After years of significant investment and persistent losses, Cambrian has finally reported its first profitable quarter since going publicRenowned as a leader in the AI chip sector in China, Cambrian's journey mirrors that of the domestic AI hardware landscapeAs its revenue begins to rise notably, the domestic AI chip market seems poised to accelerate toward a new phase of growth.

In a recent earnings forecast, Cambrian (688256.SH) indicated that its projected revenues for 2024 will be between 1.07 billion to 1.2 billion RMB, reflecting an increase of 50.83% to 69.16% year-on-yearNotably, while a net loss of approximately 396 million to 484 million RMB is expected, a significant improvement over the previous year signals a narrowing of losses by as much as 53.33%.

For the fourth quarter of 2024, Cambrian is forecasting revenues between 885 million and 1.015 billion RMB, surpassing its total revenue for all of 2023. For the first time, the company is expected to present a net profit of approximately 240 million to 328 million RMB attributable to its parent shareholders.

Since its inception, Cambrian has concentrated on developing various AI chip productsAmong its successful products are the Suyuan 270, Suyuan 290, and Suyuan 370 intelligent acceleration cardsThe venture's latest Suyuan 590 AI chip, released in 2023, uses an ASIC architecture, boasting remarkable efficiency and low power consumptionIt has been designed to support nearly all major domestic AI models and is positioned against NVIDIA’s flagship A100 product.

As the artificial intelligence wave, notably exemplified by platforms like ChatGPT, sweeps the globe, there has been an immense surge in demand for AI serversLeading cloud service providers are enhancing their investments in AI computing infrastructure, indicating a cyclical uptrend in capital expenditure

Advertisements

The AI chip market is thus experiencing rapid growth, with GPUs being a cornerstone of AI server technology.

According to data from Fortune Business Insight and Mordor Intelligence, the global GPU chip market is anticipated to rise from $48.4 billion in 2024 to $65.3 billion, with expectations of reaching approximately $274.2 billion by 2029, reflecting a compound annual growth rate of around 33.20% during this period.

China's AI chip market mirrors this growth, with projections estimating the market size to reach roughly 140.6 billion RMB in 2024, up to 178 billion RMB by 2025, marking an ongoing expansion in this critical sector.

However, this growth is set against a backdrop of increased scrutiny from the United States, as the government employs various methods to stymie the advancement of China's high-tech industriesReports indicate that chip manufacturers AMD and NVIDIA received notifications in September 2022 directing them to cease supplying advanced chips to clients based in China.

In October 2023, the U.SDepartment of Commerce announced new export regulations targeting the semiconductor sector, exacerbating sanctions against Chinese semiconductor firmsFurthermore, on January 29, 2024, a newly released regulatory framework mandated that American cloud computing companies verify the identities of their foreign users, thereby restricting the ability of Chinese enterprises to train AI models using American data centers.

In response to these restrictions, the Chinese government has implemented a series of supportive policies aiming to bolster the development of domestic AI chips.

As national policies further emphasize and guide the development of the AI sector, local governments in Beijing, Shanghai, Guangdong, and Guizhou have also introduced measures promoting the local computing power industries

Advertisements

For example, in March 2024, a collaborative action plan outlined by the Shanghai Communications Administration and ten other departments set a target for Shanghai to exceed 30 EFLOPS in intelligent computing power by 2025, accounting for more than 50% of its total computing power.

Moreover, the action plan stipulates that newly constructed intelligent computing centers must ensure that over 50% of computing power chips and storage solutions utilized are domestically produced while also supporting the development of more than ten general and specialized AI model design enterprises with international competitiveness.

Meanwhile, China’s three major telecommunications operators have also announced substantial AI server procurement projects, reinforcing the local demand for domestic AI computing resourcesNotably, China Mobile has revealed plans to procure nearly 7,994 AI servers to support the construction of its intelligent computing centers.

The procurement amount from China Mobile stands at around 19.1 billion RMB, with all successful bidders being local server providers that incorporate domestic AI chips into their offerings.

On the technological front, the divergence in the AI chip landscape illustrates that GPUs are not the only playersBesides GPUs, there are also ASIC and FPGA chipsHowever, a majority of the market share within AI chips is held by GPUs and ASICsGPUs are appreciated for their capability in high-parallel processing alongside a mature ecosystem that supports cross-platform applications, while ASICs are specifically tailored for deep learning computations, boasting advantages such as efficiency, lower power consumption, and compact form factors.

According to a report by Guotai Junan Securities, ASICs are designed for specific tasks, minimizing unnecessary hardware for general-purpose computations, which results in lower overall costs in comparison to GPUs.

Despite some gaps in computing power, with models like Google’s TPUv6 and Microsoft’s Maia100 presenting around 90% and 80% of the NVIDIA H100 capabilities, respectively, the former models are significantly less expensive, enhancing their cost-performance ratio in inference scenarios.

Data from Marvell reveals that in 2023, customized chips constituted about 16% of data center acceleration chips, amounting to $6.6 billion

Advertisements

With rising AI computational demands and increasing shares of customized chips expected, the data center market for tailored computation chips is forecasted to reach $42.9 billion by 2028, with a compound annual growth rate of 45% from 2023 to 2028.

Further optimistic predictions come from Broadcom’s CEO, who anticipates significant growth in the ASIC market, suggesting that partnerships with the top three cloud computing firms could yield opportunities worth between $60 billion and $90 billion by 2027.

Consequently, beyond GPUs, some domestic AI chip manufacturers are pivoting toward the ASIC approach, with Cambrian firmly situated on this path.

According to their financial disclosures, Cambrian's primary product lines encompass cloud products, edge products, IP licensing, and smart computing cluster systems, with the latter being their leading revenue stream—accounting for approximately 85.46% of total revenue in 2023.

Cambrian's smart computing cluster system amalgamates self-developed acceleration cards or training products with servers, networking, and storage devices provided by partners, integrated with proprietary cluster management software to form data center clusters, with its core computing power stemming from domestically developed intelligent chips.

Notably, Cambrian's Suyuan 100 chip is heralded as China's first cloud intelligent chip with peak performance capabilitiesFollowing it, the Suyuan 290 emerged as the company's inaugural cloud training chip, leveraging a 7nm process for theoretical peak performances of 1024 TOPS and 512 TOPS under 4-bit and 8-bit fixed-point arithmetic, respectivelySuyuan 370 represents Cambrian's third-generation cloud product, utilizing 7nm intelligent technology as well as featuring innovative Chiplet technology.

Adding to their portfolio, the release of the Suyuan 590 AI chip at the end of 2023 marks its arrival using an ASIC architecture, emphasizing efficiency and low power consumption while supporting nearly all major domestic models and actively competing with NVIDIA’s A100.

On the stock market, Cambrian's shares have attracted considerable interest, witnessing a surge from 95.85 RMB to an impressive peak of 777.77 RMB in 2024, with its market capitalization exceeding 300 billion RMB at one point.

In contrast, Hygon Info showcases superior performance, bolstered by its GPU and CPU product lines

Advertisements

Advertisements