What Are the Components of AI Hardware? Current Major Trends

1 Composition of AI’s Body

From ChatGPT to autonomous driving, from AI painting to intelligent customer service, behind every intelligent system, there is a complete set of AI hardware systems silently supporting it.

Today, let’s break down what exactly constitutes the “body” of AI.

2 The “Skeleton” of AI Hardware: PCB

If we compare an AI system to a person, the PCB (Printed Circuit Board) is its skeleton.

All chips, capacitors, resistors, and connectors must be soldered onto this board.

Therefore, AI servers have very high requirements for PCBs:

Many layers (dozens of layers)

Fast signals (high-speed transmission)

Strong heat dissipation (high-temperature resistance)

The core material of the PCB is CCL (Copper Clad Laminate), which consists of three layers:

Material Function

Copper Foil Conductive

Resin Insulating

Electronic Fabric (such as M9) Enhances strength, controls dielectric properties

M9 electronic fabric is currently a high-end material used in AI servers, more heat-resistant and low-loss than ordinary electronic fabric, making it a standard for “AI-grade” PCBs.

3 The “Heart” of AI Hardware: Chips and Memory

The true computing power of AI comes from these types of chips:

GPU:Graphics + parallel computing, the main force for AI training

TPU:Google’s dedicated AI accelerator

NPU:Neural network dedicated chip

FPGA/ASIC:Customized AI chips, high efficiency

4 The “Veins” of AI Hardware: CPO and Interconnect Technology

High-speed data transmission is required between chips and servers.

Traditional electrical interconnects can no longer keep up with AI’s “heartbeat,” leading to the emergence of:

CPO (Co-Packaged Optics): packages optical modules and chips together, using “light” instead of “electricity” to transmit data.

Faster speeds (Tb/s level)

Lower power consumption

Smaller latency

5 The “Sweat Glands” of AI Hardware: Cooling Systems

AI chips generate too much heat, and air cooling can no longer meet current demands.

Thus, liquid cooling has become a new trend:

Cold plate liquid cooling (indirect contact)

Immersion liquid cooling (directly immersing the motherboard in liquid)

6 Understanding the AI Hardware Industry Chain at a Glance

Upstream Materials

├── Copper Foil (conductive)

├── Electronic Fabric (M9 grade)

├── Resin (insulating)

└── Silica Powder (filler)

Midstream Components

├── CCL (Copper Clad Laminate)

├── PCB (Printed Circuit Board)

├── Power Modules

├── Cooling Modules

└── Optical Modules (CPO)

Downstream Systems

├── AI Servers (such as H100 clusters)

├── Chip Packaging

├── Complete Cabinets (liquid cooling)

└── Data Centers (10,000 card level)

7 Conclusion

Recently, the market has been more about rotation, with storage chips and quantum technology receiving good feedback. In addition, with NVIDIA’s continuous rise and the easing of US-China relations, overseas lines continue to strengthen, while domestic replacements have not shown much performance. This is the current reality, and we must acknowledge the gap. Several major overseas companies announced explosive earnings tonight, and I remain optimistic about the AI hardware direction tomorrow. Besides CPO, other branches have opportunities for rotation, such as PCB, power supply, electronic fabric, copper foil, liquid cooling, CPO, etc. Just keep an eye on these directions tomorrow and avoid chasing high prices.

I hope everyone can like, recommend, and support this. Thank you all.

Leave a Comment