A 45% year-over-year surge will push the global server market to a projected value of $366 billion in 2025, according to new data from IDC’s Worldwide Quarterly Server Tracker. The first quarter of 2025 already marked a record-setting $95.2 billion in market value – an unprecedented 134.1% increase over Q1 2024 – underscoring the growing centrality of compute power in today’s AI-accelerating economy.

The forecasted boom is driven by exponential demand from hyperscalers and cloud service providers, particularly in response to advances in generative and agentic artificial intelligence. IDC reports that x86-based servers will make up $283.9 billion of the total, reflecting a 39.9% year-over-year rise, while non-x86 servers – often optimized for specific workloads – will grow 63.7%, reaching $82 billion.

Central to this acceleration is the sharp rise in GPU-embedded server deployments. These high-performance units, essential for AI inferencing and training tasks, are projected to account for more than half of the server market’s value in 2025. Sales of these systems are forecast to rise by 46.7% annually. The market’s transformation is further emphasized by the rapid uptake of ARM-based servers, which now represent 21.1% of all shipments, growing at a rate of 70% per year, driven largely by high-density, rack-scale configurations aimed at compute-heavy AI workloads.

Fueling this demand is not just organic enterprise IT spending but a wave of future-facing infrastructure investment tied to the pursuit of artificial general intelligence (AGI). The Stargate project, a prominent AGI-focused initiative, has reasserted its commitment to invest up to $500 billion in AI infrastructure over the coming years. This announcement came just days after the launch of DeepSeek’s R1 – an AI model designed for reasoning rather than just generative responses – highlighting a growing industry realization: next-gen AI requires exponentially more computing resources.

Kuba Stolarski, Research VP of Worldwide Infrastructure at IDC, noted the shift from basic AI chat models to reasoning and agentic models is a “step-function in compute requirements,” especially for inferencing operations. “While model efficiency has improved, the capabilities that come with advanced reasoning models like R1 demand far more robust infrastructure to function at scale,” said Stolarski.

US, China, APAC, EMEA, Latin America

Regionally, the United States is leading the surge, forecasted to grow 59.7% from 2024 to 2025 and accounting for roughly 62% of all server market revenue. China, while smaller in absolute terms, is also expanding rapidly, with expected year-over-year growth of 39.5% in 2025 and contributing over 21% of global quarterly server revenue. Japan and the APeJC (Asia/Pacific excluding Japan and China) markets are projected to see respective growth rates of 33.9% and 10.8%. In contrast, EMEA and Latin America will post modest single-digit growth, while Canada is expected to see a contraction of -9.6%, attributed to an outsized acquisition in 2024 that skewed the comparative baseline.

IDC’s Server Tracker, part of its broader Quarterly Enterprise Infrastructure Tracker, provides granular data across regions and server categories, enabling vendors, cloud providers, and investors to benchmark performance and plan product strategies with precision. The tracker also dissects server architecture trends, differentiating between x86 and non-x86 systems, as well as the presence or absence of compute accelerators.

The distinction between non-accelerated and accelerated servers is increasingly important in the AI era. IDC classifies a non-accelerated server as one without a dedicated accelerator like a GPU, FPGA, or ASIC – though basic integrated graphics on a motherboard do not count as acceleration. Accelerated servers, by contrast, come equipped with discrete graphics processing units (GPUs) or other chips like field-programmable gate arrays (FPGAs) and application-specific integrated circuits (ASICs). GPUs, particularly general-purpose (GPGPU) units, are increasingly vital for AI and high-performance computing tasks.

Hyperscalers Seeking Efficiency Gains

FPGAs, which allow post-manufacturing configuration, and ASICs, which are optimized for specific applications and cannot be modified, are both gaining traction as hyperscalers seek efficiency gains. These accelerator types allow for tailored compute at lower power envelopes, providing alternatives to general-purpose CPUs and GPUs in specialized deployments.

IDC’s definition of a server remains grounded in traditional computing architecture: a multi-user device networked to provide services, typically without a human-facing interface. These machines – equipped with motherboards, memory, local storage, OS, power supplies, network interfaces, and processors – form the computational backbone of enterprises and cloud platforms alike

As AI workloads continue to evolve from training to real-time inferencing, and as demand increases for edge and core data center processing, the global server market is undergoing a foundational shift. Efficiency and sustainability remain challenges, but the broader trajectory points toward continued growth and architectural diversification. With infrastructure investment ramping up alongside technical innovation, IDC’s data signals a maturing market ready to meet the demands of a compute-first digital economy.