You might not believe this, but the price-to-book ratio of Cxmt Corporation (688825) is now nearly on par with NVIDIA's (NVDA). As of the close on August 7, Cxmt Corporation shares were at 52.48 yuan, giving it a total market capitalization of approximately 3.51 trillion yuan. Its adjusted net assets after listing are about 139.36 billion yuan, resulting in a price-to-book (P/B) ratio of roughly 25 times. At the same time, NVIDIA had a total market cap of about 5.30 trillion US dollars, with a P/B ratio of approximately 27 times. One is 25 times, the other 27 times. The difference of 2 times is almost negligible. Of course, this is about the P/B ratio, not total market value. In terms of total market cap, Cxmt is about 500 billion US dollars, while NVIDIA is around 5.3 trillion US dollars, a difference of more than ten times. But in the P/B ratio metric, the market's premium for Cxmt is indeed aligning with NVIDIA. However, the gap between these two companies is far more than just tenfold.
First, look at the business: one builds AI chips, the other makes memory modules. What NVIDIA does cannot be replaced by anyone else globally. The training and inference of AI large models rely almost entirely on NVIDIA's GPUs. In the global AI chip market, NVIDIA holds a 70% share. Every time Jensen Huang sells an H100, it effectively cashes a large check from OpenAI, Google, and Microsoft. In fiscal year 2026, NVIDIA's net profit exceeded 100 billion US dollars. What Cxmt Corporation does is DRAM—memory modules. Phones, computers, and servers all need it, but the DRAM market has long been a highly concentrated red ocean. Samsung Electronics, SK Hynix, and Micron collectively control over 90% of the global DRAM market share. Although Cxmt has made rapid progress, climbing from 3% to 8% global share in the first quarter of 2026, it is still only the fourth player, far behind the top three. One is a near-monopoly "shovel seller" for global AI computing power, and the other is the world's fourth-largest memory module maker. The gap in business value is obvious.
Next, look at revenue: the disparity is even more glaring. In the first quarter of 2026, Cxmt Corporation reported revenue of 50.8 billion yuan and a net profit attributable to the parent company of 24.762 billion yuan. It earned nearly 25 billion yuan in a single quarter, or close to 3 billion yuan per day. But when placed in the context of the global semiconductor industry, the picture changes immediately. In the same period, SK Hynix's revenue was approximately 242.9 billion yuan, meaning Cxmt's revenue was only one-fifth of that. Micron Technology's DRAM revenue was about 277.6 billion yuan, with Cxmt at 18% of that. The more striking comparison is market value: based on the August 7 closing price, Cxmt is worth about 3.51 trillion yuan. SK Hynix's market cap is around 5.58 trillion yuan, making Cxmt 63% of it, or nearly two-thirds. With revenue only one-fifth of SK Hynix's, its market cap has reached two-thirds. This ratio is unmatched across the entire A-share market. Micron Technology's market cap is about 7 trillion yuan, and Cxmt is 50% of that—while its revenue is only 18% of Micron's, its market cap has reached half. Valuation has run ahead of performance, and it's run quite far.
Industry Rules: Why Storage Companies Use PB Valuation
To understand how extreme this valuation is, you need to grasp the rules of the storage chip industry. DRAM is a classic heavy-asset, strong-cyclical industry. Building a fabrication plant costs tens of billions of dollars, and equipment depreciation is astronomical. When prices rise, profits can multiply; when prices fall, huge losses are common. Therefore, analysts primarily use P/B (price-to-book ratio) to value storage companies, not P/E (price-to-earnings ratio)—because at the peak of the cycle, profits are too inflated, and at the trough, they are all losses, making P/E unreliable. What is the typical P/B for storage companies? Micron's average P/B over the past decade is only 1.66 times, with a median of 1.83 times. Even during the classic storage super-cycle of 2017-2018, Micron's P/B peak was only 3 times. Since the start of this year, foreign investment banks have been arguing for months that "storage is no longer a cyclical stock," which has pushed valuations up somewhat. Currently, Micron's P/B is about 10 times, and SK Hynix's is about 6 times. Samsung Electronics is worse off, with its stock price falling 23% in the past month, leaving its P/B at just 1.4 to 1.6 times. And Cxmt Corporation? 25 times. That's more than four times its peers and over a dozen times the industry leader.
The Same PB, A Completely Different Story
So, how do we explain NVIDIA's 27 times P/B? NVIDIA achieves this valuation through monopoly and growth. With a 70% share of the global AI chip market, NVIDIA virtually locks in most of the profits in this track. Additionally, AI computing demand is growing exponentially—Amazon's full-year capital expenditure for 2026 has been raised to 220 billion US dollars, most of which is going to AI infrastructure. NVIDIA is the sole "arms dealer" in this arms race. A high P/B essentially means the market is willing to pay for future excess profits. For Cxmt Corporation's 25 times P/B, what is the market paying for? One factor is import substitution. DRAM is one of China's largest single import categories by value, with an extremely low domestic substitution rate. Cxmt is currently the only domestic manufacturer capable of mass-producing DRAM. If it can carve out a piece of the market from the three giants, it represents a trillion-yuan market space. Another factor is scarcity. The A-share market previously had no pure-play DRAM manufacturing target, and Cxmt fills this gap. As a "unique" asset, the market is willing to give it a premium. But the problem is that the import substitution logic needs time to materialize, while the valuation has already been over-discounted.
The Hidden Concern of a Cyclical Turning Point
There is another detail easily overlooked. Cxmt's explosive performance in the first quarter was largely driven by the surge in DRAM prices. In early 2026, the price of DDR4 rose by more than 2000% in a single month. But how long can this increase last? Morgan Stanley has already warned that the growth rate of DRAM contract prices may peak in the fourth quarter of 2026. Once prices fall, can Cxmt's profits maintain this level? The storage industry has never escaped the cycle. When prices rise, everyone is a genius; when prices fall, book profits recede like the tide. Valuing stocks based on linear extrapolation at the peak of the cycle is one of the most dangerous habits in the A-share market.
Is It Expensive? It Depends on What You Believe
Back to the initial question: Is Cxmt Corporation's 25 times P/B ratio expensive? If valued as a mere storage cyclical stock—it is outrageously expensive. Micron at 10 times, Hynix at 6 times, Samsung at 1.5 times, and Cxmt at 25 times—this is inexplicable in any traditional valuation model. But if priced as "China's only DRAM manufacturer plus core import substitution asset"—then this price is the market voting with real money. Northeast Securities gives a valuation range of 3.2 trillion to 5.7 trillion yuan, and the current 3.5 trillion is still at the lower end of the range. For a new industry player growing several times annually, this is not entirely unreasonable. The root of the disagreement lies in which story you believe. Do you believe that DRAM is ultimately a cyclical industry, that prices will inevitably fall, and that the moats of the three giants are insurmountable? Or do you believe that import substitution is an irreversible trend and that Cxmt Corporation will eventually carve out its own path in China, the world's largest semiconductor market? Two different beliefs lead to two completely different valuations. The only thing certain is that Cxmt Corporation, with a 3.5 trillion yuan market cap and a 25 times P/B ratio, has already priced in the most optimistic expectations of the "import substitution" story. The rest is up to time.
Comments