China Localises AI Chips
The latest list extends China’s trusted technology
certification system to cover AI processors as Beijing pushes the use of home-grown
chips
·
The latest assessments were jointly released by:
o
China Information Technology Security Evaluation
Centre
o
National Secrecy Science and Technology Evaluation
Centre
·
A new category titled “AI training and inference
chips” has been introduced, with certifications valid for three years.
Expansion of the Xinchuang
Initiative
·
The move broadens China’s Xinchuang
initiative, a government-backed programme aimed at replacing foreign hardware
and software in sensitive sectors with domestic alternatives.
·
Xinchuang (“information technology application innovation”)
initially focused on reducing reliance on:
o
Intel CPUs
o
Advanced Micro Devices (AMD)
o
Oracle databases
·
The campaign is now expanding into AI
infrastructure, targeting foreign AI chip suppliers such as Nvidia.
Impact of US Export Controls
·
China’s push toward domestic AI chips has
accelerated following repeated US export controls restricting access to
advanced GPUs.
·
Last week, China’s National Development and Reform
Commission said it would encourage domestic large language models to better
adapt to Chinese-made computing chips.
Chinese AI Chips Included in the Certification List
·
Newly certified products include:
o
Huawei Technologies Ascend 310 and Ascend 910 chips
o
Alibaba Group T-Head Zhenwu
M530 and M890 chips
o
GPUs from:
§ Biren
Technology
§ Hygon
Information Technology
§ Iluvatar CoreX
§ MetaX
§ Moore
Threads
Major Chinese AI Chipmakers Missing
·
Notably absent from the approved list were:
o
Cambricon Technologies
o
Kunlunxin
·
According to a source familiar with the matter,
exclusion from the certification list could affect procurement opportunities in
Xinchuang-related projects.
·
Companies are not automatically included and must
voluntarily submit products for testing under China’s “Anke V3.0” standards.
Importance of Certification
·
Inclusion in official or semi-official catalogues
is important for:
o
Government procurement
o
State-owned enterprise contracts
o
Security-sensitive technology projects
·
The secure-and-reliable certification system was
first introduced in December 2023 for:
o
CPUs
o
Operating systems
o
Centralised databases
AI Chip Market Dynamics
·
While certification matters for government-linked
procurement, the broader commercial AI chip market is still expected to depend
mainly on:
o
Performance
o
Software compatibility
o
Availability
o
Cost efficiency
·
Cloud service providers remain key buyers in
China’s commercial AI infrastructure market.
Strong Growth Expected in Domestic AI Chip Demand
·
JPMorgan Chase estimates China’s domestic AI chip
demand could grow at a compound annual rate of about 40% through 2030.
·
Growth is expected to be driven primarily by
inference workloads and rising AI token consumption rather than model training
demand alone.
Manufacturing Constraints Remain a Challenge
·
Supply constraints remain a major hurdle for
Chinese AI chip developers.
·
JPMorgan identified wafer fabrication capacity as a
major bottleneck because of:
o
Lower manufacturing yields
o
Difficulty scaling advanced chip production
·
Some Chinese suppliers are currently relying on
existing inventory to meet demand.
·
Larger shipments of domestically produced AI chips
are expected to begin in the second half of 2026.
[ABS News Service/27.05.2026]
For
the first time, China has included artificial intelligence chips in its official
“secure and reliable” technology assessments, expanding the scope of a state-backed
drive to adopt domestic alternatives to Western products.
The
China Information Technology Security Evaluation Centre and the National Secrecy
Science and Technology Evaluation Centre released the latest batch of assessments
on Tuesday evening, creating a new category for “AI training and inference chips”.
The certifications were valid for three years, according to the announcement.
As
the country’s official assessment bodies for information security and secrecy-related
technologies, their approvals are widely viewed as the definitive procurement catalogue
for party and government agencies, central state-owned enterprises and other state-linked
customers under the Xinchuang initiative.
Xinchuang, which translates as information technology
application innovation, is Beijing’s long-running campaign to purge foreign hardware
and software from sensitive and strategically important information systems.
The
move highlights how China’s technology replacement drive is pivoting towards AI
infrastructure. Initially focused on phasing out US suppliers like Intel and Advanced
Micro Devices for central processing units (CPUs) and Oracle for databases, Beijing
is now targeting Nvidia and other foreign suppliers of AI computing power.
The
push has gained urgency following successive rounds of US export controls that restricted
Chinese access to advanced graphics processing units (GPUs).
Last
week, the National Development and Reform Commission said China would guide domestic
large language models to increase adaptation to home-grown computing chips, while
promoting the integration of AI with the broader economy and society.
The
newly certified chip products include Huawei Technologies’ Ascend 310 and Ascend
910, Alibaba Group Holding’s T-Head Zhenwu M530 and M890,
as well as GPUs from Biren Technology, Hygon Information
Technology, Iluvatar CoreX,
MetaX and Moore Threads.
Alibaba
owns the South China Morning Post.
Notably
absent from the list were two of China’s most prominent AI chip developers: Cambricon Technologies and Baidu-backed Kunlunxin.
A
person familiar with the matter, who requested anonymity because they were not authorised
to speak publicly, said the exclusion from the list could affect business opportunities
in some procurement scenarios with Xinchuang requirements.
However,
each company can decide whether to submit their products for testing, as different
chips have different market positionings, the person said. If submitted for assessment,
products can only be added to the list after passing tests under the Anke V3.0 requirements.
Anke
is shorthand for “anquan kekao”,
or “secure and reliable” in Chinese, a term often used interchangeably with Xinchuang in China’s domestic technology replacement campaign.
Cambricon and Kunlunxin
did not immediately respond to requests for comment on Wednesday.
The
Xinchuang initiative started to accelerate in government
procurement around 2020, before expanding after 2022 into sectors such as finance,
energy, telecommunications and other state-linked industries.
What
began as a policy slogan has transformed into a procurement market spanning the
core layers of China’s information technology stack, from CPUs and operating systems
to databases, middleware, office software and servers.
For
domestic suppliers, inclusion in official or semi-official catalogues can help with
the bidding process and customer qualification systems, particularly in projects
where security and local control are explicit requirements.
The
secure and reliable assessment list is one of the certification mechanisms that
supports Beijing’s localisation push. China first published such assessment results
in December 2023, covering CPUs, operating systems and centralised databases.
Having
a product on the list is likely to matter the most in Xinchuang-related
procurement, where government agencies and state-owned enterprises often require
formal certification or catalogue inclusion. The larger commercial market for AI
chips, however, is still expected to be driven by cloud service providers, whose
procurement decisions are shaped more by factors such as chip availability, performance,
software compatibility and total cost of ownership.
Domestic
AI chip demand is expected to rise sharply as China’s AI industry shifts from model
development to broader deployment. JPMorgan said in an April report that China’s
indigenous AI chip demand, measured by units, could grow at a compound annual rate
of about 40 per cent through 2030, driven largely by rising token consumption from
inference workloads rather than incremental training demand.
Supply
remains one of the biggest constraints for Chinese AI chip designers. JPMorgan said
wafer fabrication capacity was a key bottleneck because of lower production yields
and the difficulty of building leading-edge manufacturing capacity. Some suppliers
had relied on existing inventory to meet demand, while significant shipments of
locally produced chips were expected to begin in the second half of 2026, the broker
said.