TrendForce’s latest research indicates that global AI server shipments experienced significant growth of 46 percent in 2024, fueled by increasing investments from cloud service providers (CSPs) and original equipment manufacturers (OEMs).
However, uncertainties could influence AI server growth in 2025, including US chip restrictions, the impact of the DeepSeek AI model, and supply chain readiness for next-generation NVIDIA GB200/GB300 rack deployments.
These factors introduce varying levels of risk, leading to three potential growth scenarios: a base case scenario with steady expansion, a worst-case scenario with more limited growth, and a bullish scenario with accelerated deployment of AI infrastructure.
The base case scenario, which is considered the most probable, anticipates that AI server shipments will continue growing at a healthy rate, driven by increased capital expenditures from major tech giants such as Microsoft, Meta, Amazon, and Google. These companies have recently announced plans to expand their cloud and AI infrastructure, with annual spending growth projected to exceed 30 percent.
Under this scenario, the global AI server market is expected to grow by approximately 28 percent year-over-year in 2025. The supply chain outlook for AI servers currently appears stable, with shipments proceeding as planned.
However, potential delays or challenges in scaling GB200/GB300 racks could lead CSPs to adjust their AI infrastructure strategies, opting instead for HGX- or MGX-based platforms, or even proprietary ASIC solutions that reduce dependency on NVIDIA’s ecosystem.
In the worst-case scenario, which has the second-highest likelihood, AI server growth could slow to a range of 20-25 percent due to a combination of regulatory and supply chain constraints. Stricter US policies on AI chip exports to China could heighten uncertainty in the market, potentially impacting demand and procurement strategies for Chinese tech firms.
Another challenge comes from the complexity of NVIDIA GB rack deployments, which may lead to delays in large-scale adoption, particularly if the GB300’s advanced specifications extend testing and validation timelines for AI server racks.
Additionally, the DeepSeek effect could influence AI server demand by shifting customer priorities away from high-end AI hardware, particularly as organizations explore alternative, cost-effective AI solutions.
In the most optimistic (bullish) scenario, AI server shipments could surge by nearly 35 percent in 2025, assuming more aggressive AI policy initiatives from both the US and China. If US-backed programs such as the Stargate Initiative successfully accelerate AI infrastructure rollouts, CSPs and enterprises may scale their AI server deployments faster than initially expected.
While DeepSeek’s growing capabilities might reduce reliance on high-end GPUs for certain AI tasks, US-based CSPs have been ramping up their procurement of proprietary ASIC-based AI servers, potentially mitigating any decline in high-end GPU shipments. DeepSeek is also expected to drive broader AI adoption, particularly in edge computing environments, which could further boost demand for AI inference servers.
A key trend emerging from these scenarios is the gradual shift in AI server demand from training-focused infrastructure to inference-based solutions. As CSPs increasingly prioritize cost-efficient AI inference, proprietary ASIC AI servers are becoming more attractive alternatives to high-end GPU-based systems.
TrendForce projects that AI inference servers will account for nearly 50 percent of total AI server shipments, marking a significant change in market dynamics. While GPUs will remain indispensable for developing AI models, the industry is becoming more segmented, with high-performance solutions catering to large-scale AI applications while cost-conscious enterprises at the edge focus on economical alternatives.
