TrendForce Consulting: Global AI server shipments are expected to increase by 46% annually in 2024.

According to the latest research from TrendForce Intelligence, the global AI server shipment volume in 2024 will benefit from strong demand from CSP and OEM, with an annual growth rate of 46%. Variables such as the international situation, the DeepSeek effect, and the supply chain readiness of NVIDIA Corporation's (NVDA.US) GB200/GB300 Rack will impact the AI server shipment volume in 2025, and TrendForce Intelligence has proposed three scenario estimates based on this. The first scenario is the base case, which is judged to have the highest probability of occurring. Major CSP companies such as Microsoft, Meta, Amazon, and Google have all announced plans to expand their capital expenditure on cloud or AI infrastructure this year, with average annual growth expected to exceed 30%, which will support the demand for AI servers. In this scenario, TrendForce Intelligence estimates that the global AI server shipment volume in 2025 is expected to increase by nearly 28%. However, if challenges arise in the shipment volume of GB200/GB300 Rack this year, end customers such as CSP may prioritize HGX, MGX, or other models, or focus on building their own ASIC infrastructure. The next most likely scenario is the worst-case scenario, with annual growth estimates for the overall AI server shipment volume ranging from 20% to 25%. TrendForce Intelligence states that this estimate is based on three major considerations: first, the impact of international developments on AI chips; second, the complexity of the NVIDIA Corporation GB Rack solution will result in bulk shipments in the second half of 2025, with GB300 expected to be released at the same time, leading to longer testing and verification periods for the overall Server Rack system. Lastly, the DeepSeek effect may cause customers to reduce their reliance on high-end AI solutions. The bullish case scenario is based on the positive trend in AI policies in China and the United States, such as the U.S. Stargate plan, which is expected to accelerate the deployment of AI servers by various companies. Additionally, despite the potential impact of DeepSeek on high-end GPU adoption, American CSPs are purchasing their own ASIC AI servers faster than expected, and the anticipated benefits of the DeepSeek effect will help expand AI application scenarios, thereby driving the development of edge AI servers. Overall, the optimistic estimate for AI server shipments in 2025 is a nearly 35% annual increase. TrendForce Intelligence points out that under the influence of DeepSeek, CSPs are expected to actively develop lower-cost proprietary ASIC solutions and shift their focus from AI training to AI inference, gradually increasing the proportion of AI inference servers to nearly 50%. GPUs will continue to support various AI algorithm developments in the future, but CSPs focusing on large-scale algorithms and edge enterprise customers emphasizing cost considerations will have different demands.