What makes Cambricon's valuation reach one trillion?
The STAR 50 Index has soared rapidly. Who drove up the market? It's a company called Cambricon. During intraday trading, its stock price rose by over 8%, reaching 1,613 yuan, and its total market capitalization exceeded one trillion yuan.
After a decade of the STAR Market, it is the first stock to reach a market capitalization of one trillion yuan. It is a company specializing in chips.
01
540 million years ago, an extremely extraordinary event occurred on Earth. For billions of years before that, life forms were basically single - celled organisms, with little variety.
Suddenly, within a time frame of less than 20 million years, the vast majority of animal phyla emerged all at once. Trilobites, anomalocarids, and those strange ancient creatures you've seen in museum displays all appeared simultaneously, as if they had made an agreement.
The scientific community calls this the "Cambrian Explosion."
Do you think it was due to a sudden genetic mutation of a certain species that gave it an edge? No way. It was a change in the Earth's environment. When the ice caps melted, the oxygen content in the seawater increased significantly, and a large area of new shallow - sea territory emerged. With such a change in the aquatic environment, it was hard for life not to thrive.
In 2016, two brothers from Jiangxi, who graduated from the Juvenile Class of the University of Science and Technology of China, founded an AI chip company in Beijing named "Cambricon."
The older brother, Chen Yunji, is in charge of chip architecture, while the younger brother, Chen Tianshi, focuses on AI algorithms. One deals with hardware, and the other with software. They are betting on the same thing: that artificial intelligence will experience its own explosion, just like life did 540 million years ago.
Back then, AI was not very popular, and making specialized chips for AI was even less so. Chen Tianshi once mentioned that they couldn't even secure a research grant of 200,000 yuan because no one thought this venture would succeed.
At this point, I need to explain what Cambricon actually manufactures.
Both your mobile phones and computers have processors. The most common CPU is a general - purpose device that can handle various tasks. Later, NVIDIA transformed the GPU, originally used for rendering game graphics, into the main force for AI computing.
This device is good at performing a large number of repetitive mathematical operations simultaneously and can run AI models very quickly. Currently, AI companies around the world are scrambling for NVIDIA's GPUs.
Cambricon has taken a third path. It doesn't produce general - purpose GPUs but instead designs processors specifically customized for AI.
Let me give you an analogy:
A GPU is like a Swiss Army knife that can handle various scenarios, while the processors made by Cambricon are more like scalpels, designed to perform a single task extremely well, with lower power consumption and higher efficiency.
The closest ordinary people got to Cambricon was probably in 2017.
That year, Huawei released the Kirin 970 chip. For the first time, Huawei phones could "recognize" whether the photos you took were of landscapes or human faces. The AI processor responsible for this recognition was provided by Cambricon.
Ten years have passed. Cambricon's chips have grown from small modules in mobile phones to large - scale devices in cloud data centers.
The new - generation chip, MagicMind 690, which went into mass production at the beginning of this year, can already run large models with trillions of parameters. ByteDance, Alibaba, and Tencent are all testing or deploying it.
I checked, and it has been exactly ten years since the company was founded.
The founders once said that Cambricon's chips aimed to be the "stepping stones for future AI applications." Ten years later, this stepping stone has reached a market capitalization of one trillion yuan, a result that even the founders themselves might not have anticipated back then.
However, the real story worth telling about this company is hidden within these ten years because there was a period when the company almost went out of business.
02
I dug into its historical financial reports. In 2017, Huawei almost supported Cambricon entirely. Huawei's purchases accounted for 98% of Cambricon's revenue. For a startup company, getting tied to a giant like Huawei just one year after its establishment seemed like a stroke of good fortune, right?
However, this good fortune didn't last even two years.
In 2019, Huawei unveiled its self - developed Da Vinci architecture and started producing its own NPUs. Cambricon received a notice saying that they were no longer needed.
How did they part ways?
There are three versions in the industry. One version is that Huawei wanted an exclusive partnership, but Cambricon firmly insisted on being a neutral supplier and refused. Another version is that Huawei thought Cambricon's products were not good enough. The third version is that Huawei never intended to rely on others in the long run, and this cooperation was just a transition.
It doesn't matter which version is true.
The consequence was obvious: Huawei's share of Cambricon's revenue dropped from 98% to 15%, and the terminal IP business line basically shut down.
What was even more fatal was that not only did Huawei stop cooperating, but it also became a direct competitor of Cambricon. The product line of Huawei's Ascend chips almost completely overlapped with that of Cambricon.
Since 2019, Cambricon has fallen into a long period of darkness. Let me show you its financial situation to give you a sense of the desolation of entrepreneurship:
In 2020, it lost 435 million yuan; in 2021, 825 million yuan; in 2022, 1.257 billion yuan; in 2023, 848 million yuan; and in 2024, it was still in the red with a loss of 452 million yuan. In five years, nearly 4 billion yuan was burned.
The year 2022 was the darkest. The U.S. Department of Commerce added Cambricon to the Entity List, and its stock price dropped to 46.6 yuan, with a market capitalization of only over 20 billion yuan. Compare that with today's one - trillion - yuan market capitalization; it was a 50 - fold difference.
In the same year, early - stage investors who had accompanied the company since the angel round one by one fled. Yuanhe Origin, SDIC Venture Capital, and Paleozoic Venture Capital basically reduced their holdings and exited completely around that time. The verdict of the capital market was straightforward: the story couldn't continue.
There is a detail that few people mention:
In 2022, Cambricon's revenue was 729 million yuan, while its R & D expenses reached 1.523 billion yuan. I calculated that the R & D expenditure was more than twice the revenue. For every yuan earned, two yuan was invested in R & D. Losses were inevitable. This was not a business failure but a desperate bet on something.
The MagicMind 590 chip was ready around 2023, with performance equivalent to 80% of NVIDIA's A100. Logically, it should have been put on the market.
I checked the data for 2023. The situation was cruel: that year, the sales volume of Cambricon's cloud - based chips plummeted by 88%. The chips were produced, but no one bought them.
You may ask why.
In 2023, large companies could still obtain NVIDIA's products. Although the H100 was restricted, the special - edition H20 was still available. As long as NVIDIA's products were on the market, few customers were willing to take the risk of switching to domestic products. It had nothing to do with whether Cambricon's products were good or not. The key was that there was no compelling reason for customers to use them.
Two years later, the situation completely changed.
It became increasingly difficult for NVIDIA's products to enter the market, and the supply of the H20 was also restricted. Large companies were finally left with no choice. The same MagicMind 590 chip with the same performance parameters suddenly became a best - seller.
In 2025, Cambricon's full - year revenue was 6.497 billion yuan, and its net profit was 2.059 billion yuan. It was the first time in five years since its listing that the company made a profit for the whole year. In the first quarter of this year, the performance was even more impressive, with a single - quarter revenue of 2.885 billion yuan and a net profit of 1.013 billion yuan. For the first time, the operating cash flow turned positive.
It took three years for the stock price to rise from 46.6 yuan to 1,613 yuan and for the market capitalization to increase from 20 billion yuan to one trillion yuan.
Looking at this from the other side, it is quite counter - intuitive: in 2023 and 2025, Cambricon basically had the same chip in hand. What changed was the environment around the chip.
If you grow eyes in a completely dark room, are they useful? No. It is only when the lights are turned on and the oxygen concentration reaches a critical point that the species with eyes suddenly become valuable.
So, stop just focusing on Cambricon. Take a look at the "seawater" under its feet and see what has changed.
03
Three years ago, NVIDIA accounted for 95% of the Chinese AI chip market.
What does this number mean?
Almost all of China's AI computing power was built on the foundation of an American company. For large - model training, voice - assistant responses, and intelligent short - video recommendations, in most cases, the chips running in the background were NVIDIA's.
I checked the latest data from IDC, and in 2025, this number dropped to 55%.
In just over a year, 40 percentage points disappeared. Did NVIDIA's technology regress? No. Its products couldn't enter the market.
The United States imposed one round of restrictions after another. First, it banned the A100, then the H100, and finally restricted the supply of the H20, a specially - made version for the Chinese market. By the way, this data is the latest, and the data for 2026 is not available yet.
The shelves are empty, but the demand won't wait.
Tencent invested 18 billion yuan in AI in 2025 and plans to more than double it to over 36 billion yuan in 2026. Alibaba and ByteDance are also increasing their capital expenditures significantly. All this money will eventually be used to purchase rows of chips for data centers. If NVIDIA's chips can't be bought, we can't just leave the data centers idle, right?
This is the change in the "seawater." The "continental shelf" vacated by NVIDIA is so large that there is room for all domestic "species" to stand.
I looked at the shipment volume of domestic manufacturers:
In 2025, Huawei's Ascend chips shipped 812,000 units, ranking first among domestic manufacturers and capturing about 20% of the market share. Alibaba's T - Head shipped 265,000 units. Baidu's Kunlun Chips and Cambricon each shipped 116,000 units, tying for the third place among domestic manufacturers.
Pay attention to one thing:
It is more than a dozen companies working together to fill this gap. Moore Threads, Muxi, Biren, Suiyuan, and Daysci all went public in a cluster from the end of 2025 to the first half of 2026. People in the industry summarized the situation as "one super - large company, several large companies, and four rising stars."
This scene is exactly the same as the explosion 540 million years ago. There is no single species dominating; instead, an entire ecosystem is simultaneously activated.
There is also another force accelerating the development: domestic large - scale models.
In April this year, DeepSeek released its V4 model with trillions of parameters. Remember that a year ago when the V3 model was launched, the industry was still arguing about "whether domestic chips could run large - scale models."
This time, take a look. On the day of the model release, Huawei's Ascend, Cambricon, Hygon, Moore Threads, Muxi, Kunlun Chips, and T - Head were all compatible with the model.
On that day, Day 0, everything ran smoothly. Domestic models ran on domestic chips. The compatibility of domestic chips with domestic models made them more attractive, and more customers were willing to pay. The data from these customers, in turn, provided more training data for the models.
Once this positive - feedback loop starts, the market will drive itself. So, Cambricon's one - trillion - yuan market capitalization is, in fact, one of the results of the awakening of the entire domestic AI chip ecosystem.
04
After explaining the above, there is actually a more confusing thing. Cambricon ranks third in terms of domestic shipment volume, which is even less than a fraction of the first - ranked company. However, its market capitalization is one trillion yuan today, the highest among domestic AI chip companies.
Think about it. How can we understand this discrepancy?
I calculated that the market's valuation of Cambricon includes at least three factors.
The first is the realization of performance. After eight years of losses, the company finally turned a profit. The power of this turning point is astonishing. Goldman Sachs has directly set the target price at 2,406 yuan, and Morgan Stanley has set it at 1,528 yuan. The stock price has already exceeded Morgan Stanley's benchmark price during intraday trading today.
The second is the scarcity of its ecological niche.
Huawei's Ascend chips have a dominant position in terms of shipment volume, but Huawei is not publicly listed. T - Head belongs to Alibaba, and Kunlun Chips belong to Baidu, and they are also not listed. You can't buy a single share of their stocks in the capital market.
Moore Threads is listed and follows the general - purpose GPU route. Hygon Information is also listed, and its main business is CPUs and DCUs.
If you want to buy a "pure, independent third - party AI chip company" that is also profitable in the A - share market, after a thorough search, there is really almost only Cambricon left.
This scarcity gives it an additional valuation premium. Morgan Stanley's scenario - based valuation is a good example:
In the optimistic scenario, the target price is 2,887 yuan, while in the pessimistic scenario, it is 770 yuan, a nearly four - fold difference. How much is a company worth? The market is actually betting on how much an ecological niche is worth.
The third is a question mark of 373 times.
Cambricon's current trailing P/E ratio is around 373 times, while NVIDIA's is 60 times and Kweichow Moutai's is 21 times. Even if you have confidence in it, this number is still a big test for investors' nerves.
Those who are bullish on the stock have also calculated that the 373 - times P/E ratio is obtained by dividing the current market capitalization by the profit of the past year. However, institutional investors are betting on the profit two years later. If the net profit reaches over 12 billion yuan in 2027, the P/E ratio corresponding to the same stock price will drop to around 80 times, which is not that outrageous.
Those who are bearish also have a point. The customer concentration is too high. The top five customers contribute nearly 90% of the revenue. The annual report doesn't disclose who they are, and the market speculates that they are ByteDance, Alibaba, and other leading companies, but no one has confirmed this.
If one of the large companies manages to develop its own chips or cuts its budget one day, the company's performance could decline sharply.
In my opinion, the 373 - times P/E ratio is composed of three factors: technical strength, scarcity premium, and the bonus of the overall market