According to the latest AI Server research report from TrendForce, North American cloud service providers (CSPs) are persistently intensifying their investments in AI infrastructure, which is projected to drive a year-on-year increase of over 28% in global AI server shipments for 2026. Furthermore, the immense computational load generated by AI inference services is ushering general-purpose servers (General Servers) into a cycle of replacement and expansion. Consequently, TrendForce also anticipates that global server shipments (including AI servers) will experience a 12.8% year-on-year growth in 2026, representing an acceleration compared to the growth rate projected for 2025. TrendForce indicates that the server market in 2024-2025 is primarily focused on training advanced large language models (LLMs), relying on AI servers equipped with GPUs and HBM for parallel processing. Starting in the second half of 2025, the continuous development of AI inference services—such as AI Agents, LLaMA model applications, and upgrades to Copilot—is prompting CSPs to actively pivot towards inference services to develop monetization and profit models. Depending on the specific application scenario, AI inference can be supported not only by AI server racks but also by general-purpose servers to handle the computational and storage requirements before and after the inference process. Statistics from TrendForce reveal that the total capital expenditure of the top five North American CSPs—Alphabet (Google), AWS (Amazon Web Services), Meta, Microsoft, and Oracle—is expected to surge by approximately 40% year-on-year in 2026. This increase is partly for deploying large-scale infrastructure and partly for replacing general-purpose servers purchased during the cloud investment boom of 2019-2021. Alphabet (Google) and Microsoft are forecasted to be the most aggressive in boosting their procurement of general-purpose servers to meet the daily, physically demanding inference traffic requirements of services like Copilot and Gemini. Observing the 2026 AI server market, the primary drivers for shipment growth will stem from North American CSPs, sovereign cloud projects by various governments, and the accelerated development of proprietary ASICs by major CSPs and edge AI inference solutions. Analyzing the AI chips used, GPUs are projected to maintain the largest share at 69.7%; models equipped with NVIDIA's GB300 are expected to become the mainstream for shipments, with the VR200 models gradually ramping up volume after the latter half of the year. However, with North American players like Alphabet (Google) and Meta actively expanding their self-developed ASIC solutions, the shipment share of ASIC AI servers is anticipated to rise to 27.8% in 2026, the highest level since 2023, with their growth rate also surpassing that of GPU AI servers. Alphabet's (Google) investment in its proprietary ASICs is significantly stronger than that of most CSPs, positioning it as the leader in the ASIC market; its TPUs are not only used internally for the Google Cloud Platform infrastructure but are also actively sold externally to companies like Anthropic.
Comments