Silang Technology recently filed for an IPO on the Shanghai Stock Exchange's STAR Market, aiming to raise 4.426 billion yuan at a valuation of 8 billion yuan. The company, known as a "national-level unicorn," boasts a shareholder list that includes Kweichow Moutai, CATL, and SMIC. Its core products, the "Tianqiong" 3D scientific computer and the UCP series of communication baseband chips, are designed to break foreign technology monopolies by providing scientific computing infrastructure. Revenue grew impressively from 251 million yuan to 673 million yuan over three years, but a closer look reveals a critical issue: from 2023 to early 2026, a single client, Yangtze River Computing, contributed 100%, 95.39%, 99.03%, and 97.4% of revenue, respectively. This suggests Si Lang Tech effectively has only one customer.
The financial data is equally concerning. Cumulative R&D spending of 1.713 billion yuan has surpassed total revenue, and net losses have exceeded 1 billion yuan over three years. In the first quarter of 2026 alone, the company posted a loss of 283 million yuan, nearly matching the entire loss for 2024. While the gross margin is an impressive 89%, the inventory turnover rate is a mere 0.14, meaning it takes over two years to sell products from stock. This paints a picture of a company that is technologically robust but commercially fragile. The MaPU architecture is 100% proprietary, and the "Tianqiong" 3D scientific computer is comparable to the US's Anton2, demonstrating strong academic credentials. However, issues like extreme customer concentration, persistent losses, inventory buildup, and a lack of an ecosystem are all critical concerns.
Solid Technology
The company's foundation is strong. Founder Wang Donglin, born in 1956, is a former director of the Institute of Automation at the Chinese Academy of Sciences and has led the development of seven national strategic processors, winning two second prizes in the National Science and Technology Progress Award. He began developing the MaPU architecture in 2009 and founded Si Lang Tech after retiring in 2016. Now 69, he serves as Chief Scientist. The MaPU architecture's key innovation is merging the efficiency of ASICs with the flexibility of CPUs and GPUs, achieving a core utilization rate of over 90% and, for specific scientific tasks, delivering equivalent computing power dozens of times greater than traditional CPUs. The "Tianqiong" 3D scientific computer is one of the few products globally capable of microsecond-scale, million-atom-level simulations, comparable to the US Anton2. Its molecular dynamics simulation capabilities are unique in China, and it has served clients like CATL, the National New Materials Big Data Center, and China Pharmaceutical University, contributing to dozens of papers in top-tier journals. The R&D team of 539 people makes up 79.26% of employees, and the company holds 150 patents. In June 2026, Si Lang Tech was valued at 8 billion yuan on the Hurun Global Unicorn List. This technological prowess is indeed top-tier in China's scientific computing chip sector.
Fragile Business
However, strong technology requires buyers. The problem for Si Lang Tech is that there are too few of them. During the reporting period, Yangtze River Computing, a state-backed computing infrastructure operator under China Information and Communication Technologies Group, was the sole source of revenue. The top five clients accounted for nearly 100% of revenue. This extreme concentration means Si Lang Tech's pricing power resides not in its own hands but in the procurement budget of a single entity. The high 89% gross margin, while indicating strong product pricing, is built on a monopolistic transaction with one buyer, not market competition. Any change in Yangtze River Computing's procurement strategy could cause Si Lang Tech's revenue to plummet. The losses are severe: net losses of 319 million yuan in 2023, 300 million yuan in 2024, 406 million yuan in 2025, and 283 million yuan in the first quarter of 2026, totaling over 1.3 billion yuan in three years and three months. Cumulative R&D spending of 1.713 billion yuan accounts for 115.37% of total revenue, meaning for every 1 yuan earned, 1.15 yuan is spent on R&D. While the STAR Market allows unprofitable companies to list, investors must understand this is not a short-term loss for growth but a long-term, high-intensity investment for technological leadership with highly uncertain commercial returns. The inventory data is also worrying. The book value of inventory was 181 million yuan, 300 million yuan, 420 million yuan, and 411 million yuan at the end of each reporting period, with turnover rates of 0.49, 0.26, 0.19, and 0.14. A turnover rate of 0.14 means it takes an average of 2.5 years to sell inventory. While high-value, customized scientific computers may have long production cycles, a turnover rate of 0.14 is extremely low for manufacturing.
Absence of Ecosystem
Si Lang Tech has chosen the most difficult path: a fully proprietary architecture that does not support CUDA and relies on no overseas IP. This contrasts with most domestic AI chip companies like Cambricon, Haiguang Information, and Biren Technology, which are compatible or partially compatible with the CUDA ecosystem to reduce customer migration costs. The MaPU architecture is entirely self-designed from the instruction set to the microarchitecture, meaning customers must rewrite software, retrain teams, and rebuild development processes. In a market where NVIDIA's CUDA has dominated scientific computing for over a decade, commercializing a new architecture is extremely challenging. The prospectus acknowledges this, stating the company must cultivate its software ecosystem from scratch, facing high market entry barriers. Currently, clients are mainly government and research institutions, with very few market-oriented corporate clients. The growth in the scientific computing market is primarily from government-led supercomputing centers and research institute procurement, with extremely low penetration in the enterprise sector, except for CATL. Another concern is the management structure. Wang Donglin, 69, holds 33.67% of voting rights as Chief Scientist, while Zha Hao, 36, with a bachelor's degree from Shanghai University of Finance and Economics, joined in 2021 and serves as Chairman and CEO. They signed a unanimous action agreement in December 2022, where Zha Hao's decisions on major matters follow Wang Donglin's opinion. Essentially, the company's technology roadmap and strategic decisions are ultimately controlled by the 69-year-old founder. For a company needing rapid marketization and ecosystem building, this governance structure's effectiveness is debatable.
Si Lang Tech's story is a microcosm of Chinese hard-tech entrepreneurship: scientists leaving the lab with world-class technology to face uncertain commercial prospects. The MaPU architecture and "Tianqiong" are undeniably robust, meeting a genuine demand for domestic alternatives in the scientific computing field, supported by strong policy. However, investing is not about technology alone; it is about the business. Key commercial metrics—over 95% customer concentration, persistent massive losses, an inventory turnover rate of 0.14, and a near-zero ecosystem—do not question its technical strength but challenge its ability to commercialize. With a planned 4.426 billion yuan fundraising and an 8 billion yuan valuation, how long can this valuation hold if the company cannot quickly expand its client base, reduce dependence on Yangtze River Computing, and build a developer ecosystem after listing? The STAR Market allows unprofitable companies to list, but listing does not mean perpetual losses are acceptable. Is Si Lang Tech's IPO the glorious beginning of a "first scientific computing stock" or another chapter in a hard-tech valuation bubble? The answer will not be in the prospectus but in the roster of customers over the next three years.
Comments