Explosive Growth and Technical Demands of AI Server PCBs: A Comprehensive Analysis

1. Explosive Growth in AI Server PCBs

In recent years, AI applications have experienced explosive growth. The exponential growth in demand for AI computing power from businesses and research institutions has directly driven the explosive growth of the AI server market. According to TrendForce data, AI server shipments are expected to reach nearly 1.2 million units in 2023, a year-on-year increase of nearly 38.4%. By 2026, the number of AI servers is projected to exceed 2.3 million. The rapid growth of AI servers is primarily driven by the following factors.First, the training and inference of large models require extremely high computing power. For example, the GPT-3 model has 175 billion parameters. Training such a model requires a massive amount of computing resources, and AI servers provide powerful parallel computing capabilities to meet the demands of large-scale model training. Second, the widespread application of AI in various industries, such as intelligent customer service, image recognition, and intelligent recommendations, has also prompted businesses to increase their investment in AI servers to improve their business efficiency and competitiveness. Furthermore, with the development of emerging technologies such as 5G and the Internet of Things, data volumes are exploding, and the demand for data processing and analysis is further driving the growth of the AI server market.

2. Special Requirements for PCBs in AI Servers

The performance requirements of AI servers differ significantly from those of traditional servers, placing higher demands on PCBs. Traditional servers are primarily used for data storage and routine computing tasks, with relatively low requirements for computing speed and data transmission rates. AI servers, on the other hand, need to process massive amounts of data and perform high-speed calculations to meet the real-time demands of AI algorithms.

To meet the high-performance demands of AI servers, PCBs require upgrades in terms of layer count, materials, and processes. Traditional server PCBs typically have 6-16 layers, while AI server PCBs typically have over 20 layers, sometimes even up to 30. This increased number of layers provides more wiring space, reduces signal interference, and improves data transmission stability and speed. AI servers require copper-clad laminates with low dielectric constant (Dk) and low dielectric loss (Df) to reduce signal transmission losses and enhance signal integrity. Furthermore, AI server PCBs require more advanced manufacturing techniques, such as backdrilling, resin plugging, and polyolefin (POFV), to enhance board reliability and stability.

3. Analysis of the Reasons for the Explosive Growth in AI Server PCB Demand

In AI servers, GPU modules are core components that deliver powerful computing power. NVIDIA's A100 and H100 processors, for example, are widely used in AI computing, and demand for PCBs supporting these processors has surged accordingly. To enhance inter-GPU communication, chips must be pre-embedded on GPU adapter boards, making GPU boards the primary driver of the growing demand for AI server PCBs. Compared to traditional servers, AI servers utilize multiple GPUs, significantly increasing the value of their PCBs. For example, the NVIDIA DGX A100 boasts a 5.6-fold increase in PCB value compared to a typical server, with approximately 94% of this increase coming from GPU boards.

The proportion of AI servers in data centers continues to grow, evolving from a limited number of applications to becoming a mainstream feature. According to relevant data, AI servers are expected to account for over 50% of all data center servers in the next few years. Data center infrastructure upgrades also include network and storage equipment, which also utilize PCBs, further expanding market demand for PCBs.