Summary of Key Points
TianShuZhiXin's revenue nearly tripled in the first half of 2026 (a year-on-year increase of 191.6%), mainly due to the surge in demand for AI inference chips. However, the gross margin plummeted from 50.1% to 17.2%. The company's first year without a loss was not due to core business profitability but rather from unrealized gains from stock investments amounting to 760 million yuan. While the company has a competitive edge in the general-purpose GPU segment, its market share of AI computing chips is limited, and the industry competition is fierce, indicating that the window of opportunity to prove its capabilities is narrowing.
Detailed Analysis
1. Revenue Doubling: Inference Chips as the “Growth Engine,” but Losses Caused by “Side Businesses”
- Growth Driver: Surge in Inference Chip Demand
With the adoption of applications like AI Agents, the demand for inference computing power has increased significantly. TianShuZhiXin's revenue from inference chips rose by 651.8% year-on-year, accounting for 69.2% of total revenue (compared to 26.8% last year). For example, ByteDance is in talks to purchase its inference chips, and if the deal goes through, it would become ByteDance's third domestic GPU supplier.
- The Truth Behind the Loss Turnaround: Investment Gains Saved the Day
The company's net profit for the first half was 106 million yuan, but 760 million yuan of this came from unrealized stock investment gains (acquiring a 0.27% stake in a listed company, resulting in a sevenfold increase in value). Excluding this gain, the core business actually suffered a loss of about 654 million yuan, which is more than the same period last year. The so-called “adjusted net profit of 246 million yuan” includes the company's own share-based payments and listing expenses, without deducting the investment gains, indicating a significant amount of inflated figures.
2. Gross Margin Halved: Low-Margin Products and Non-Core Business Impacts
- Product Structure Drags Down Margins
The unit price of inference chips is only 9,200 yuan, while that of training chips is 30,400 yuan (more than three times higher). The proportion of inference chips in revenue jumped from 27% to 69% in the first half, significantly lowering the overall gross margin. Even considering the core business, the gross margin of general-purpose GPU products decreased from 50.3% to 43.9%.
- Additional Losses
The company sold previously stocked server components at low prices, incurring a loss of 171 million yuan (with a gross margin of -1157%), and it also recognized an inventory impairment of 72 million yuan. These factors further reduced the gross margin by 26 percentage points to 17.2%. Excluding these non-core impacts, the core business's gross margin would be around 43.6%, still much lower than that of competitors (Moore Threads at 56.95% and Cambricon at 55.25%).
3. Industry Positioning: Fifth in General-Purpose GPUs, but Limited Market Share
- General-Purpose GPU Market: A Small Footprint
According to Frost & Sullivan, TianShuZhiXin ranks fifth in the general-purpose GPU market (with a 0.3% share), behind NVIDIA, AMD, Hygon Information, and Muxi. Among domestic manufacturers, it ranks third (with a 9.8% share).
- Overall AI Computing Chips: Limited Market Share
Including all AI computing chips (both dedicated ASICs and others), IDC data shows TianShuZhiXin in 9th place, with domestic competitors such as Huawei, Alibaba PingTouGe, and Baidu KunlunXin ahead. In 2025, China shipped 4 million AI acceleration cards, and TianShuZhiXin's annual shipments are only at a mid-range level among domestic manufacturers.
4. Competitive Pressure: Fierce Competition and Narrowing Window of Opportunity
- Crowded Field with Clear Leaders
Huawei Ascend has the largest shipments and a complete ecosystem. Alibaba PingTouGe and Baidu KunlunXin have the advantage of their own cloud services, ensuring domestic demand. Mid-tier companies like Moore Threads and Muxi have also gone public and have access to capital. Almost all manufacturers are competing for the inference chip market, making the field increasingly crowded.
- Capital Market Hierarchy: TianShuZhiXin in the Third Tier
Cambricon and Hygon Information have market values of around 600 billion yuan (first tier), Muxi and Moore Threads around 250 billion yuan (second tier), while TianShuZhiXin's market value is only around 100 billion Hong Kong dollars (third tier), indicating lower recognition from the market.
- Supplier Selection by Major Clients: Easy to Enter, Hard to Maintain a Share
Major downstream clients (such as ByteDance) typically choose multiple suppliers, making it difficult for TianShuZhiXin to secure a significant share, even if it gets a order. With existing suppliers like Huawei and Cambricon, TianShuZhiXin's role is more of a backup, making it challenging to establish a strong position.
5. Future Challenges: Moving from Surviving to Thriving
TianShuZhiXin's main issue is the lack of self-sustaining profitability in its core business, and the competition is intensifying. To break through, it needs to address two challenges:
- Enhance Product Competitiveness: Either scale up production of inference chips to reduce costs or develop products with higher margins (such as high-end training chips).
- Reduce Dependence on Investment Gains: Only by making the core business profitable can the company gain a solid position in the capital market and the industry. Otherwise, as the window of opportunity narrows, it may fall behind the leading players.
In summary, TianShuZhiXin has seized the opportunity in inference chips, but to achieve sustainable development, it must improve product competitiveness or make technological breakthroughs. Otherwise, its temporary success in turning losses into profits may be short-lived. The AI arms race continues, and the company that turns its core business into a cash-generating asset will be the one to succeed.