Summary of Key Points
Recently, domestic GPU manufacturer Suoyuan Technology announced its upcoming listing on the STAR Market, with an issue price of 142.18 yuan, corresponding to a valuation of 61.2 billion yuan. However, the company’s annual revenue for 2025 is less than 1 billion yuan, with 83% of its income coming from its largest shareholder, Tencent, and it is still in a state of annual losses in the tens of millions. Three years ago, a technology company that relied on a single giant for funding and had not yet turned a profit could be valued at 60 billion yuan, all due to the scarcity of domestic GPUs. But now, with companies like Cambricon, MooreThread, and Muxi already listed, the domestic GPU sector has moved from the stage of determining availability to the stage of competing for dominance. The valuation logic for the entire sector is about to change completely. The previous myth of companies soaring to valuations of 200-300 billion yuan upon listing will soon face the test of actual performance.
Detailed and Easy-to-Understand Explanation
1. Tencent’s “Double Boost” for Suoyuan: Investing and Buying Products, but This Can’t Last Forever
Many think it’s a bad thing for a company to get 80% of its revenue from one customer, but for a startup in the GPU industry just starting out, having Tencent as a partner is like winning the lottery. GPUs are not just phone cases; they need to be integrated into servers, tested for stability over half a year, and prove their value in large-scale models before they can convert into actual sales orders. Most startups fail on the path of trying to attract customers. Tencent, as China’s largest consumer of computing power, has its own social media, gaming, large-scale models, and cloud services. By investing in Suoyuan and placing orders for its products, it’s like providing a guaranteed 80% of sales volume right from the start, saving the company the effort of finding customers. Suoyuan’s revenue has tripled in the past three years thanks to this partnership. However, this relationship is also a hurdle that must be overcome. If Tencent changes its procurement strategy or Suoyuan’s products can’t keep up with its rapid iteration, the company’s revenue could be halved. All domestic GPU companies currently have a high concentration of customers; a large order from a data center can consume half a year’s production capacity for a small company. The key question is not whether Tencent’s 83% share is high or not, but whether Suoyuan can leverage this partnership to secure orders from Alibaba, ByteDance, and other organizations. They can’t rely on this support forever.
2. Domestic GPUs Are No Longer “Rare”; the Capital Market Is Pushing for Real Results
It’s meaningless to discuss whether Suoyuan’s 61.2 billion yuan valuation is high in isolation. By comparing several domestic GPU companies, the logic becomes clear: Cambricon has already made a profit of 2.3 billion yuan in half a year; MooreThread’s revenue for the first half of the year was 1.7 billion yuan, just one step away from profitability; Muxi has already made money from its main business in the second quarter. Suoyuan is still losing money, relying on a 300% growth rate this year to maintain its appearance. A few years ago, the valuation logic for domestic GPUs was simple: NVIDIA was a bottleneck, and China needed its own GPUs. As long as a company could produce a working chip, capital was willing to pay for its expected profits over the next decade, focusing on who could bring out a usable product first and reap the benefits of scarcity. Now, with multiple companies in the market, the scarcity factor is no longer significant. The capital market’s question has shifted from “Does China need you?” to “Why would customers choose your products over others?” Companies not only have to compete for the market left by NVIDIA but also for market share, revenue, ecosystems, and customers. Those who fall behind will see their market value halved immediately.
3. Selling GPUs Is About Offering Convenience to Customers
Many wonder why no one can compete with NVIDIA, not because its chip performance is several times better, but because what it sells is more than just a silicon chip—it’s a complete software ecosystem called CUDA. For over a decade, Chinese AI engineers have learned to use CUDA, and all large-scale models are optimized for it, with data centers built according to NVIDIA’s standards. It’s like asking someone who has used Apple computers for 10 years to switch to Windows—everything would be unfamiliar, and the hassle of migration would be greater than buying 10 new computers. Even if domestic GPUs are half the price of NVIDIA’s, if customers had to spend several times as much on code adjustments and retraining engineers to migrate their models, no one would take the risk. Domestic GPUs focus on practical solutions: they don’t aim to match NVIDIA’s top-tier chip performance but instead minimize the hassle of switching chips. Using their chips requires almost no code changes, reducing migration costs from 100% to 10%, making it more attractive for customers. AI has moved beyond the stage of large-scale model training to the application implementation phase, where the focus is on efficiency and cost-effectiveness. Domestic GPUs start by addressing scenarios where NVIDIA’s products were seen as less profitable and unreliable, gradually gaining market share.
4. The 61.2 Billion Yuan Issue Price Is Just the Starting Point; a Valuation Bubble of Several Hundred Billion Yuan Is About to Burst
Previously listed domestic GPU companies have shown the market’s frenzy: MooreThread’s valuation was only 53.7 billion yuan at launch, but its stock price quadrupled on the first day, reaching a market value of 320 billion yuan; Muxi’s market value also soared from several billion yuan to 300 billion yuan. Suoyuan’s listing is even more dramatic, with only 4% of its shares tradable, meaning less than 2.5 billion yuan in circulating shares. With so many investors competing, a small amount of capital can drive the stock price sky high, potentially reaching 200-300 billion yuan. However, this price is unsustainable. If Suoyuan’s revenue is less than 1 billion yuan and it’s still losing money, with 80% of its income coming from Tencent, how much should Cambricon, which has already made a profit of 2.3 billion yuan in half a year, be valued? If MooreThread, with higher revenue and impending profitability, saw its market value drop to 180 billion yuan, does that mean the previous valuation was incorrect? The primary market may be optimistic about ten GPU startups simultaneously, but the secondary market can’t value all ten companies as if they will become the next NVIDIA. In the end, only a few companies will survive and dominate the market. Suoyuan’s listing could burst the valuation bubble for the entire sector. Previously, companies could inflate their valuations based on the concept of “domestic substitution,” but now they must compete on performance and market share. Those that can’t stand the test will see their market values plummet.
In summary, while Suoyuan’s 61.2 billion yuan issue price is a significant milestone, the current valuation bubble of several hundred billion yuan is likely to burst soon. The capital market is no longer just interested in the concept of “domestic substitution” but in real performance and market share. Only the most competitive companies will survive, and the valuation system for the entire sector will be reshaped.