Summary of Key Points
Domestic AI chips have seen explosive growth in a context of severe computational power shortages: Multiple manufacturers have seen their revenues double year-on-year, with the demand for intelligent computing power exceeding supply by more than three times. However, companies are generally facing cash flow pressures (due to upstream costs for securing production capacity and delayed payments from downstream clients). The industry has divided into three main camps, and the focus of competition will soon shift to customer loyalty, supply chain capabilities, and technological approaches. Leading companies are expected to start making profits around 2027, marking a transition from general growth to a phase of value selection.
1. Why Have Domestic AI Chips Suddenly Become So Profitable?
The core reason is the gap between supply and demand:
- Demand Side: The United States is restricting China's access to the most advanced AI chips from companies like NVIDIA (e.g., the H100) and even attempting to limit the use of foreign computing resources. Meanwhile, the construction of domestic intelligent computing centers has surged, with the scale of intelligent computing power increasing by 177% year-on-year in the first half of the year, and by 303 EFLOPS in the second quarter alone (equivalent to the addition of many “AI supercomputers”).
- Supply Side: Domestic chip production capacity is still ramping up, but there are challenges in ensuring compatibility with cloud and model manufacturers. According to data from the China Academy of Information and Communications Technology, demand is growing three times faster than supply, leading to a surge in revenues for domestic chip manufacturers.
2. Profits Are on the Horizon, but Cash Flow Is a Concern
Although future profits are projected, companies are currently facing significant cash flow challenges:
- Profit Signs: Cambricon turned a profit last year, with a revenue target of 47.6-59.5 billion yuan by 2028 through equity incentives. Companies like Muxi and Suiyuan plan to become profitable between 2026 and 2027.
- Cash Flow Difficulties: Chip companies are experiencing financial strains due to capital being tied up at both ends: they need to pay upfront to secure production capacity from suppliers (e.g., wafer factories), and payments from downstream clients (cloud and model companies) are delayed. For example, Muxi had a net cash outflow of 1.3 billion yuan, and Moore Threads had a net outflow of 2.1 billion yuan. Huawei’s overall cash flow changed from a net inflow of 31.1 billion yuan last year to a net outflow of 39.8 billion yuan in the first half of this year, with inventory increasing by 45% (indicating a buildup of inventory).
3. The Three Major Camps: Who Are the Leading Players?
The domestic AI chip industry is divided into three tiers based on shipment volume and commercialization:
- First Camp: Huawei Ascend, Alibaba Pingtouge, Baidu Kunlun Core, Cambricon, and HaiGuang Information. All shipped over 100,000 chips last year. Apart from Huawei and Alibaba (whose valuations are not publicly available), the other three have market values exceeding 50 billion US dollars. Huawei Ascend, for instance, has established strong partnerships with operators like China Mobile, which placed large orders for Ascend devices in the first half of the year (totaling over 3.3 billion yuan). Reuters reports that Huawei plans to ship 750,000 950PR chips in 2026.
- Second Camp: Moore Threads, Muxi, Birun, Tianshu ZhiXin, and Suiyuan. These companies shipped fewer than 50,000 chips last year but have market values below 50 billion US dollars.
- Third Camp: Unicorns such as Xiwang and Dongfang SuanXin. They have not yet shipped large volumes but have valuations over 1 billion US dollars and are focusing on differentiated approaches, such as optimizing chip performance for specific AI tasks (e.g., improving the speed of answering questions).
4. What Will Determine Success in the Future?
The key factors will be customer loyalty and supply chain strength:
- Customer Loyalty: Companies that can establish close relationships with major clients (e.g., Alibaba, Tencent, operators) will have a competitive advantage. For example, Suiyuan is a major shareholder of Tencent (with 20% ownership) and 80% of its revenue comes from Tencent, making it a “directly affiliated” company with Tencent.
- Supply Chain Capability: The ability to secure essential components (memory, networking, etc.) is crucial, as it will determine the future direction of computing power. For instance, Alibaba Pingtouge’s next-generation chips emphasize strong interconnectivity, which could replace traditional large-scale training solutions. Huawei Ascend 950DT covers the entire process of large-scale model training and inference.
- Technological Approaches: The third camp’s chips, designed for specific use cases (e.g., AI chat, image generation), may have opportunities to gain a foothold in niche markets due to differentiated requirements.
5. Variables That Could Change the Landscape
Several key factors could alter the current industry landscape in the coming years:
- Alibaba Pingtouge: As the largest cloud provider and leader in open-source models in China, Pingtouge’s chips can integrate well with Alibaba’s ecosystem. If it goes public or increases production, it could challenge the leading camps.
- Huawei Ascend: With the launch of the 950DT chip in the second half of the year and its large production capacity, its position in the first camp is likely to be solid.
- Third Camp’s Chips: As AI applications shift from large-scale training to everyday inference (e.g., intelligent customer service, video generation), chips optimized for inference could see increased demand, potentially giving these startups an opportunity to rise.
By 2027, the industry will not experience a complete reshuffle but rather a phase of value selection. Demand will continue to grow, but company values will diverge, with some firms thriving through strong customer relationships and supply chain capabilities, while others may fall behind due to technical or financial constraints. It will be only when the industry growth slows down that true competition will reveal the winners and losers.
(The entire analysis is written in plain language, making it easy for non-financial professionals to understand the current state and future prospects of domestic AI chips.)