIC event in Shanghai, Huawei publicly displayed its
next-generation AI computing cluster named the CloudMatrix 384. As reported by
Reuters, this public showcase drew a “large crowd” to the company’s booth,
attesting to strong industry and media interest. Since its initial
announcement in April 2025, global analysts have speculated about its
significance for both domestic and international markets. The system is
perceived as a direct response to the ongoing restrictions that prevent Chinese
firms from accessing most of Nvidia’s advanced AI chips.
What Are the Main Features of Huawei’s CloudMatrix 384
System?
According to analysis from the South China Morning Post and
SemiAnalysis, and as summarized by several international outlets, the
CloudMatrix 384:
- Incorporates
384 Ascend 910C neural processing units (NPUs) and 192 Kunpeng server
CPUs, bound in a “supernode” architecture. - Provides
ultra-high bandwidth and low latency interconnectivity. - Delivers
system-level performance that rivals—and on some tasks, exceeds—Nvidia’s
GB200 NVL72, which contains only 72 of Nvidia’s latest B200 chips. - Achieves
up to 300 PFLOPS of processing power and 48TB of high-bandwidth memory in
a single system, making it optimized for demanding AI workloads such as
large language model (LLM) serving. - Is
already operational on Huawei Cloud since June, according to Huawei Cloud
CEO Zhang Pingan.
How Does the CloudMatrix 384 Compete With Nvidia’s Flagship
Offerings?
In an April 2025 analysis cited by Reuters, Dylan Patel of
SemiAnalysis stated that, “Huawei has developed AI system capabilities that
could surpass those of Nvidia”. Patel emphasized that although Nvidia’s
individual chips may lead in raw power, Huawei’s innovation is in its robust
multi-chip interconnect, allowing for system-level performance that outpaces
Nvidia’s in certain configurations.
Nvidia CEO Jensen Huang, in an interview with Bloomberg and
as echoed by Huaweicentral, acknowledged that,
“Huawei’s technology…is probably
comparable to an H200. They’ve been moving quite fast. They’ve offered this new
system called CloudMatrix, which scales up to even a larger system than our
latest generation, Grace Blackwell”.
Huang further admitted that Huawei
could now deliver solutions within China that Nvidia no longer had permission
to sell under tightened U.S. export rules.
What Motivated Huawei to Develop Its Own AI Infrastructure?
As noted by Reuters, the unveiling of the CloudMatrix 384
comes at a crucial time for China’s AI sector. Shrinking access to Nvidia chips
due to U.S. sanctions drove domestic companies to seek homegrown alternatives. Huawei
responded proactively, ramping up the design and shipment of its Ascend series
chips—the foundational components of the CloudMatrix 384 and similar
high-performance AI computing systems.
Huawei has previously emphasized its “Full-Stack,
All-Scenario AI Portfolio,” aiming to make China less dependent on foreign
technology by controlling every layer of the stack—from chips to cloud
platforms to AI frameworks such as MindSpore.
What Are Industry and Analyst Reactions?
- Dylan
Patel (SemiAnalysis) in comments highlighted by Reuters on April 21:
“Huawei’s new system could beat Nvidia’s performance in some
configurations by leveraging aggressive system-level design and
interconnect”. - Nvidia
CEO Jensen Huang told Bloomberg in May, as relayed by Huaweicentral:
“Huawei’s technology is probably comparable to an H200…They’ve been moving
quite fast.” Huang identified CloudMatrix as a notable example of the
Chinese company’s progress. - A
recent SCMP tech report by Huawei and SiliconFlow researchers described
the CloudMatrix 384 as “a specialized AI supernode purpose-built for
extensive AI workloads, reshaping the foundation of AI infrastructure”.
How Is Huawei Navigating U.S. Trade Restrictions?
Multiple outlets, including Reuters and SCMP, pointed out
that despite being placed on the U.S. Entity List, Huawei has pressed forward
with in-house semiconductor and software innovation. The CloudMatrix project is
cited as a direct testament to Huawei’s resilience and determination to capture
growing domestic demand for advanced digital infrastructure.
Huawei officials at the WAIC event declined to answer
detailed questions about the CloudMatrix 384, demonstrating caution around
disclosures under international scrutiny.
What Are the Implications for China’s and the Global AI
Supply Landscape?
With CloudMatrix 384, Huawei has firmly positioned itself as
China’s most viable supplier of advanced AI chips for data centers and cloud
platforms. The installation and use of the new system on Huawei Cloud were
confirmed by CEO Zhang Pingan in June. Industry experts expect this to
accelerate development and deployment of next-generation AI applications within
China, cushioning the blow from ongoing U.S. restrictions.
As reported in the South China Morning Post, the CloudMatrix
384 “surpassed the performance of some of the world’s most prominent systems in
running DeepSeek’s 671-billion-parameter R1 reasoning model”. This
suggests that China may now rely less on foreign technology for foundational AI
computing.
What Is Huawei’s AI Strategy Moving Forward?
Huawei’s broader AI plans are anchored in investing in
research, developing a full portfolio from chips to cloud platforms, and
nurturing an open AI ecosystem ready to serve everything from public and
private clouds to edge and IoT devices. MindSpore, its answer to
TensorFlow, is being positioned as the backbone of its AI framework strategy.
What Statements Did Key Stakeholders Make?
- Dylan
Patel, founder, SemiAnalysis (as quoted by Reuters and TradingView): “Huawei
now had AI system capabilities that could beat Nvidia”. - Jensen
Huang, CEO, Nvidia (interview with Bloomberg, cited by Huaweicentral): “Huawei’s
technology…probably comparable to an H200. They’ve been moving quite fast”. - Zhang
Pingan, CEO, Huawei Cloud (commented in June, as per Reuters and Yahoo
Finance): Confirmed CloudMatrix 384 operational status on
Huawei’s cloud platform.
The unveiling of Huawei’s CloudMatrix 384 is a landmark
moment in the ongoing global race for AI infrastructure. By demonstrating a
system that can challenge Nvidia’s top products and by scaling to meet domestic
and international AI workloads, Huawei cements itself as not just a survivor of
U.S. sanctions but as an active leader in next-generation computing. The coming
months will reveal how this shift alters the competitive tech landscape for
both Chinese and global enterprises.
