Background: Recently, BofA Securities invited Mr. Greg Ye, founding partner of Delta Capital, to conduct an in-depth discussion on the latest developments in the localization of China’s semiconductor industry. Mr. Ye has over 20 years of experience in private equity, executive management, and consulting, focusing on semiconductor investments.
Core Highlights
-
Strong adoption of domestic AI chips amid geopolitical headwinds
-
Chinese companies have made rapid progress in the high-performance AI accelerator field, with leaders including Huawei/HiSilicon, Cambricon, and the “Four Little Dragons” (Biren, Enflame, MetaX, and Moore Threads). The policy headwinds from the US and China have forced Chinese AI chip users (such as internet giants) to turn to domestic suppliers, driving revenue growth and improving profitability for local companies. Even if the US lifts restrictions, domestic companies are expected to maintain significant growth potential because: 1) chip users will maintain a multi-supplier strategy to reduce supply risks; 2) the performance gap is narrowing, with some domestic chips now able to match or even surpass Nvidia’s H20. In the medium to long term, the industry may consolidate into 3-5 major suppliers. Meanwhile, domestic startups are also expanding in the high-performance edge AI chip sector, leveraging design and architecture advantages—7nm technology is sufficient to support this. Mr. Ye also sees significant opportunities in data center supporting chips, such as connectivity chips and next-generation memory.
-
Advanced wafer foundry capacity remains tight
-
Chinese wafer foundries have begun mass production of 7nm nodes. However, capacity is fully loaded and primarily allocated to tier-1 customers like Huawei, resulting in overall supply falling far short of actual demand. In the 5nm node, some leading foundries have started small-batch production but have not yet achieved mass production. The shortage of advanced chip manufacturing capacity may last at least 2-3 years, constrained by shortcomings in key equipment technology. For example, photolithography equipment manufactured by Chinese suppliers lags ASML by about 7-8 years in technical capability.
-
EDA/IP localization is underway
-
According to Mr. Ye, domestic EDA accounts for about 20% of the market share—leading local EDA suppliers are typically five years behind global leaders (such as Cadence), while some rapidly growing startups may have narrowed the gap to 2-3 years. As domestic foundries advance to 7nm and below nodes, it is expected that the market share of domestic EDA suppliers will approach 50% within the next five years. The IP sector is more fragmented: CPU IP is still dominated by ARM (80-90% share), but local suppliers are expected to capture 30% of the open-source RISC-V architecture IP market within five years.
Q&A Summary
Q: Compared to 2-3 years ago, the Hong Kong IPO market is thriving and concerns about ADR delisting have eased. What impact does this have on VC investment in China’s semiconductor industry?
A: In the foreseeable future, Hong Kong IPOs will remain the mainstream exit channel for VC/PE investments in China’s semiconductor industry, especially after the strong performance of the Hong Kong stock market this year. Of the approximately 200 projects Delta Capital has invested in over the past 15 years, about one-third have achieved exits, with another 120 projects in preparation.
While Hong Kong IPOs remain the primary channel, A-share IPOs are returning and opening up to more companies, although the number of IPOs has not yet recovered to the levels of 2020-21. He emphasized the rise of the “Four Little Dragons” in the high-performance AI chip sector, half of which have been approved for listing. With the China Securities Regulatory Commission (CSRC) now allowing cross-industry mergers and acquisitions, the pace of backdoor listings (or reverse mergers) in the A-share market is accelerating. For example, according to public records, a traditional gas station operator may be approved to acquire control of a leading semiconductor IP developer. Backdoor listings are another viable exit channel for VC-backed tech startups.
Regarding listings in the US, due to geopolitical concerns, Chinese companies in the deep tech sector are unlikely to be approved for US listings in the short term. Currently, Hong Kong and mainland China will remain the main exit channels.
Q: In which areas is the technology gap between China and the US most evident, and is it narrowing or widening?
A: Overall, the technology gap between China and the US is narrowing in most deep tech fields, and he does not believe there is any area where the gap is necessarily widening. China has made substantial progress in areas that do not require significant capital equipment, such as high-end chip design in the EDA/IP field and advanced packaging like 2.5D packaging.
However, he also emphasized the two areas with the largest technology gaps—1) in the semiconductor equipment sub-industry, particularly photolithography machines, where Chinese companies lag ASML by 7-8 years; 2) in advanced processes of 7nm or below. Some leading Chinese foundries can now produce a considerable volume of 7nm chips, but supply is far behind actual demand. Some domestic foundries may be developing 5nm processes, but large-scale production is unlikely in the short term.
The progress in chip design may lag because Chinese chip design companies are still 2-3 years behind Nvidia, which is the same as 2-3 years ago, primarily constrained by weak semiconductor manufacturing capabilities. Even if Chinese companies can design 3nm chips, TSMC’s manufacturing capacity is limited by geopolitical factors, and there are currently no foundries in China capable of handling this task.
Opportunities in the semiconductor sub-industry: From the perspective of early investors, the entire industry chain is full of opportunities. The focus in the next 2-3 years will be on AI-related chip design and upstream bottleneck areas.
In AI, the high-performance cloud AI chip sector is already somewhat crowded. Major players include the “Four Little Dragons,” Cambricon, and Huawei’s HiSilicon. There are already 5-6 companies of considerable scale, some of which are already listed or about to be listed. He also mentioned opportunities in the following two areas:
1. Chips entering data centers, including a) connectivity chips, which can compete with or replace NVLinks; b) scale-up/scale-out chips: while Nvidia is dominant, it is not as strong in this area as in the GPU field; c) storage chips, such as HBM (high bandwidth memory) and new types of memory that can replace DRAM in high bandwidth data storage and transmission.
2. Another opportunity beyond GPUs is how to push AI computing power to the edge and client side. Currently, there is no dominant player globally—Nvidia does not dominate this field. Qualcomm is a major player. Mr. Ye noted that many Chinese startups have made good progress in this area, as these chips rely more on design techniques or intelligent architectures rather than advanced manufacturing nodes (high-quality client AI chips can be achieved using 7nm technology without relying on 5nm/3nm technology). However, the challenge lies in how to address power consumption issues and adapt to the architecture of limited space in mobile phones and laptops.
Q: To what extent can China’s semiconductor supply chain meet the needs of local chip manufacturers, especially for advanced processes?
A: For the past two years, 7nm capacity has been very tight, almost 100% occupied by Huawei for mobile Kirin chips or AI chips. Although capacity has increased, demand has also piled up. Currently, apart from Huawei, 7nm capacity is available for a few customers, but supply is far behind demand. He believes this situation will persist for at least the next 2-3 years.
Q: You mentioned that the technology gap in photolithography is about 7-8 years. Does this refer to the comparison between China’s DUV (Deep Ultraviolet) and ASML’s most advanced DUV? Have you considered EUV (Extreme Ultraviolet) technology?
A: China’s largest photolithography machine developer, SMEE (Shanghai Micro Electronics Equipment), reportedly can achieve 28nm through single exposure or 7nm through multi-patterning. This is similar to ASML’s 1980i model launched 10-15 years ago—this is the level we are currently at. China may increase production in the next two years. To his knowledge, there may be a national effort to deliver prototypes of EUV machines, which may debut by the end of this year or early next year. Therefore, they may launch the first generation of EUV, similar to ASML’s machines from about 7-8 years ago.
However, ASML has continued to advance during this time. They have begun transitioning a few customers to the second generation of EUV. SMEE will need another 2-3 years to achieve mass production. With government support and the joint efforts of the entire industry chain, it is only a matter of time before Chinese companies achieve this goal.
Q: If the US lifts restrictions and resumes AI chip supplies to China, will Chinese buyers return? What market share can US AI chips occupy?
A: B30 (Nvidia’s new chip) may be available to China next year, while B200 is currently only sold to the Western world. B30 is a streamlined version of B200 but is still much better than H100 and may also outperform any products that China can offer. If that day comes, Chinese companies will welcome it. On the other hand, based on past experience, Chinese companies will not rely 100% on it. This is not only a government mandate but also requires different judgments. Some of the largest customers need at least two suppliers, including Nvidia and a domestic supplier, to guard against changes in US-China relations.
Another point is that data centers require both training and inference capabilities. The growth rate of inference demand may outpace training. Chinese chips are sufficient for inference tasks, often offering better cost-performance ratios than Nvidia chips. Some data center builders prefer to purchase a small number of Nvidia chips for training and a large number of domestic chips for inference. Some AI infrastructure software can even mask heterogeneous setups, such as pairing one Nvidia B200 with ten local accelerators.
Q: With over a dozen companies in China developing AI chips, are you concerned about increased competition or consolidation?
A: Consolidation will happen sooner or later, as ultimately the Chinese market does not need more than five or six GPU developers. China already has large enterprises in the AI field, along with another five or six companies engaged in graphics rendering. Many Chinese customers prefer to be tightly bound to specific suppliers, which is different from the US, where the “Mega-7” all purchase from a large AI chip supplier. In this market structure in China, there may be enough space for more than three AI GPU providers to survive, but not more than five.
Q: Current market share of domestic companies in the EDA/IP field, and views on the fastest localization and key players in the segment
A: Domestic EDA vendors hold about 20% of the market share, while Synopsys and Cadence still dominate. He expects local share to rise significantly and mentioned three A-share listed companies and rapidly growing private enterprises. Currently, leading Chinese suppliers’ tools lag global leaders by about five years, although some digital realization solutions are only 2-3 years behind. Progress in EDA needs to synchronize with the advanced process capabilities of foundries. Once domestic 7nm and 5nm capacities mature, the share of domestic EDA vendors may reach 50% within five years.
In the IP sector, he described a fragmented market. ARM holds 80-90% of the CPU IP business, while Chinese startups focusing on RISC-V, with strong government support, may capture 30% of the domestic market. In relatively low-value IP categories like interface IP, local share has already exceeded 30% and may surpass 50% within 3-5 years. While a technology gap still exists compared to global leaders, this creates opportunities for domestic startups and investors.
Q: Growth prospects for China’s EDA/IP in the context of surging chip design and increasing complexity
A: Driven by high single-digit growth in the global EDA market and localization opportunities in China, the overall EDA market in China is expected to grow at least 20% annually. He added that three listed EDA companies are achieving over 30% annual growth through a combination of organic growth and acquisitions, while smaller players may see even higher growth. In the IP sector, domestic interface IP vendors are expected to grow by 30% annually, and well-funded RISC-V companies may double their revenue annually over the next five years.
Q: Synopsys and Cadence have shifted to AI-driven EDA tools. Do you see Chinese EDA suppliers taking a similar direction in AI empowerment?
A: Chinese EDA vendors are actively embracing AI in various ways. He emphasized: 1) Companies are upgrading existing tools and launching new products to better support AI chip developers; 2) Companies are applying AI to their workflows, such as using AI accelerators like Nvidia H100 or Huawei Ascend for simulation software calculations instead of relying on CPUs; 3) Companies are integrating AI and machine learning algorithms into their software, such as layout and routing tools, to provide better performance.
Q: The technology gap between mainland China and Japanese/Taiwanese companies in 12-inch wafer production, and the possibility of large-scale production in the next 1-2 years
A: The gap is relatively small. He noted that several domestic 12-inch wafer manufacturers have produced wafers of acceptable quality for leading Chinese foundries. Major foundries, including SMIC and Huawei-affiliated factories, are sourcing some wafers from domestic suppliers while continuing to import from Taiwan and Japan.
Guest Speaker Bios
Mr. Greg Ye has over 20 years of experience in private equity, executive management, entrepreneurship, and consulting, and is one of the founding partners of Delta Capital. His areas of focus include information technology and medical devices. He has led successful early or growth-stage investments in companies such as Zhongji Xuchuang (300308), Daqo New Energy (NYSE: DQ), Chuan Yi Co., Ltd. (SHSE: 603100), Sijinch Power Electronics (HKSE: 580), Yuanwanggu (SHSE: 002161), Obit (300053), and Zebao (300464). Before founding Delta, Mr. Ye served as managing partner at New Margin Ventures and as director of corporate strategy and business development at Cadence Design Systems Inc. He has also served as chairman of the North America China Semiconductor Association and holds an MBA from Harvard Business School.
「Long press the QR code below
Enter to get this research report for free」
