Gartner：企業數據中心永不消亡

• AI-optimized server architectures designed for high-density GPU/accelerator workloads, with annual worldwide server spending projected to climb 37% in 2026[3] • Power and cooling infrastructure for AI datacenters: AI server power supply market projected to grow from $1.5 billion (2024) to over $31 billion (2028); liquid cooling market expected to grow from $300 million (2024) to over $30 billion (2028)[5] • High-speed networking infrastructure: Gen AI networking equipment projected to grow from $8 billion (2023) to $34 billion (2028), with accelerated shift toward fiber optics and advanced interconnect solutions (Ethernet, InfiniBand, copper-based and fiber-optic connections)[5] • Datacenter hardware composition: Tech hardware and equipment (power, cooling, storage, networks) expected to account for one-third of $582 billion in 2026 datacenter spending, with chips comprising the remaining two-thirds[5] • AI infrastructure spans from rack-level systems in gigawatt-scale hyperscale datacenters to smaller deployments in medium-sized enterprise wiring closets[5]

The persistence and acceleration of datacenter investment signals a fundamental shift in enterprise IT strategy away from the 2019 cloud-first consolidation narrative. Rather than wholesale migration to public cloud, organizations are adopting hybrid models with substantial on-premises infrastructure investments. This trend is primarily driven by AI workload requirements, which demand specialized hardware, custom cooling solutions, and high-speed interconnects that are economically justified only at scale. The 31.7% datacenter growth rate—more than double any other IT category—indicates that datacenters will remain central to enterprise computing through at least 2027-2028. This contradicts earlier predictions of datacenter obsolescence and suggests that cloud providers and enterprises will coexist in a bifurcated infrastructure landscape, with datacenters serving as critical nodes for AI model training, inference, and specialized workloads while cloud services handle general-purpose computing and elasticity requirements.