Global annual AI server shipments, 2024-2025

Despite unfavorable factors such as the ongoing global tariff wars, persistent doubts about the slow return on generative AI investments, and the US's continued expansion of technical and product restrictions on China's AI computing capabilities, major IT players remain undeterred, continuing to invest heavily in AI datacenter development.

Meanwhile, the performance and applications of large language models (LLMs) are steadily advancing, driving sustained investment and demand for AI servers. As a result, global AI server shipments are projected to reach 1.81 million units in 2025, with shipments of high-end models equipped with high bandwidth memory (HBM) surpassing one million units, a 40% increase from the previous year, according to DIGITIMES estimates.

Regarding the procurement of high-end AI servers, due to rapid advancements in commercial GPU cluster performance and the accelerated adoption of AI ASICs by some cloud users, North America's major cloud service providers are expected to account for 70% of procurement in 2025.