The year 2023 is expected to witness a significant surge in shipments of AI servers worldwide, according to TrendForce.
Approximately 1.2 million units of AI servers, equipped with GPUs, FPGAs, and ASICs, are projected to be delivered, reflecting a robust growth rate of 38.4 percent. This increase aligns with the growing demand for AI servers and chips, indicating that AI servers are set to constitute nearly 9 percent of total server shipments, with a projected increase to 15 percent by 2026.
TrendForce has revised its Compound Annual Growth Rate (CAGR) forecast for AI server shipments between 2022 and 2026 to an ambitious 22 percent. Furthermore, there is an impressive 46 percent increase anticipated in AI chip shipments for 2023.
NVIDIA’s GPUs currently dominate the AI server market, capturing an impressive market share of 60-70 percent. Following closely behind are ASIC chips developed independently by Cloud Service Providers (CSPs), which claim over 20 percent of the market.
NVIDIA’s market dominance can be attributed to several crucial factors. Firstly, the demand for NVIDIA’s A100 and A800 models is high among both American and Chinese CSPs. The demand curve for the second half of 2023 is expected to gradually shift towards the newer H100 and H800 models, which carry a higher Average Selling Price (ASP) of 2-2.5 times that of the A100 and A800, enhancing their appeal. Additionally, NVIDIA is actively promoting its self-developed machine solutions like DGX and HGX to further bolster the attractiveness of these models.
The profitability of high-end GPUs, particularly the A100 and H100 models, also plays a crucial role in NVIDIA’s market position. NVIDIA’s strong position in the GPU market allows for a significant price variation for the H100 model, potentially resulting in a price difference of nearly US$5,000 based on the buyer’s purchase volume.
Furthermore, the influence of ChatBOTs and AI computations is predicted to persist and expand across various professional fields, such as cloud and e-commerce services, intelligent manufacturing, financial insurance, smart healthcare, and advanced driver-assistance systems in the latter half of the year.
This trend has led to an increased demand for AI servers, particularly cloud-based servers equipped with 4-8 GPUs and edge AI servers with 2-4 GPUs. TrendForce forecasts an astonishing annual growth rate of 50 percent in the shipment of AI servers featuring NVIDIA’s flagship A100 and H100 models.
HBM, a high-speed RAM interface utilized in advanced GPUs, is poised for a significant surge in demand. NVIDIA’s upcoming H100 GPU, scheduled for release this year, will incorporate the faster HBM3 technology, enhancing the computation performance of AI server systems.
As the demand for cutting-edge GPUs, including NVIDIA’s A100 and H100, AMD’s MI200 and MI300, and Google’s proprietary TPU, continues to escalate, TrendForce projects a remarkable 58 percent increase in HBM demand for 2023, with an estimated further boost of 30 percent expected in 2024.
