As the development of AI technologies such as self-driving cars and Chat GPT accelerates, the number of companies introducing AI servers for large-scale, high-speed data processing is expected to increase.

According to TrendForce announced on the 8th, AI server shipments are expected to increase by an average of 10.8% annually from 2022 to 2026 as cloud service providers (CSPs) make large-scale investments in AI server development.

It is estimated that AI servers equipped with general-purpose GPUs will account for close to 1% of global annual server shipments last year. This year, AI server shipments are expected to increase by 8% year-on-year due to increased demand for generative AI, etc.

In particular, the four major cloud companies, Google, AWS, Meta, and Microsoft, are expected to account for 66.2% of the total annual AI server demand in 2022. In China, ByteDance accounted for 6.2% in 2022, followed by Tencent, Alibaba, and Baidu with 2.3%, 1.5%, and 1.5%, respectively.

Currently, NVIDIA has about 80% market share in server GPUs, while AMD has about 20%. In the server GPU market used for AI-related computing, mainstream products are NVIDIA's H100, A100, A800 and AMD's MI250, MI250X series.

Among them, ChatGPT mainly adopts Nvidia's A100 and utilizes Microsoft Azure's cloud. Accordingly, Microsoft's AI server demand in 2023 is expected to reach a total of 25,000 units. Baidu is expected to reach a total of 2,000 units.

Also, according to TrendForce, the expansion of ChatGPT, an OpenAI investment by MS, is driving demand for HBM. As hyper-large AI requires high-performance hardware to process large amounts of data, demand for HBM is also expected to increase.

According to the survey, HBM currently accounts for about 1.5% of the total DRAM market. The major suppliers of HBM solutions are Samsung, SK Hynix, and Micron. Among them, SK Hynix is the only company that can mass-produce the HBM3 solution adopted by NVIDIA, so it is expected to become the dominant supplier of HBM3 solutions.

Demand for HBM solutions has increased significantly due to inventory build-up amid supply chain shortage concerns caused by the COVID-19 pandemic, but this demand growth is expected to slow down in 2023. TrendForce currently projects that the HBM solution market will expand by over 40-45% by 2025.

Meanwhile, AI servers require high-performance computing resources and large storage space for high-level data processing performance. Accordingly, the importance of large-scale data centers is being emphasized in order to operate AI servers. Since data centers provide computing environments including servers, storage, and network equipment, investment in data centers is also expected to accelerate.