Driven by the wave of artificial intelligence technology, the global demand for AI computing power has surged, and the AI server market has also entered a stage of rapid expansion.
AI servers are mainly used to process massive amounts of data required for deep learning workloads, especially during training and inference processes, which involve key requirements such as large memory capacity, high bandwidth, and system cache consistency. Compared with ordinary servers, AI servers usually adopt a "heterogeneous form", with stronger computing power, faster processing speed, and larger storage space, which can efficiently support high load and complex computing tasks.
00265a405abdeff0e077a8df0813c1f1.jpg
Wide market prospects
The traditional server market has an annual shipment volume of about 12 million units and a market size of 120 billion US dollars. Compared to traditional servers, AI servers demonstrate market potential far beyond traditional servers due to their higher standalone value and computing power requirements.
01. The market is growing rapidly as the "engine" of computing power, and the demand for AI servers is showing explosive growth. According to the latest report released by Counterpoint Research, the global server market revenue increased by 35% year-on-year in the second quarter of 2024, reaching $45.422 billion. Among them, the revenue share of AI servers has increased to 19%.
According to IDC's forecast, the global AI server market is expected to reach a size of 21.1 billion US dollars in 2023, and is expected to exceed 31.7 billion US dollars by 2025, with an average annual growth rate (CAGR) of 22.7% from 2023 to 2025. From the perspective of the domestic market, the size of the accelerated server market is expected to reach 5 billion US dollars in the first half of 2024, a year-on-year increase of 63%. Among them, GPU servers continue to dominate with a market size of 4.3 billion US dollars, while the market growth rate of non GPU accelerated servers such as NPU, ASIC, FPGA is astonishing, with a year-on-year growth rate of 182% and a market size of nearly 700 million US dollars.
02. Leading the way in shipment volume
Despite the overall slowdown in global server shipment growth, the shipment volume of AI servers continues to increase. In 2022, global server shipments increased by 4.8% to 13.6 million units; It is expected that the growth rate will slow down to 1.5% in 2023. According to TrendForce data, the shipment of AI servers is expected to exceed 1.2 million units in 2023, accounting for nearly 9% of the total global server shipments, with an annual growth rate of up to 38.4%. It is expected that by 2026, the shipment volume of AI servers will reach 2.37 million units, accounting for 15% of the total, and the compound annual growth rate (CAGR) is expected to remain around 25%.
Types and application scenarios
The AI server adopts a heterogeneous architecture, which can flexibly allocate computing power according to the needs of different applications and provide customized computing support for different scenarios. According to the type of processor, AI servers can be divided into combinations such as CPU+GPU, CPU+FPGA, CPU+ASIC, etc. GPU remains the preferred chip for data center acceleration, but non GPU chips such as NPU, ASIC, etc. are gradually occupying a place. According to IDC's forecast, the market share of non GPU chips will exceed 20% by 2025. In terms of performance, ASIC performs outstandingly in certain specific application scenarios, but its programmability and flexibility are weak; GPU has more advantages in deep learning training and general scenarios.
According to their usage, AI servers can be classified into training and inference types. Training servers typically have extremely high requirements for storage, bandwidth, and computing power, and are often designed with 8-GPU; Reasoning servers, on the other hand, have more flexible requirements for computing power. Depending on different business scenarios, they can be equipped with different chips such as GPUs, NPUs, or CPUs, and even achieve reasoning tasks through AI accelerators with PCLe interfaces.
1-25022010494b43.png
Global market distribution
North American dominance, Asia Pacific rise
In terms of regional market distribution, North America will occupy 38.2% of the global AI server market in 2023, mainly due to the leadership and early adoption of AI technology by tech giants such as Silicon Valley. The European market share is 27.5%, driven by government policy support and sustained investment in AI research and development. The demand for AI technology in the Asia Pacific region, especially in emerging economies such as China and India, is constantly increasing, and it is expected to become the fastest-growing region in the world by 2024, with a market share expected to reach 22.1%.
From the surge in computing power demand to the accelerated construction of AI infrastructure, AI servers will undoubtedly continue to lead the wave of technological innovation in the coming years. As a leading electronic component distributor in China, Faben Electronics has a rich product line and brand resources, including Taiyo Yuden, ROHM, Renesas, Nexperia, Nichicon, Wanyu, Fengbin, etc., which can provide high-value component products and services for AI server applications. In the future, Faben Electronics will work with more partners, fully utilize its technological capabilities and market resources, and jointly create value for the AI server industry chain.
|
