AI Servers: The Savior of the Supply Chain, Examining Key Industries

NVIDIA’s robust financial report reveals the true impact of AI on the technology industry, particularly in the AI server supply chain.

The surge in AI-driven demand has prompted research institutions to revise their estimates. According to TrendForce, an industry research firm, the shipment volume of AI servers (including those equipped with GPUs, FPGAs, ASICs, etc.) is projected to reach nearly 1.2 million units in 2023, with a year-on-year growth of 38.4%, accounting for nearly 9% of the total server shipments. TrendForce has even raised the compound annual growth rate for AI server shipments from 2022 to 2026 to 29%.

So, what sets AI servers apart from conventional servers?

AI servers are indeed part of the overall server market, but they are specifically designed for cloud-based AI model training or inference. In terms of specifications, AI servers, in the broad sense, refer to servers equipped with AI chips (such as GPUs, FPGAs, ASICs mentioned earlier), while the narrow definition includes servers with at least one GPU.

The reason AI servers have become the “savior” of the industry in 2023 is twofold. Firstly, the performance of various end markets has been significantly weak this year, and even the previously stable server market has experienced a downturn. The surging demand for expensive AI servers has not only generated buzz but has also directly driven revenue growth throughout the industry.

According to TrendForce’s estimation, AI servers will account for 9% of the total server market’s shipment volume this year, and it is projected to reach 15% by 2026. In terms of unit price, AI servers are priced approximately 15 to 20 times higher than the servers traditionally used by cloud service providers. The increased demand for computational power, power management, and improved cooling technology has directly contributed to the usage of server components.

When it comes to the key components of AI servers, they include GPUs, CPUs, memory, smartNICs, casings, motherboards, cooling systems, power supplies, and assembly testing. From a cost perspective, GPUs remain the most critical part, accounting for around 70% of the overall cost, while components like cooling systems, power supplies, and casings make up less than 1%.

Although NVIDIA, the leading supplier of GPUs, is the biggest winner in terms of cost, the increased usage of components and the upgrades in specifications and technology have still benefitted the Taiwanese supply chain, which has played a vital role in the server industry.

Looking at the upstream industry, the largest beneficiary is TSMC, which handles manufacturing for NVIDIA. Regarding power supplies, the demand from AI servers is two to three times higher than before, which undoubtedly brings positive news for power supply manufacturers such as Delta Electronics and Lite-On Technology.

While cooling systems account for less than 1% of the overall cost, they are critical for the operation of AI servers, prompting major server suppliers to invest heavily in this area. Currently, the leading providers of AI server cooling solutions include Auras Technology, AVC Technology, and Sunon, which offer 3D Vapor Chamber (VC) technology. As for assembly and manufacturing, the key players are traditional server suppliers such as Quanta, Inventec, Foxconn, Wistron, Wiwynn, Supermicro, and Gigabyte.

TrendForce analyst Frank Gong states that in recent years, the server industry has been moving towards standardization, and server suppliers have faced pressure from customers to diversify their supply chains. However, with the significant surge in AI server demand, it is driving not only revenue growth but also providing an opportunity for companies to showcase their technical advantages in the field of AI servers. Furthermore, the high level of complexity in AI server design and the substantial rise in customized requirements have resulted in improved customer stickiness, which is highly welcomed and anticipated by the supply chain.

(Photo credit: Google)