Research: Global AI Server Shipments to Reach 3.7 Million Units This Year, Estimated to Grow Over 50%
Research firm TrendForce predicts global AI server shipments will reach approximately 3.7 million units in 2026, a year-on-year increase of over 50%, driven by the rapid development of generative AI and the increasing demand for computing power and storage.
📋 Article Processing Timeline
- 📰 Published: May 12, 2026 at 13:54
- 🔍 Collected: May 12, 2026 at 14:02 (7 min after Published)
- 🤖 AI Analyzed: May 14, 2026 at 06:51 (40h 49m after Collected)
Central News Agency
(Central News Agency reporter Pan Zhiyi, Taipei, May 12) The latest forecast data from research firm TrendForce shows that global AI server shipments will reach approximately 3.7 million units in 2026, an increase of 51.3% compared to last year. AI servers are expected to maintain double-digit growth from 2027 to 2028, with global AI server shipments estimated to approach 5 million units by 2028.
TrendForce stated that in early May, US AI company Anthropic publicly announced that its Q1 2026 revenue and product usage annualized growth reached 80 times, significantly exceeding the company's previous expected growth target of 10 times. It also mentioned the current severe shortage of computing power, highlighting the huge current global demand for AI computing power and massive storage needs.
TrendForce explained that the growth in AI server shipment scale leads to an increase in single-machine storage capacity. Compared to traditional servers where a single device relies only on multiple CPUs to complete calculations, with low computing density and small memory capacity, the growth of generative AI large model parameters, longer context lengths, and the surge in multimodal data have made heterogeneous computing the core technological path for AI servers.
TrendForce pointed out that taking NVIDIA's Blackwell and Vera Rubin as examples, their architectures will deeply strengthen heterogeneous collaborative capabilities. Through the NVLink high-speed interconnect bus, they will connect the data links between central processing units (CPU), graphics processing units (GPU), high-bandwidth memory (HBM), and double data rate synchronous dynamic random access memory (DDR) memory, achieving integrated coordination of computing, storage, and scheduling.
In addition, TrendForce stated that Google's Gemini computing cluster also adopts a heterogeneous architecture of CPU plus Tensor Processing Unit (TPU). Under heterogeneous architectures, multi-chip high-frequency interaction and real-time flow of Token data require both large-capacity DDR as system memory and high-bandwidth HBM to complete instantaneous computing throughput. (Editor: Yang Kaixiang) 1150512
Choose to stand with facts, every sponsorship you provide is a force for protecting press freedom.
Download the Central News Agency "First-Hand News" APP to stay updated with the latest news.
The text, images, and audio/video on this website may not be reproduced, publicly broadcast, or publicly transmitted and used without authorization.
(Central News Agency reporter Pan Zhiyi, Taipei, May 12) The latest forecast data from research firm TrendForce shows that global AI server shipments will reach approximately 3.7 million units in 2026, an increase of 51.3% compared to last year. AI servers are expected to maintain double-digit growth from 2027 to 2028, with global AI server shipments estimated to approach 5 million units by 2028.
TrendForce stated that in early May, US AI company Anthropic publicly announced that its Q1 2026 revenue and product usage annualized growth reached 80 times, significantly exceeding the company's previous expected growth target of 10 times. It also mentioned the current severe shortage of computing power, highlighting the huge current global demand for AI computing power and massive storage needs.
TrendForce explained that the growth in AI server shipment scale leads to an increase in single-machine storage capacity. Compared to traditional servers where a single device relies only on multiple CPUs to complete calculations, with low computing density and small memory capacity, the growth of generative AI large model parameters, longer context lengths, and the surge in multimodal data have made heterogeneous computing the core technological path for AI servers.
TrendForce pointed out that taking NVIDIA's Blackwell and Vera Rubin as examples, their architectures will deeply strengthen heterogeneous collaborative capabilities. Through the NVLink high-speed interconnect bus, they will connect the data links between central processing units (CPU), graphics processing units (GPU), high-bandwidth memory (HBM), and double data rate synchronous dynamic random access memory (DDR) memory, achieving integrated coordination of computing, storage, and scheduling.
In addition, TrendForce stated that Google's Gemini computing cluster also adopts a heterogeneous architecture of CPU plus Tensor Processing Unit (TPU). Under heterogeneous architectures, multi-chip high-frequency interaction and real-time flow of Token data require both large-capacity DDR as system memory and high-bandwidth HBM to complete instantaneous computing throughput. (Editor: Yang Kaixiang) 1150512
Choose to stand with facts, every sponsorship you provide is a force for protecting press freedom.
Download the Central News Agency "First-Hand News" APP to stay updated with the latest news.
The text, images, and audio/video on this website may not be reproduced, publicly broadcast, or publicly transmitted and used without authorization.