Data Center Capex Surges in Q2 2026 as Agentic AI and Component Costs Drive Infrastructure Spend

Dell'Oro Group's latest 2Q 2026 Data Center IT Capex report reveals a marked acceleration in global data center capital expenditure. The surge was characterized by broadening investment across both top-tier hyperscale cloud service providers and rapidly expanding neocloud operators and AI model builders. Hardware spending was heavily anchored by Nvidia Blackwell Ultra deployments and proprietary hyperscaler custom silicon, alongside a noticeable rise in server average selling prices fueled by higher memory and storage component costs. In the server market, Dell maintained the leading server OEM revenue position, followed by Supermicro and Lenovo, while white-box server shipments attained record revenue levels. This growth profile reflects a pivotal technical transition for cloud architects and infrastructure teams. Unlike earlier phases of the generative AI boom that were almost exclusively centered on isolated GPU training clusters, current infrastructure demand is being reshaped by agentic AI architectures. Autonomous agents and multi-step reasoning models require sustained runtime inference, extensive retrieval-augmented storage access, and complex orchestration layers. Consequently, general-purpose compute, distributed low-latency memory fabrics, and complementary high-speed networking are commanding a much higher share of data center capital outlays than in previous quarters. Within the broader cloud and DevOps landscape, the report illustrates how specialized AI compute providers—neoclouds—are transitioning from niche GPU rental services into major infrastructure buyers that actively partner with and challenge traditional hyperscalers. Simultaneously, hyperscalers continue expanding their custom accelerator roadmaps to mitigate single-vendor exposure and curb rising operational costs. However, the compounding effect of higher server unit costs due to elevated DRAM and solid-state storage pricing means that expanding physical capacity is becoming significantly more capital-intensive on a per-rack basis. For enterprise practitioners and DevOps teams, these infrastructure dynamics demand stricter workload optimization and cost-governance practices. Organizations should anticipate continuing upward pressure on cloud instance pricing for high-memory, accelerated, and storage-optimized tiers. To counter these cost headwinds, engineering organizations should prioritize efficient inference runtimes, quantization techniques, and architectural optimizations that minimize token latency and memory footprint before committing to large-scale reserved cloud or colocation footprints.