Tianzhi Chip Hong Kong IPO Valuation Analysis: Are Domestic GPU Technology Barriers Overestimated?

Tianzhi Chip officially launched its public offering on December 30, 2025, and plans to list on the main board of the Hong Kong Stock Exchange on January 8, 2026 (stock code: 9903). The IPO will issue 25.4318 million H-shares at a price of HK$144.60 per share, with estimated net proceeds of approximately HK$3.479 billion and an IPO market capitalization of approximately HK$35.442 billion[1][2]. Notably, the price-to-sales (PS) ratio corresponding to this IPO is about 50-60x, significantly lower than that of its A-share listed peers.

According to the financial data disclosed in the prospectus, Tianzhi Chip has maintained strong growth momentum in recent years. From 2022 to 2024, the company’s revenue reached RMB 189 million, RMB 289 million, and RMB 540 million respectively, with a three-year compound annual growth rate (CAGR) of 68.8%[1][3]. In the first half of 2025, it achieved revenue of RMB 324 million, a year-on-year increase of 64.2%, showing steady growth momentum. In terms of customer expansion, the number of customers served by the company jumped from 22 in 2022 to 181 in 2024, achieving a “triple jump” growth. As of June 30, 2025, the products have been deployed more than 900 times in over 290 customers, covering multiple fields such as financial services, healthcare, and transportation[2].

The offering has introduced a cornerstone investor group consisting of 18 top investors, with a total subscription amount of approximately HK$1.583 billion. The cornerstone investors include well-known institutions such as ZTE Corporation (Hong Kong), UBS AM Singapore, Huascend Technologies (Hong Kong), 4Paradigm, China Universal Asset Management (Hong Kong), China Asset Management (Hong Kong), Xinxing Holdings, and Mingshan Capital[1][3]. This shows the high recognition of industrial capital and institutional investors for the domestic general-purpose GPU track. Previously, the company has received support from many well-known institutions such as Grand Capital, Princeville Capital, Starry Night Capital, Sequoia China, and Yuanhe Capital. In 2025, it completed Series D and D+ financings, raising more than RMB 1.4 billion and RMB 2.05 billion respectively, with a pre-money valuation of RMB 12 billion in the Series D+ round[2].

One of the most notable focuses of this IPO is the huge valuation gap between Tianzhi Chip and its A-share peers. According to public data, the PS ratio corresponding to Tianzhi Chip’s Hong Kong IPO is about 50-60x, while the PS-TTM of Muxi Semiconductor and Moore Threads listed on A-shares are as high as 139x and 303x respectively[1]. This means that with similar revenue scales, the Hong Kong market capitalization is only one-tenth of the peak value of its A-share peers.

This valuation gap reflects the截然不同 pricing logics of the two markets for tech stocks. The Hong Kong market is relatively more mature and rational, focusing more on a company’s profitability, cash flow status, and business model sustainability. Tianzhi Chip is still in a loss-making state, and the prospectus clearly states that “net loss will increase significantly in 2025” without giving a timeline for achieving profitability[1]. In the eyes of Hong Kong investors, continuous R&D investment and unfulfilled commercial prospects make high valuations lack solid financial support.

In contrast, the valuation of domestic GPU companies in the A-share market focuses more on the strategic value of technological breakthroughs and market substitution space. Against the backdrop of computing power sovereignty rising to a national strategy, the capital market’s pricing of domestic general-purpose GPU companies has gradually moved away from short-term profit orientation, instead focusing on whether the company has sustainable technical assets, delivery capabilities independent of external ecosystems, and substitution certainty driven by both policy and demand[2]. This valuation logic gives technology breakthrough companies more room for imagination, but it also means that valuations may deviate from traditional financial analysis frameworks.

From the perspective of industry horizontal comparison, the domestic GPU “Four Little Dragons”—Moore Threads, Muxi Semiconductor, Tianzhi Chip, and Biren Technology—have扎堆 listed between 2024 and 2025, forming an IPO relay race at the end of the year[1][3]. These companies represent the latest development level of the domestic GPU industry, but the differentiated pricing of the capital market also reveals investors’ different judgments on the industry’s prospects.

Whether the technology barriers of domestic GPU companies are overestimated needs to be systematically analyzed from multiple dimensions such as computing power performance, software ecosystem, manufacturing process, production capacity guarantee, and customer trust.

In terms of computing power performance, domestic GPUs still have significant gaps with international giants. Taking Biren Technology as an example, its first flagship general-purpose GPU BR100 adopts data flow parallelism + Chiplet technology + TSMC N7 process, with 16-bit floating-point computing power exceeding 1000 TFLOPS and 8-bit fixed-point computing power exceeding 2000 TOPS, which is close to or even surpasses NVIDIA’s A100 and H100 in some indicators[4]. However, whether this “brute force computing” strategy is commercially feasible remains to be verified. In contrast, as the first Chinese company to achieve mass production of both training and inference general-purpose GPUs, Tianzhi Chip has launched the TianGai series training GPUs and ZhiKai series inference GPUs based on a full-stack self-developed architecture, with certain advantages in technical integrity[2].

The software ecosystem is the biggest challenge facing domestic GPUs. As of October 2025, Muxi’s MXMACA software stack had only 150,000 registered users and more than 13 million network API calls[4][5]. Compared with NVIDIA’s CUDA ecosystem built over 20 years, this scale is negligible. The CUDA ecosystem has gathered millions of developers, thousands of optimized applications, and a huge knowledge base, forming a strong network effect and switching costs. A technical director of an AI company frankly said: “We will not easily change GPU suppliers because the migration cost of applications developed based on CUDA is too high. It takes months just for testing and adaptation, and we will not take risks unless domestic GPUs have overwhelming advantages in performance and cost.”[5]

Although Tianzhi Chip has laid out software ecosystem construction and built the DeepSpark open-source community, which has gathered more than 400 training models and 80 inference models[1], it still has a huge gap compared with NVIDIA’s CUDA ecosystem. This ecosystem dependency puts domestic GPUs at a natural disadvantage in customer competition. Even if technical indicators are close, it is difficult to gain customer trust and adoption.

In terms of manufacturing process, domestic GPU companies generally face the challenge of limited access to advanced processes. Although they can obtain 7nm processes through wafer foundries such as TSMC, they still have gaps in energy efficiency ratio and transistor density compared with NVIDIA’s next-generation products using more advanced processes[4]. In addition, as a chip design company, Tianzhi Chip is highly dependent on supply chain links such as wafer foundry and packaging testing. Geopolitical risks in the global semiconductor supply chain may cause uncertainty to its business[1].

From the perspective of production capacity guarantee, although domestic GPU companies have made breakthroughs in design capabilities, they are still受制于 others in key links such as advanced packaging and HBM memory. This supply chain shortcoming not only affects product performance but also limits the ability of large-scale mass production. When choosing suppliers, customers consider production capacity stability as an important factor, and the disadvantages of domestic GPUs in this regard further weaken their market competitiveness.

Comprehensive evaluation shows that domestic GPUs are about 40% of international giants in computing power performance, only 10% in software ecosystem, about 50% in manufacturing process, about 60% in production capacity guarantee, and about 30% in customer trust. These data indicate that the technology barriers of domestic GPUs are still at a low level overall, and there is still a long way to go to form an effective competitive moat.

Even if the capital market gives high valuations to domestic GPU companies, their commercialization landing still faces many challenges. From Tianzhi Chip’s prospectus, it can be seen that the company is trapped in a dilemma of “high investment, difficult to profit”.

High customer concentration is the primary risk. From 2022 to 2024, revenue from the top five customers accounted for 94.2%, 73.3%, and 73.4% of total revenue respectively[1]. This highly concentrated customer structure means that the company’s revenue is highly dependent on a few large customers. Once major customers adjust their procurement strategies or reduce orders, it will have a significant impact on the company’s performance. The situation of Moore Threads is even more severe: in the first half of 2025, the revenue from the top five customers accounted for as high as 98.29%, of which the largest single customer accounted for 56.63%[5]. Muxi Semiconductor’s top five customers contributed more than 88% of sales in the first three months of 2025. This customer structure is formed because domestic GPU companies have not yet formed large-scale commercialization capabilities and can only rely on a few large customers to open the market.

Price wars further compress profit margins. In the case of no advantage in technology and ecosystem, domestic GPU companies can only compete for customers by low prices, which directly affects gross profit margin performance. Tianzhi Chip’s gross profit margin dropped from 59.4% in 2022 to 49.1% in 2024[5], which is a direct reflection of industry involution.

The uncertainty of the profit timeline also brings concerns to investors. Tianzhi Chip’s prospectus clearly states that “net loss will increase significantly in 2025” without giving a timeline for achieving profitability[1]. In contrast, Muxi Semiconductor expects to reach break-even as early as 2026, while Moore Threads expects to achieve consolidated statement profitability as early as 2027[1]. This means that investors may need to wait a long time to see returns.

From the perspective of the industry as a whole, domestic GPU companies are generally in the stage of “burning money for market share”. About 80% of the funds raised by Tianzhi Chip in this IPO will be invested in R&D of products and solutions[1], reflecting the industry’s high dependence on continuous R&D investment. This development model can be maintained during the period of abundant capital, but once the financing environment tightens or market expectations change, enterprises will face huge survival pressure.

Based on the above analysis, whether Tianzhi Chip’s Hong Kong IPO valuation of 50-60x PS is reasonable needs to be judged from multiple angles.

From a positive perspective, the domestic GPU industry is in a development opportunity period driven by both policy dividends and market demand. Computing power sovereignty has risen to a national strategy, and AI applications continue to deepen, providing a broad market space for domestic GPUs. As the first domestic company to achieve mass production of both training and inference GPUs, Tianzhi Chip has certain first-mover advantages and technical accumulation. The company’s revenue maintains a CAGR of 68.8%, customer数量 expands rapidly, and commercialization landing results are significant[2]. The active participation of cornerstone investors also reflects industrial capital’s recognition of the company.

From the risk perspective, the technology barriers of domestic GPUs are indeed suspected of being overestimated. The gap with NVIDIA’s software ecosystem is difficult to bridge in the short term, and issues such as high customer concentration, profit erosion from price wars, and continuous losses are prominent. More importantly, the relatively rational valuation environment in the Hong Kong market means that the company is difficult to replicate the speculative valuation levels of its A-share peers.

From the perspective of valuation comparison, Tianzhi Chip’s Hong Kong valuation of 50-60x PS provides a larger safety margin compared with the 139-303x PS of its A-share peers. This valuation gap is not simply a market bias, but reflects different pricing logics of different markets for risk and growth. For investors with lower risk appetite, Hong Kong valuations may be more attractive; for investors pursuing high elastic returns, they need to weigh the valuation risks of A-share peers.

The technology barriers of domestic GPUs are still in the construction stage overall, and it will take time to form an effective competitive moat. The capital market’s optimistic expectations for the industry are partly based on the long-term logic of “domestic substitution”, but technical breakthroughs and commercialization landing still face many challenges in the short term. For investors, while chasing hot concepts, they need to pay more attention to changes in the company’s fundamentals and valuation rationality to avoid blind following.

[1] Eastmoney.com - Tianzhi Chip Hong Kong IPO Special Report (https://caifuhao.eastmoney.com/news/20260101234703435799170)
[2] Shanghai Securities News - Tianzhi Chip’s IPO Market Capitalization Will Reach HK$35.4 Billion (https://finance.eastmoney.com/a/202512313606769760.html)
[3] 21st Century Business Herald - Tianzhi Chip to List Soon: Offering Price HK$144.6, Market Capitalization About HK$35.4 Billion (https://www.21jingji.com/article/20251230/herald/2e727c0a28c0736bcf8ffb4ec0cdd2e0.html)
[4] Sina Finance - The Great Era of Domestic GPUs (https://finance.sina.com.cn/roll/2025-12-23/doc-inhcucqk8415999.shtml)
[5] DoNews Column - Domestic GPU Four Little Dragons List in Crowd, Seizing the “Chinese Version of NVIDIA” (https://www.donews.com/article/detail/7430/94635.html)