According to TrendForce's latest AI server research report, North American cloud service providers (CSPs) continue to increase their investment in AI infrastructure, which is expected to drive global AI server shipments to grow by more than 28% year-on-year in 2026. Furthermore, the massive computing load generated by AI inference services is also driving a replacement and expansion cycle for general-purpose servers. Therefore, TrendForce forecasts that global server shipments (including AI servers) will also grow by 12.8% year-on-year in 2026, a larger growth rate than in 2025.

TrendForce indicates that the server market in 2024-2025 will primarily focus on advanced large-scale language model (LLM) training, relying on AI servers equipped with GPUs and HBMs to perform parallel computations. From the second half of 2025, AI inference services such as AI agents, LLaMA model applications, and Copilot upgrades will continue to develop, with CSPs actively shifting towards inference services to develop monetization and profit models. Depending on the application scenario, AI inference can utilize both AI server racks and general-purpose servers to support the computational and storage needs before and after inference.

According to TrendForce statistics, the total capital expenditure of the five major North American CSPs—Google, AWS, Meta, Microsoft, and Oracle—is projected to increase by 40% year-on-year in 2026. This is partly due to the deployment of large-scale infrastructure and the replacement of general-purpose servers purchased during the cloud investment boom of 2019-2021. Google and Microsoft are expected to be the most aggressive in increasing their general-purpose server purchases to meet the daily demand for physical Copilot and Gemini inference traffic.

The proportion of ASIC category is increasing.

Looking at the AI server market in 2026, shipment momentum will mainly come from North American CSPs, local government sovereign cloud projects, and the accelerated development of in-house ASICs and edge AI inference solutions by large CSPs. Analyzing the AI chips used, GPUs are estimated to account for 69.7%, remaining the largest segment; models equipped with NVIDIA GB300 will become the mainstream, while VR200 will gradually increase in volume after the second half of the year.

However, with North American players like Google and Meta actively expanding their self-developed ASIC solutions, the shipment share of ASIC AI servers is expected to rise to 27.8% in 2026, the highest since 2023, and its shipment growth rate will surpass that of GPU AI servers. Google's investment in self-developed ASICs is significantly stronger than that of most CSPs, making it the leader in the ASIC market. In addition to using its own TPUs in the Google Cloud Platform cloud service infrastructure, it is also actively selling them to companies such as Anthropic.