Financial Services Association, December 19 (Reporter Wang Biwei) Another chip company has entered the “trillion dollar club.” As performance skyrocketed and expectations raised, Broadcom (AVGO.US) recently became the ninth company in the US with a market capitalization exceeding trillion dollars, and the term “next NVDA.US (NVDA.US)” also emerged.

Unlike Nvidia, which mainly sells general-purpose GPUs, Broadcom mainly sells custom chips called ASICs. Wu Quan, founding partner of Huaxin Jintong Semiconductor Capital, told the Financial Federation that ASIC architectures generally achieve higher speed and efficiency in artificial intelligence than GPU and FPGA architectures. Under these circumstances, Morgan Stanley expects the AI ASIC market to grow from $12 billion in 2024 to $30 billion in 2027, with a compound annual growth rate of 34%.

With the popularity of ASIC, there are also quite a few companies in the A-share market whose stock prices have risen due to the ASIC concept. It is worth noting that most companies' ASIC chip business is far from Broadcom's business direction. “Although they are all ASIC chips, the range of ASIC chips is very broad. We mainly use consumer electronics and smart homes, which are different from Broadcom's application direction.” A person from the securities department of Rockchip (603893.SH) told the Financial Federation reporter.

Broadcom's High Expectations Bring Fire to ASIC

AI chip architectures mainly include GPUs, ASICs, FPGAs, etc. Among them, ASIC is a dedicated integrated circuit, which is an integrated circuit designed and manufactured in response to specific user requirements and the needs of specific electronic systems.

Broadcom's high expectations for ASIC are an important reason why the ASIC concept has recently received market attention.

Broadcom's performance report released on Friday showed that Broadcom's AI-related revenue (data center Ethernet chip+AI ASIC) soared 220% year over year, reaching a total of 12.2 billion US dollars for the entire fiscal year. Broadcom CEO Chen Fuyang also said at the performance conference that sales of the company's artificial intelligence products will increase 65% in the next fiscal quarter, far faster than the overall semiconductor industry's growth of about 10%; at the same time, it is also predicted that the market demand for ASIC will be 60 billion to 90 billion US dollars in 2027.

As far as the AI industry is concerned, if general-purpose GPUs represented by Nvidia GPUs are “all-encompassing,” ASIC is a more “accurate breakthrough.”

“The main requirement for ASIC is that customers can perform specific AI tasks and optimize computational performance, while reducing unnecessary functions to achieve better energy efficiency. Furthermore, it can also achieve better cost control and customization requirements.” Qiu Peiwen, an analyst at TrendForce Jibang Consulting, told the Financial Federation.

Wu Quan told reporters that there are two main reasons why demand for ASICs will explode at this point in time:

First, it is due to the transformation of AI models from training to inference. “Take the TPU in collaboration between Google and Broadcom as an example. Its architecture is mainly suitable for machine learning and deep learning. Its functions focus on inference rather than other aspects such as training and rendering.”

According to the Barclays Research Report, at present, the demand for AI inference calculation will increase rapidly. It is estimated that it will account for more than 70% of the total computational demand for general artificial intelligence. The demand for inference calculation can even exceed the demand for training calculation, reaching 4.5 times that of the latter.

Second, Wu Quan told the Financial Federation reporter that this is also related to downstream players wanting to get rid of their dependence on Nvidia. “Basically, big players who want customized chips are big players, similar to Byte, Meta, and Google. They are concerned about whether Nvidia chips can be supplied stably over a long period of time. By choosing ASIC, downstream has more control over the product and can enhance the purchasing voice.”

In a research report published on the 15th, Morgan Stanley believes that with targeted optimization and cost advantages, ASIC is expected to gradually gain more market share from Nvidia GPUs. Under these circumstances, Nvidia's stock price declined for several consecutive trading days.

However, on December 17, Citibank analysts Atif Malik and Papa Sylla released a report supporting Nvidia, reaffirming that “these two chips will coexist,” and it is estimated that by 2028, the total market size (TAM) of AI accelerators will reach 380 billion dollars, of which AI GPUs will dominate, accounting for 75% of the share, while ASIC will only account for 25%.

“Custom chips are very demanding on usage scenarios, and only a few key large customers will have them.” An AI practitioner told reporters. Qiu Peiwen further told reporters that ASICs are not mass-produced like standard products, the initial development cost is high, and after development is completed, they are usually only executed for specific customers or applications, and the flexibility is low.

Domestic ASIC surges

As Broadcom's expectations rose, the A-share market also ushered in a new narrative. The stock prices of many companies that have indicated that they designed ASIC chips have soared.

By the close of trading on December 18, the stock price of the ASIC concept stock Cambrian-U (688256.SH) rose more than 8% on the same day, a record high; Quanzhi Technology (300458.SZ) stock prices rose 9.24%; Rockchip and Fumanwei (300671.SZ) stock prices all increased by more than 8%; and the stock prices of Shanshi Nettech (688030.SH) and VeriSilicon (688521.SH) all rose more than 5%.

It should be noted that although they are all ASIC companies, most ASIC products are not AI chips, but SOC and other directions, making it difficult to benefit from this wave of AI ASICs.

For example, Rockchip has publicly stated that the company can “provide ASIC chips designed for specific functions, including interface conversion, wireless connectivity, MCU chips, etc.” However, Rockchip's securities department staff told the reporter that unlike ASICs customized by Broadcom for customers to use for big model reasoning training, the company's downstream products are mainly IOT. “Actually, this industry is very detailed, and the gap between companies is quite large.” In addition, the ASIC security chips developed by Yamashi Network Technology are more downstream of consumer electronics, communication equipment, the Internet of Things, etc.

In fact, ASIC is a very common architecture in the semiconductor industry.

According to Everbright Securities, WSTS divides ASIC types into special analog circuits, special logic circuits, digital signal processors, etc. In 2023, the ASIC market reached 248.57 billion US dollars, up 3.4% year on year, accounting for 47.2% of the semiconductor market.

It is true that many domestic AI chip manufacturers have chosen the ASIC direction. AI chip companies such as Cambrian and Yuntian Lifei-U (688343.SH) are all moving towards ASIC. However, Wu Quan told reporters that this does not mean that domestic producers currently have an opportunity to stand side by side with Broadcom.

“Compared with chip design in general fields such as CPUs, GPUs, and FPGAs, it may be easier to design ASICs in a single field, which has opened up a field for various design startups, including some domestic AI chip manufacturers. However, due to reasons such as a small market and insufficient product strength, few highly specialized ASICs are widely commercialized in China.”

Take the Cambrian Era, which has a market capitalization of over 250 billion, as an example. In the first half of this year, Cambrian's revenue was only about 65 million yuan. Although Q3's revenue in a single quarter reached 0.121 billion yuan, which greatly increased the total revenue for the first three quarters, it also included non-recurring profit and loss items of nearly 59.19 million yuan.

Under these circumstances, other segments driven by ASIC may have received more attention.

Guotai Junan believes that Broadcom's optimistic expectations for the ASIC market show the determination that cloud factories will continue to invest in AI infrastructure on a large scale in the next 2-3 years, which will drive the continued prosperity of upstream optical modules, switches, PCBs, and high-speed cables.

Currently, leading domestic optical modules include Guangxun Technology (002281.SZ), Huagong Technology (000988.SZ), etc. High-speed switch leaders include Ruijie Network (301165.SZ) and Ziguang (000938.SZ), etc., and those in the field of high-speed connectivity include Zhongji Xuchuang (300308.SZ), Xinyisheng (300502.SZ), and Lixun Precision (002475.SZ).