Chinese semiconductor thread II
- Thread starter vincent
- Start date
40％ of Korean memory productions are in China.
Aiming for 2000 Tokens/s! Hanxu Technology, a dark horse in the Peking University-affiliated streaming inference chip industry, raises tens of millions of yuan in funding.
Hanxu Technology, a Beijing University-affiliated startup specializing in ultra-fast streaming inference chips, has successfully raised tens of millions of yuan from investors including Qigao Capital and Saiyi Industrial Fund. This funding supports the company's mission to develop dedicated hardware for real-time AI applications and large model inference.
- After two years of R&D, the company's first chip prototype passed on-chip testing with a bandwidth performance of 100 GB/s/mm². Its architecture utilizes weight-based streaming to eliminate hardware scheduling bottlenecks.
- While conceptually similar to Groq, Hanxu claims superior per-unit-area weight access capabilities. The company aims to surpass Groq's limitations regarding storage density and power consumption by adopting a pioneering "MRAM+SRAM" hybrid memory architecture.
- The next-generation chip targets a performance of 2000 Tokens/s, designed specifically for low-latency scenarios such as embodied intelligence and real-time simultaneous transmission.
- As the industry shifts from training to inference, Hanxu seeks to overcome traditional GPU "memory walls" through architectural innovation, positioning itself as a critical domestic force in the AI chip sector.
Chinese vendor memory productions are tripling this year.
Xinkailai's wholly-owned subsidiary, Qiyunfang Technology, increased its capital to 50 million yuan, representing a 400% increase.
On March 11, business registration information showed that Wuhan Qiyunfang Technology Co., Ltd. recently underwent a change in its registered capital, increasing from 10 million yuan to 50 million yuan, an increase of 400%.
According to Tianyancha, Wuhan Qiyunfang Technology Co., Ltd. was established in July 2023, with He Jingwei as its legal representative. Its business scope covers the development of basic artificial intelligence software, the development of artificial intelligence application software, and information system integration services. In terms of shareholder structure, the company is wholly owned by Shenzhen Xinkailai Technology Co., Ltd.
This significant capital increase may indicate that Qiyunfang Technology will further expand its presence in areas such as artificial intelligence software and information system integration, building up its financial strength for future business expansion. As a wholly-owned subsidiary of Xinkailai Technology Co., Ltd., Qiyunfang Technology's capital expansion has also drawn market attention to its future development plans.
Public information shows that Shenzhen Xinkailai Technology Co., Ltd. has a wide range of investments in the field of technology. This significant capital increase in its subsidiary demonstrates its continued optimism about the artificial intelligence and information system integration sector.
Yixintong Precision, a high-end connector manufacturer, has completed a nearly 100 million yuan Series A financing round.
Recently, Jiangsu Yixintong Precision Electronics Co., Ltd. (hereinafter referred to as "Yixintong"), a leading domestic provider of high-end precision connector solutions, completed a nearly RMB 100 million Series A financing round. This round was led exclusively by Quande Xue Capital, and the funds will be primarily used for capacity expansion, R&D upgrades at the research institute, and the rapid rollout and order fulfillment of new products.Established in 2017, Yixintong focuses on the R&D, production, and sales of high-end connectors. With a factory area of 73,000 square meters and a cumulative investment of 500 million RMB, its core team boasts over 20 years of experience in mold design and precision manufacturing. The company has achieved large-scale mass production in three major sectors: new energy vehicles, energy storage, and consumer electronics, while simultaneously developing and expanding into products such as high-speed connectors for AI computing servers. Yixintong has strategically positioned itself in the new energy vehicle and energy storage sectors, and has already established deep strategic partnerships with leading domestic energy storage customers and OEMs, holding a leading market share in these areas and experiencing rapid growth.
Huahong Group invests in and establishes a semiconductor company
According to Qichacha, Shanghai Huayao Semiconductor Co., Ltd. was recently established, with Zhou Limin as its legal representative and a registered capital of 1 billion yuan. Its business scope includes: integrated circuit manufacturing; integrated circuit chip and product manufacturing; integrated circuit sales; integrated circuit design; integrated circuit chip design and services, etc. Qichacha's equity penetration analysis shows that the company is jointly held by Shanghai Huahong (Group) Co., Ltd. and its subsidiary, Shanghai Hualibo Enterprise Management Partnership (Limited Partnership).
China's AI Semiconductor Self-Sufficiency in 2028 is Expected
Morgan Stanley, a securities firm, predicted that China will have enough artificial intelligence (AI) semiconductor production capabilities to meet its core demand in the country from 2028.
According to Hong Kong investment magazine AAStox on the 12th, Morgan Stanley released a report and reported, "China is making significant progress in establishing a self-sufficiency system for artificial intelligence graphics processing units (GPUs)."
Morgan Stanley evaluated that China has made great progress in acquiring foundry production lines and equipment necessary for artificial intelligence semiconductor manufacturing with its own capabilities over the past year under the leadership of the government.
Some predict that by 2028, it will secure enough mass production capacity to meet its core domestic demand.
This means that the need for China to rely on overseas foundry companies or import equipment is greatly reduced.
Morgan Stanley assessed that Chinese semiconductor companies quickly paid off in the early stages thanks to the government's support policy.
Another factor contributing to strengthening the competitiveness of Chinese companies is that cost competitiveness is becoming more important than semiconductor performance in the field of artificial intelligence reasoning.
This is because artificial intelligence semiconductors of Chinese companies are performing less than products of overseas competitors such as Nvidia and AMD, but are superior in terms of cost and power efficiency.
Morgan Stanley said, "China's strategy to overcome its technological weakness by expanding the scale of semiconductor production is leading to steady results," adding, "The distribution of locally developed GPUs has the potential to lead to overseas exports."
However, some predicted that if Chinese companies have difficulty maintaining their competitiveness, semiconductor companies could eventually suffer from oversupply and restructuring.
Morgan Stanley predicted that China's artificial intelligence semiconductor market will reach 67 billion U.S. dollars by 2030. This means that the market will grow at an annual average of 23 percent from 2024.
Some predict that China's self-sufficiency in artificial intelligence semiconductors will hit 76% by 2030.
Morgan Stanley added that based on this, it maintains an optimistic view on Chinese big tech companies investing in SMIC, Naura, and semiconductor sectors.
Morgan Stanley, a securities firm, predicted that China will have enough artificial intelligence (AI) semiconductor production capabilities to meet its core demand in the country from 2028.
According to Hong Kong investment magazine AAStox on the 12th, Morgan Stanley released a report and reported, "China is making significant progress in establishing a self-sufficiency system for artificial intelligence graphics processing units (GPUs)."
Morgan Stanley evaluated that China has made great progress in acquiring foundry production lines and equipment necessary for artificial intelligence semiconductor manufacturing with its own capabilities over the past year under the leadership of the government.
Some predict that by 2028, it will secure enough mass production capacity to meet its core domestic demand.
This means that the need for China to rely on overseas foundry companies or import equipment is greatly reduced.
Morgan Stanley assessed that Chinese semiconductor companies quickly paid off in the early stages thanks to the government's support policy.
Another factor contributing to strengthening the competitiveness of Chinese companies is that cost competitiveness is becoming more important than semiconductor performance in the field of artificial intelligence reasoning.
This is because artificial intelligence semiconductors of Chinese companies are performing less than products of overseas competitors such as Nvidia and AMD, but are superior in terms of cost and power efficiency.
Morgan Stanley said, "China's strategy to overcome its technological weakness by expanding the scale of semiconductor production is leading to steady results," adding, "The distribution of locally developed GPUs has the potential to lead to overseas exports."
However, some predicted that if Chinese companies have difficulty maintaining their competitiveness, semiconductor companies could eventually suffer from oversupply and restructuring.
Morgan Stanley predicted that China's artificial intelligence semiconductor market will reach 67 billion U.S. dollars by 2030. This means that the market will grow at an annual average of 23 percent from 2024.
Some predict that China's self-sufficiency in artificial intelligence semiconductors will hit 76% by 2030.
Morgan Stanley added that based on this, it maintains an optimistic view on Chinese big tech companies investing in SMIC, Naura, and semiconductor sectors.
沪电股份 invested 4.3B RMBin 2024Q4 to build new high-end PCB expansion project for AI chips. Construction started in 2025/06 and expect to start trial production in 2026H2 and then gradually increase. This will help meet high end PCB for high speed computing server and AI. Its base in Thailandentered production in 2025Q2.