Table of Contents
Dell’Oro Group VP of research Baron Fung told RCRTech that accelerated servers equipped with AI accelerators are expected to account for roughly 40% of the $3 trillion in DC capital expenditure forecast
In sum – what to know:
AI drives spending – Accelerated servers with AI accelerators are expected to account for roughly 40% of the projected $3 trillion in data center capex.
Inference reshapes demand – Inference is already driving accelerator demand, while general-purpose servers and storage forecasts have more than doubled.
Power is the bottleneck – Fung identifies power availability as the biggest constraint, with grid expansion and onsite generation becoming increasingly important.
Power availability is emerging as the biggest constraint on the data center industry’s ability to sustain the rapid expansion of AI infrastructure, according to Dell’Oro Group vice president of research Baron Fung.
Fung said accelerated servers equipped with AI accelerators are expected to account for roughly 40% of the $3 trillion in data center capital expenditure forecast by Dell’Oro through 2030, highlighting the central role of AI infrastructure in the market’s growth.
“Accelerated servers with AI accelerators are expected to account for roughly 40% of the $3 trillion data center capex, so definitely AI infrastructure is a major driver. In addition to accelerated servers, specialized networking for AI clusters and storage are used to support AI workloads, whether directly or indirectly”, the executive told RCRTech.
Dell’Oro’s latest forecast represents a significant increase from its January outlook. The company said its 2030 data center capex forecast has nearly doubled, reflecting higher hyperscaler capex guidance, increased projections for global data center power capacity and higher commodity costs.
The role of AI is also extending beyond training workloads. Fung said high-end accelerators are already supporting inference demand and that inference-driven requirements are creating additional demand for general-purpose servers and storage.
“High-end accelerators are already supporting the demand for inference workloads. In fact, it’s likely that the accelerators demanded by inference have exceeded that of training clusters. The increase in inference demand creates additional demand for general-purpose servers and storage, such as more KV cache storage and CPUs to manage the agentic AI workloads. Thus, we also increased our forecast for general-purpose servers and storage by more than 2X,” he added.
Dell’Oro expects general-purpose server demand to benefit from growing inference, agentic AI and storage workloads, broadening infrastructure growth beyond accelerated computing.
The analyst sees power as the most significant constraint on the industry’s ability to maintain its current investment trajectory.
“I think power availability is probably the biggest bottleneck. The industry is racing to add more power capacity. In addition to more grid power, hyperscalers are relying more on onsite power generation. Demand for power generation equipment is very strong with long lead times. Some chip vendors have quoted higher capex figures than my $3 trillion figure suggests, but haven’t considered the power availability aspect,” he added.
Fung also identified near-term constraints in memory and storage components, although he expects those bottlenecks to improve over the next one to two years.
Beyond infrastructure availability, the long-term sustainability of AI investment will depend on whether the technology generates sufficient returns. “The ROI on AI is key for sustainable long-term growth,” he said.
Dell’Oro’s forecast also points to a broader shift in the customer mix. The newly added AI-specialized cloud segment, comprising AI model builders and neocloud service providers, is projected to grow at nearly a 60% CAGR through 2030, outpacing other customer segments.
At the same time, the company expects high-end accelerators to remain the largest share of data center capex and the primary driver of spending growth through 2030, while general-purpose servers benefit from the expansion of inference, agentic AI and storage workloads.
