Dell'Oro lifts data centre chip forecast on AI demand

Dell'Oro Group has raised its forecast for the global data centre IT semiconductors and components market to USD $1.8 trillion by 2030, citing stronger expectations for data centre capital expenditure and power capacity additions.

Artificial intelligence is expected to remain the main driver of infrastructure spending. AI accelerators are forecast to account for the largest share of component revenue, while demand also spreads to memory, storage, central processing units and network interface cards.

The updated outlook points to broader knock-on effects from AI workloads beyond specialist chips. Inference, storage and agentic AI are expected to lift demand for general-purpose servers, extending growth into more conventional parts of the server market.

Power demand

One of the clearest constraints is electricity use. Server and storage components are projected to consume more than 200GW of power over the next five years, indicating that data centre operators will need to expand site power and supporting infrastructure alongside hardware deployments.

The estimate highlights how semiconductor growth is now tied not only to chip design and manufacturing supply, but also to the physical limits of data centre construction. As operators add more compute-intensive systems, component demand will increasingly depend on whether enough power and cooling can be brought online.

Memory and storage also remain under pressure, with supply constraints expected to keep pricing elevated, although prices should moderate over the forecast period.

That pricing matters because memory and storage are central to AI server configurations as well as broader enterprise and cloud deployments. If prices ease gradually rather than sharply, suppliers could continue to benefit from strong spending while buyers face a slower return to more normal procurement patterns.

Baron Fung, Vice President of Research at Dell'Oro Group, said AI would remain the primary driver of data centre infrastructure spending, with accelerators accounting for the largest share of the component market.

Fung said memory, storage, CPUs and NICs would innovate and scale alongside accelerators, while supply constraints would sustain elevated pricing for memory and storage. He added that inference, storage and agentic AI would increase demand for general-purpose servers, supporting growth across broader component categories.

He also said significant development was under way to improve accelerator performance per watt and reduce cost per token by scaling architectures from the package through the rack level. Heterogeneous computing tailored to different workload domains could further improve efficiency at the data centre scale.

Custom silicon

The forecast also points to a growing shift towards custom silicon. AI accelerators, CPUs and NICs are expected to move further in that direction as hyperscale operators seek better efficiency and lower costs.

The trend reflects a wider change in data centre procurement. Large cloud groups have spent years developing in-house chips for networking, inference and general compute, and the latest forecast suggests that effort will reach deeper into the component stack as AI systems become larger and more specialised.

Network interface cards are another category expected to grow faster. Front-end demand is forecast to rise from general-purpose servers used for inference, agentic AI and storage, while scale-out AI networking is expected to strengthen back-end demand.

The distinction matters because AI infrastructure requires investment across several layers at once. Operators need components inside servers, interconnects between servers and networks that can handle traffic generated by large training and inference clusters. That broadens spending beyond a narrow group of accelerator suppliers.

Broader market

The report covers revenue and unit shipments for accelerators, CPUs, memory, NICs and storage drives, including hard disk drives and solid-state drives. Together, those categories show how AI spending is reshaping the wider data centre supply chain, not just the market for graphics processing units and dedicated accelerators.

The upward revision from Dell'Oro's earlier forecast suggests the firm now expects stronger and more sustained spending than it did at the start of the year. For suppliers, that implies continued demand across a range of parts. For operators, it signals that the race to build AI infrastructure will remain tied to energy availability, component pricing and the ability to design systems that use power more efficiently.