Why is GigaDevice, a company with no in-house manufacturing operations, valued at 500 billion yuan?

I noticed someone went to the interactive platform to ask the secretary of the board of GigaDevice Semiconductor: Do you use hafnium metal in your DRAM? How much do you use? Where do you buy it from?

The secretary of the board was very polite. His response went around in circles, but the core message was just one sentence: Since the day we were founded, we haven't been involved in manufacturing. Others handle everything from wafer production, packaging to testing.

01

The person who asked this probably thought GigaDevice had its own factory. To be honest, those who have held the stock for half a year but still can't tell what it sells probably think the same way.

There are three types of people in the chip industry.

The first type is like Samsung, which is the most vertically - integrated; from drawing blueprints, operating machines to packaging and shipping products, they handle the entire process by themselves. It's like a workshop that goes from growing cotton to selling ready - made clothes.

The second type includes companies like Foresee and Biwin, which are at the other end of the supply chain; they buy ready - made memory chips and assemble them into memory sticks and solid - state drives for sale. They are like tailors.

GigaDevice belongs to the third type: it only draws blueprints.

It determines what the chips look like. After drawing the blueprints, it hands them over to foundries to produce, and then sells the products under its own brand. A company of more than two thousand people doesn't own a single lithography machine. It's really a light - asset company.

With such a division of labor, for a company that only draws blueprints and doesn't operate machines, most of its fate depends on one thing: What do you draw?

When Zhu Yiming founded GigaDevice in 2005, it wasn't the turn of a startup in Zhongguancun to draw the blueprints for mainstream memory like DRAM and NAND. He chose SRAM, and then NOR Flash.

The only common point is that the market size is small. Samsung can see it but doesn't bother to get involved.

What are these two things? Let me give you a brief introduction:

SRAM is like a scratch paper for chips; when a machine is working, it needs to temporarily record some numbers in the middle of calculations. After use, it erases the data, and all information is lost when the power is off. It's really fast, but it's expensive and can't store much data.

NOR Flash is like an alarm clock for devices. When you unplug and then plug in your router, it recovers in a few seconds, thanks to this small chip inside. The startup process is written in it. Once powered on, the machine follows the instructions.

It retains data when the power is off and costs only a few dollars each. It can be found everywhere in TVs, headphones, and cars.

A scratch paper and an alarm clock together make up a market worth two or three billion US dollars a year. Next to them, the DRAM and NAND markets are like two huge granaries, worth more than 160 billion US dollars a year. Samsung doesn't want to play in this small market.

A turning point occurred around 2008. Samsung thought the profit from NOR was too thin and strategically withdrew from the market. The market leader, Spansion, was hit by the financial crisis and entered bankruptcy protection.

The people at the table got up and left, but the food was still hot. GigaDevice took their seats.

In 2013, it made the same kind of choice; it released the first domestic 32 - bit MCU with a Cortex - M3 core. It's the small chip that controls household appliances, electricity meters, and charging piles. The model naming and pin definitions are all based on STMicroelectronics' STM32.

You don't need to understand these technical terms or remember them. Just know that when engineers replace the chip, the code can still run almost the same. The tables of the giants are too high, and it entered from the side of the table legs.

After making such choices for twenty years, what has it achieved? I checked. According to Frost & Sullivan's statistics based on 2024 sales:

In NOR Flash, it ranks second globally, with a market share of 18.5%; in SLC NAND, a small - capacity flash memory, it ranks sixth globally; in niche DRAM, which specializes in small - capacity memory, it ranks seventh globally; in MCU, it ranks eighth globally and first in the Chinese mainland.

None of its four product lines ranks first globally, but all four are in the top ten. Individually, none of them is very impressive, but when put together, it's quite interesting: Among the companies that draw blueprints in the world, there are very few that have all four lines in the top ten.

In terms of revenue, it's more intuitive. In 2025, its revenue was 9.203 billion yuan. Memory chips accounted for just over 70%, MCU about 20%, and the rest were sensors and analog chips.

By choosing to draw blueprints that large companies don't bother with, it has built a business worth nine billion yuan.

However, blueprint - drawing companies have an inherent problem: You don't have the final say. No matter how good your blueprints are, you need someone to produce according to them. Whoever controls the production equipment holds half of your fate.

GigaDevice has suffered a big setback in this regard and has also made a move that outsiders don't quite understand. That production equipment is called CXMT.

02

Actually, I need to be more accurate; CXMT only produces DRAM for GigaDevice. For NOR, NAND, and MCU products, they each have their own production equipment, which has nothing to do with CXMT.

With this scope defined, let's answer the question that has been hotly debated on the stock forum: If I buy GigaDevice, is it the same as buying CXMT?

This has to start with a deal that didn't go through.

In 2008, during the financial crisis, GigaDevice was on the verge of collapse. At that time, an American memory factory called ISSI came to acquire it. GigaDevice didn't sell and managed to hold on.

Seven years later, ISSI was privatized by a group of Chinese capital, delisted from NASDAQ, and put into a shell company called Beijing Sigmastar.

One more year later, in August 2016, GigaDevice was listed on the Shanghai Stock Exchange. Just one month after the listing ceremony, it suspended trading. The planned deal was quite dramatic: to acquire Beijing Sigmastar for 6.5 billion yuan.

The company that wanted to buy it in 2008 became its target.

For a blueprint - drawing company to grow, buying others' blueprints is the shortest way. Sigmastar happened to have the blueprint that GigaDevice was most interested in: DRAM, the chip that plays a major role in memory sticks.

The deal was negotiated for more than a year, and in August 2017, it fell through.

Who sabotaged the deal? One of ISSI's suppliers directly said: If you merge with GigaDevice, I have the right to stop supplying. According to the reports at that time, this supplier was Nanya Technology, a memory factory in Taiwan, China.

How could a supply contract veto a 6.5 - billion - yuan deal? GigaDevice explained everything clearly at the investor briefing: ISSI's main product is DRAM, and there is no factory in the Chinese mainland that can produce DRAM on a contract basis. If you want to change the factory, you have to wait for two years just for customer certification.

Let me put it another way: You can afford the blueprints, but the production equipment is in someone else's hands. If they don't produce for you, the blueprints are just a piece of paper.

A blueprint - drawing company's attempt to acquire another blueprint - drawing company was vetoed by a piece of production equipment. It was a serious setback.

Two months later, in October 2017, GigaDevice announced a cooperation with Hefei Industrial Investment to build a 12 - inch memory wafer factory. The project was worth about 18 billion yuan. Hefei Industrial Investment would cover four - fifths of the cost, and GigaDevice would contribute about 3.6 billion yuan. The terms stated that the output would be given priority to meet GigaDevice's needs.

This project later got the name CXMT. In July 2018, Zhu resigned as the general manager of GigaDevice, only keeping the position of chairman. He went to Hefei and became the chairman and CEO of CXMT.

Now it's clear:

CXMT is not a subsidiary of GigaDevice, nor is it a factory under GigaDevice. With the same founder, outside the listed company, it built a piece of production equipment for China's DRAM industry and for GigaDevice.

The word "outside" is the answer.

There are two "pipes" connecting the two companies. One is thin, and the other is thick.

The thin one is equity. GigaDevice has invested a total of 2.3 billion yuan in CXMT, resulting in a shareholding of about 1.8%. Whether CXMT goes public and how high its valuation soars, this investment is recorded under "Other equity instrument investments" in GigaDevice's financial statements, and its fluctuations don't affect the income statement.

Those who expect GigaDevice's profits to skyrocket once CXMT goes public have got the "pipes" wrong.

The thick one is business. GigaDevice gives the blueprints for niche DRAM to CXMT. After production, CXMT sells the products under GigaDevice's brand.

I looked into the volume of the thick "pipe": The related purchases between the two companies were 764 million yuan in 2023, and the company expects it to reach 5.711 billion yuan in 2026. In four years, it has increased by more than six times.

Let me also clarify a confusing point. In the early days, when CXMT was just starting out, it used GigaDevice's channels to sell its own products. This was called consignment sales, which stopped in the second half of 2023. Now the situation is the opposite. CXMT produces for GigaDevice. There are still people who get the direction of who is working for whom wrong.

Some people must be asking: One person is the chairman of both companies, and the related purchases are increasing year by year. Isn't this just moving things from one hand to the other?

All these things are done transparently. For related transactions, Zhu abstains from voting at the GigaDevice board of directors. The independent directors hold a special meeting first. The amount and pricing principles are publicly announced and can be easily checked.

As for what this growth rate means, everyone has their own interpretation.

When CXMT goes public, these two "pipes" will remain as they are. The thin "pipe" is for riding on the valuation, not affecting the profit. The thick "pipe" is tied to the production capacity and produces for you.

03

Okay, we've clarified the "pipes" issue. What about the money? GigaDevice's net profit in the first quarter increased by five times. Where did this money come from? To understand GigaDevice in 2026, we need to look back at GigaDevice in 2023.

I looked at its annual report for that year: The annual net profit attributable to the parent company was 161 million yuan, a year - on - year decrease of 92%.

Not only did it lose money, but it also made an asset impairment provision of about 612 million yuan that year. The most significant cut was in goodwill. The fingerprint chip company SenseInfo, which was bought for 1.7 billion yuan in 2019, had its book value reduced by about 380 million yuan. It was originally part of the product line puzzle, but when the market turned cold, it became a big hole in the financial statements.

What big event happened that year? Nothing. It was just that the industry cycle had entered the downward phase. Downstream customers stopped placing orders, and the price of products kept falling.

Blueprint - drawing companies have a small advantage during the downward cycle. Since they don't own the machines, they don't have to bear the depreciation costs. Big companies like Samsung and SK Hynix, which operate dozens of production lines, burn money every day when the production lines are shut down. GigaDevice doesn't have this burden.

However, when the product price falls, all blueprints have to be sold at a low price. The gross profit is still severely affected. Although a light - asset company can avoid a heavy fall, it can't avoid hunger.

The company that was in such a difficult situation in 2023 and the one with a five - fold increase in net profit in the first quarter of 2026 are the same. The customers for the blueprints haven't changed. It still sells the same products: scratch paper (SRAM), alarm clock (NOR Flash), and small - capacity memory.

What has changed is the external environment. In the past two years, AI has heated up the computing power market. The gross profit of producing HBM, a high - end product, is several times that of producing ordinary products. Samsung, SK Hynix, and Micron have shifted their production capacity one by one. No one wants to produce the old products anymore.

In the second half of 2025, overseas large companies withdrew from the 2D NAND market, creating a big gap in the supply of small - capacity flash memory. The same is true for niche DRAM. Large companies withdrew even faster. Since the second quarter of 2025, both the volume and price of GigaDevice's products in this line have increased.

Look, the seats are empty again. In 2008, it started from this opportunity. Seventeen years have passed, and the same scenario is playing out in the market with the same players.

At this point, the cyclical nature of this company becomes apparent. It's a bit different from ordinary cyclical stocks.

What's the difference?

In the case of pork stocks, at least the supply and demand within the industry determine the cycle. The markets that GigaDevice serves, such as set - top boxes, household appliances, industrial control, and automobiles, have been relatively stable in terms of market size in recent years. There is no such thing as a sudden increase.

Its ups and downs depend entirely on where the production capacity of the giants is. When the giants enter the market, it has to stand aside. When the giants leave, it fills the empty spaces. Whether it has a good time or not depends on whether the giants move their production equipment.

Zhi Yuan calls this a "shadow cycle": The light source is in someone else's hands, and it is the shadow on the wall.

To be fair, it's not entirely a shadow. The MCU product line meets the real demand from the industrial and automotive sectors. The significant increase in the shipment volume in the first quarter has nothing to do with the product price. There are real orders. This product line has a weak cycle and is relatively stable, but it's too small to support the whole company.

What really matters is the length of the pendulum's rope. The giants shifted their production equipment to produce HBM because the gross profit there is really high. No one wants to turn down money.

When the HBM production capacity is full and the price drops, and the giants shift their production equipment back to produce the old products, it's just a matter of a production schedule for Samsung. The old product market is very small. Once a giant's production equipment returns, GigaDevice has to give up a share.

It has no veto power over this. It can't control when the "wind" stops.

04

The market obviously knows this. But the current price doesn't seem to reflect this knowledge.

What's really inside the price tag? Let me show you. As of the close on July 2nd, the market value of GigaDevice Semiconductor has dropped to about 487.6 billion yuan. Its annual net profit attributable to the parent company in 2025 was 1.648 billion yuan.

With these two numbers in front of you, use your calculator. Don't rush to scold after you're done.

This price may not be completely irrational. The problem is that it's not the price of a single thing. Let me break it down for you. This single price tag includes three things. The shelf - lives of these three things are very different.

The first thing: The money from the cycle.