What makes AI data centers so expensive and technically demanding? A new report from the Semiconductor Industry Association (SIA) and Deloitte points to a solution: the overall semiconductor stack. The study finds that chips account for more than 95% of the content cost inside a main AI server rack, displaying that AI infrastructure relies upon a far more than headline-grabbing GPUs.
The report,Powering AI: The Semiconductor Ecosystem at the Foundation of Data Centers, evaluates AI infrastructure via a virtual teardown of a advanced AI data server rack. Instead of targeting only on AI accelerators, the report emphasize the broad variety of chip technologies needed to help large-scale AI workloads.
As per the research, semiconductors signifies more than 50% of the total capital value required to built and function an AI data center. That figure shows how deeply chips sit inside each a part of AI infrastructure, from compute and memory to networking, power distribution, storage, and system control.
AI Infrastructure Needs Thousands Of Chips Per Rack
The report evaluate that a single AI server rack consists of more than 4,500 packaged chips. These include advanced logic chips, together with AI accelerators, ASICs, FPGAs, CPUs, DPUs, and networking chips. They also include memory technology including high-bandwidth memory, DRAM, SRAM, and NAND flash.
Analog and foundational chips also play a major role. Power chips, transceivers, controllers, and sensors support AI systems perform reliably at scale. For data scientists, engineers, and infrastructure teams, this finding strengthens an crucial point: AI performance relies on the coordinated behavior of many technologies, not one chip category alone.
The report also venture major growth ahead. SIA and Deloitte estimate that annual revenue from semiconductors deployed in AI data centers should attain more than $1.2 trillion by 2028. That would mark a nearly tenfold increase over 4 years and exceed overall global semiconductor sales from 2025 throughout all end uses by more than 50%.
AI Demand Will Drive Data Center Investment
The research evaluate that governments and industry will invest more than $4 trillion in new data center infrastructure by 2028. Of that overall, up to $2.8 trillion could go towards semiconductors.
This scale of funding suggests why AI infrastructure has become a strategic priority for policymakers, cloud providers, chipmakers, and enterprise technology leaders. As AI models emerge as larger, more multimodal, and extra embedded in production systems, demand for compute, memory bandwidth, networking capacity, and efficient power management will persist to rise.
The report also arrives as policymakers targets on U.S. Competitiveness in AI and semiconductor supply chains. SIA president and CEO John Neuffer emphasized that global AI adoption will relies on the full array of semiconductor technologies powering the field.
For technical teams, the message is obvious. AI infrastructure strategy now needs a systems-level view. Model development, deployment, inference efficiency, and data center design all relies on the underlying semiconductor ecosystem.
