📌AI Industry Chain Research Lab｜Issue 1

Many people follow AI, NVIDIA, and TSMC daily, and know terms like CPU, GPU, and HBM, but few can clearly explain the relationships between them.
Without understanding the semiconductor industry chain, it's impossible to make money from AI.
Some still can't figure out why some semiconductor companies do design, others manufacturing, and others only packaging.
Today, I'll explain these questions in 5 minutes, using one main thread:
How does a grain of sand become a chip?
By understanding this main thread, you'll not only grasp the semiconductor industry but also know where a company's value comes from.
🔔① What exactly is the relationship between semiconductor, chip, and CPU?
(Corresponding to Image 01)
Many people's first mistake when approaching semiconductors is mixing these three terms.
In fact, they form an inclusion relationship.
Semiconductor refers to the entire industry.
It includes all links: materials, equipment, design, manufacturing, packaging, and testing.
A chip is a product manufactured using semiconductor materials—essentially an integrated circuit with a large number of transistors.
And CPU is just one category of chips.
Besides CPU, there are GPU, memory chips, analog chips, RF chips, AI accelerator chips...
So remember one sentence:
Semiconductor is the industry, chip is the product, and CPU is just one type of chip.
Many people studying semiconductor stocks like to directly discuss a particular company.
But in reality, before discussing companies, it's more important to first build this industry chain map.
Otherwise, it's like analyzing a car company without knowing the parts of a car—easy to get lost.
🔔② Why is it called "semiconductor"?
(Corresponding to Image 02)
Materials in the world can be roughly divided into three categories.
The first is conductors, like copper, silver, aluminum. Electric current can pass through almost freely.
The second is insulators, like plastic, rubber, glass, which hardly conduct electricity.
Semiconductors lie between the two.
Their biggest feature is not "conducting a little electricity," but that we can artificially control whether they conduct or not.
The most commonly used material for modern chips is silicon.
Silicon itself is not a particularly good conductor, but after doping with elements like boron or phosphorus, its conductivity can be precisely controlled.
Transistors were invented using this property.
It can be said that without silicon, there would be no modern computers. Hence the whole industry is called the semiconductor industry.
🔔③ How does a grain of sand become a chip?
(Corresponding to Image 03)
The starting point of a chip is actually the most common quartz sand, which after high-temperature purification yields high-purity polysilicon.
But chips cannot be made yet at this stage.
Because the crystal arrangement inside polysilicon is disordered, causing electron movement to be easily disturbed.
So engineers use a process called the Czochralski method to slowly pull polysilicon into a single crystal silicon ingot. Only then can electrons move stably along the designed path.
Next, the silicon ingot is sliced into thin round wafers less than 1mm thick.
This is the most important basic material in the entire semiconductor industry—the wafer. Many mistakenly think wafers are chips.
Actually, they are not. A wafer is more like a blank sheet of paper.
All circuits are first drawn on this blank paper.
Finally, it is cut into individual chips.
So, if you see a company's business described as "silicon wafers" or "wafers," they are not selling chips yet—just the most basic raw material for manufacturing chips.
🔔④ How is a chip "carved out"?
(Corresponding to Image 04)
Having wafers is far from enough. What truly determines chip performance is the subsequent manufacturing process.
Many think chips are "produced."
Actually, to be more precise, they are carved out layer by layer.
First, the chip design company completes the circuit design. Then, the fab coats the wafer surface evenly with photoresist, and uses a lithography machine to "expose" the designed circuit pattern onto the wafer surface.
Which areas need to be kept and which need to be removed are already pre-designed.
Next, using etching equipment, the unwanted parts are gradually "corroded" away.
Then, through deposition, ion implantation, CMP polishing, and other processes, new materials are stacked layer by layer.
Then lithography, etching, deposition again...
Advanced chips often repeat this process hundreds of times.
Eventually, tens of billions of transistors are built on a silicon chip the size of a fingernail.
This is the true birth process of a chip.
At this point, a wafer has completed its most complex manufacturing steps.
But it still cannot be used directly.
Why?
Because it is still a "bare chip."
🔔⑤ Why can't chips be sold directly after they are made?
(Corresponding to Image 05)
After thousands of steps like lithography, etching, and deposition, a wafer is finally completed. But at this point, it still cannot be installed in a computer or a phone.
The reason is simple:
It's too fragile.
A real chip is only a few square millimeters to tens of square millimeters in size.
After being cut, it's just a piece of exposed silicon.
It has no protective layer, no pins, and cannot connect to the motherboard.
So it must go through two final steps:
Packaging and testing.
Packaging is not just "wrapping it up."
It also performs three important tasks:
First, protecting the chip.
Second, helping with heat dissipation.
Third, connecting the chip to external circuits.
Finally, after testing to confirm that performance, power consumption, and stability all meet requirements,
A chip truly ready for sale is born.
Many think packaging is just the final step.
In fact, in the AI era, advanced packaging has become one of the most important technologies in the entire industry chain.
Why?
Because GPUs are getting bigger, HBM is increasing, and chiplets are becoming more complex.
Packaging no longer just determines whether a chip can be used, but also the upper limit of chip performance.
So in recent years, advanced packaging has become one of the hottest directions in the industry.
🔔⑥ Why is the division of labor in semiconductor companies becoming more detailed?
(Corresponding to Image 06)
If you observe the semiconductor industry, you'll notice a very interesting phenomenon: almost no company can do everything.
Why?
The answer is two words:
Too expensive.
Building an advanced fab often requires hundreds of billions of dollars in investment. Developing a generation of advanced process nodes takes years.
Plus, each link—equipment, materials, processes—requires long-term accumulation.
So the industry has gradually formed a specialized division of labor, with each company concentrating its resources on the part it does best.
This is how today's semiconductor industry chain was formed.
🔔⑦ Why does AI boost the entire industry chain?
(Corresponding to Image 08)
Many think that the AI rally is just the NVIDIA rally.
In fact, that's only one link in the chain.
An AI server doesn't just have a GPU.
It also needs:
CPU for scheduling,
HBM for high-speed memory,
PCB for connections,
High-speed switches for communication,
Optical modules for transmission,
Advanced packaging to integrate everything together.
If any link fails, the entire AI server cannot function properly.
So for every dollar invested in AI, it's not just GPU manufacturers that benefit, but the entire semiconductor industry chain.
That's why in the past two years, we haven't only seen NVIDIA rise.
TSMC, Broadcom, Micron, SK Hynix, Samsung Electronics, Applied Materials, ASML, and other companies have also continuously benefited.
🔔⑧ When studying a semiconductor company, first answer one question.
(Corresponding to Image 09)
When we see a semiconductor company, don't rush to look at the PE ratio or the stock price.
First ask yourself:
Where does it stand in the industry chain?
Because position in the chain determines what it makes money from.
Materials companies make money from consumables.
Equipment companies make money from selling equipment.
Design companies make money from intellectual property.
Fabs make money from manufacturing capability.
Packaging companies make money from advanced processes.
Different positions mean completely different business models.
Understanding this makes the valuation logic of many companies very clear.
Final Words
Many people study AI by focusing on just one company.
But what truly drives the AI revolution is never a single company.
It's a complete industry chain spanning materials, equipment, design, manufacturing, packaging, servers, and cloud computing.
Understanding this chain means you no longer just see stock prices, but the underlying logic of capital flow throughout the AI era.
Next issue, we'll continue unpacking one of the most confusing topics:
CPU, GPU, NPU, FPGA, ASIC—what are the differences?
Why is AI training almost inseparable from GPU?
Why are inference chips beginning to diversify?
Where does the real competition in AI chips take place?