1. Executive summary

A recent SIA and Deloitte study argues that semiconductors account for more than 95% of the value of a leading AI data server rack. The report also projects that annual semiconductor revenue used in AI data centers could reach $1.2 trillion by 2028, highlighting the scale of the AI infrastructure buildout.

2. Key observations

· AI servers are not only about GPUs. The value chain includes HBM/DRAM, CPUs, NICs/DPUs, switch chips, optical components, power management, and foundational chips.

· As rack power density increases, memory bandwidth, networking, interconnects, and power efficiency become as important as raw accelerator performance.

· AIDC competitiveness will depend not only on the number of GPUs, but also on cluster network efficiency, failure rate, scheduling, storage throughput, and energy efficiency.

· Procurement decisions should move from headline device price to total cost of ownership and effective compute output.

3. Implications for AI data center and equipment investment

When comparing B300, H200, Pro 6000, or other platforms, the analysis should normalize for units per MW, rack power density, network cost, liquid-cooling requirement, expected rental price, target customer, and three-year residual value. This is more useful than comparing GPU prices alone.

4. Source information

Source: Semiconductor Industry Association / Deloitte, June 1, 2026

Original link: https://www.semiconductors.org/new-report-finds-semiconductors-account-for-95-of-an-ai-data-server-racks-value-encompassing-the-full-stack-of-chip-technologies/

Note: This document is an original brief prepared from public reporting and research materials, not a reproduction of the source article.