Tianshu Zhixin: Last year's losses worsened, reasoning products sacrificed price for volume, gross profit margin under pressure, and cash flow continued to drain.

Securities Star Li Ruohan

Recently, Tian Shu Zhi Xin (09903.HK) disclosed its first financial report after listing. Although the company’s revenue continues to grow at a high rate, its loss for the year has further worsened, with a cumulative net loss of over RMB 3.2 billion (RMB 3.2 billion+) over four years.

Securities Star noted that the company has achieved revenue growth for its inference product series by proactively cutting prices to sacrifice short-term profits, and the gross margin of the series has consequently declined. High R&D expenditures and a surge in share-based payment expenses resulting from equity incentives have further squeezed profit margins. As the business scale expands, the company’s inventories and trade receivables, as well as accounts receivable bills, rise in tandem, thereby affecting the company’s cash flow performance.

Against the backdrop of a new stage driven by AI industry efficiency and the ongoing contest between general-purpose and application-specific chip routes, the company is seeking breakthroughs through measures such as co-design of software and hardware, and rolling out products for new application scenarios. However, whether it can balance generality with specialized efficiency, and convert technological advantages into market advantages, remains to be seen.

Inference product gross margin declines, AI compute solution business cools

Public information shows that Tian Shu Zhi Xin’s general-purpose GPU products cover two key application scenarios—training and inference—with different requirements, corresponding to two major product series, “Tianxia” and “ZhiKai.”

As shipments and the number of customers grow, the company’s revenue has maintained an upward trend in recent years, increasing from RMB 189 million in 2022 to RMB 540 million in 2024. However, while revenue grows, losses have continued to expand. In 2022, 2023, and 2024, the company’s net loss amounts were RMB 554 million, RMB 817 million, and RMB 892 million, respectively.

Based on 2024 revenue data, Tian Shu Zhi Xin’s share in China’s general-purpose GPU market is about 0.3%. The market shares of both training-type and inference-type products are below 1%, meaning the company’s market share still needs further improvement.

Since 2025, the company’s loss situation has not improved. The financial report shows that in 2025, the company achieved revenue of RMB 1.03B, up 91.6% year over year; the loss for the year was RMB 1B, up 12.5% year over year. Calculated, since 2022, the company’s cumulative net loss has reached RMB 1.16B.

By product, both the “Tianxia” and “ZhiKai” series of Tian Shu Zhi Xin have seen significant growth. Corresponding revenue was RMB 584 million and RMB 339 million, accounting for 56.5% and 32.8% of total revenue, respectively.

Securities Star noted that while revenue from the company’s inference series products grew significantly, its gross margin declined. In 2025, due to the company sacrificing short-term profits by proactively cutting prices in order to capture market share and accelerate inventory sales, the gross margin of its “ZhiKai” series products was 39.2%, down 7.5 percentage points year over year.

From the perspective of technology roadmap, Tian Shu Zhi Xin has chosen a technology route with deep compatibility with NVIDIA’s CUDA ecosystem. Its advantage lies in compatibility with mainstream AI frameworks and the CUDA ecosystem, which lowers customer migration barriers and facilitates large-scale deployment.

As key technological innovation deepens, customers not only focus on peak chip performance, but also on effective computing power in real business scenarios, migration costs, development experience, and ecosystem compatibility. Whether it can achieve deep optimization and instant adaptation for mainstream AI frameworks and inference engines, whether it can provide a complete toolchain and developer support, and whether it can co-build an open ecosystem with upstream and downstream players in the industry chain, are key factors that will determine its market position.

It is worth noting that although Tian Shu Zhi Xin’s DeepSpark community has now gathered more than 600 mainstream algorithm models, the maturity level of its ecosystem is unlikely to match CUDA in the short term. Based on industry views, whether it is MUSA from Moore Threads or Tian Shu Zhi Xin’s Tian Shu Zhi Suan software stack, although each company is developing its own programming model, they have not yet formed a unified force.

In addition to general-purpose GPU products, the company’s AI compute solutions are heavily affected by fluctuations in project delivery cycles and the market supply cycle of supporting equipment. At the time, revenue declined 42.2% year over year, dropping to RMB 96 million.

High R&D spending erodes profits; cash flow remains negative

Securities Star noted that because the AI chip and general-purpose GPU markets have the characteristics of heavy upfront investment and a long product commercialization cycle, losses are the status quo commonly faced by domestic GPU companies.

Persistently high R&D spending is the main reason for Tian Shu Zhi Xin’s losses. From 2022 to 2025, the company’s R&D expenditures were RMB 457 million, RMB 616 million, RMB 773 million, and RMB 974 million, respectively, accounting for 241.8%, 213.1%, 143.2%, and 94.2% of revenue in the same periods.

Besides R&D investment, the company’s sales and distribution expenses as well as administrative expenses have also grown to varying degrees. In 2025, the two items mentioned above were RMB 152 million and RMB 482 million, respectively, representing year-over-year increases of 23.9% and 87.3%. The company’s revenue scale cannot cover the aforementioned expenses. Regarding the sharp increase in administrative expenses, Tian Shu Zhi Xin explained that it was mainly due to an increase in share-based payments made to administrative personnel and the growth in professional service fees.

The 2025 financial report shows that as the company implemented equity incentives, it recognized a large amount of share-based payment expenses. The share-based payments increased from RMB 248 million in 2024 to RMB 526 million in 2025, a rise of 112%, further squeezing the company’s profit space. If this portion is excluded, the company’s adjusted net loss was RMB 438 million, narrowing by 32.1% year over year.

In its financial report, the company mentioned that the current AI industry is facing scale implementation, and the development of large models is no longer limited to a simple expansion of parameter scale, but has entered a new stage driven by efficiency. How to significantly reduce training and inference costs while ensuring model capability has become the core proposition of competition in the industry. This shift puts entirely new requirements on compute infrastructure.

It is worth noting that in choosing between general-purpose and application-specific chip routes, the industry exhibits a pattern of “each has its pros and cons, and they contend with each other.” General-purpose chips have advantages such as high flexibility, but in terms of cost-effectiveness in specific scenarios, the general-purpose architecture has certain limitations. Application-specific chips have a longer development cycle and a more closed ecosystem, but for single-task inference in specific AI applications, they show clear advantages in energy efficiency ratio and unit cost.

In response to changes in industry trends, Tian Shu Zhi Xin is addressing this by implementing measures such as co-design of software and hardware, and launching its “Tongyang Series” robot brain products for robot and edge computing scenarios. This is also seen externally as the company’s attempt to seek a balance between generality and specialized efficiency. However, whether it can truly connect the intermediate path between general-purpose and specialized use cases, and achieve continuous conversion of technological advantages into market advantages, still needs to be tested over time.

Securities Star also noted that as the business scale expands, Tian Shu Zhi Xin’s inventories and trade receivables, as well as accounts receivable bills, rise in tandem, thereby affecting the company’s cash flow performance. In 2025, the corresponding amounts for the two above indicators were RMB 710 million and RMB 577 million, respectively, up 107% and 53% year over year, respectively. In the same period, the net cash used in operating activities was -RMB 1.162 billion, down 88.02% year over year. Over the past four years, the company has “bled” more than RMB 3.1 billion cumulatively. (This article was first published by Securities Star. Author | Li Ruohan)