Chips account for 95% of the value of an AI data server rack, according to a new report from the Semiconductor Industry Association (SIA). The finding frames AI infrastructure spending as, in effect, a semiconductor procurement exercise. Everything else in the rack — power delivery, cooling, chassis, networking hardware outside the chip stack — sits inside the remaining 5% of value.

The report, released by the SIA, concludes that AI server racks encompass the full stack of chip technologies. That scope spans logic processors, memory, storage, interconnect and packaging. In an AI rack, the accelerator silicon and the memory that feeds it dominate the bill of materials.

What does the 95% figure actually measure?

The SIA measured value at the server rack level, the standard unit of deployment in AI data centers. Racks built for AI training and inference differ sharply from general-purpose server racks. In traditional enterprise servers, chips typically represent a materially smaller share of system value, with storage, boards and integration taking a larger cut. The AI rack inverts that balance.

The report's framing gives the semiconductor industry a direct claim on the capital spending surge around AI. When cloud operators and enterprises buy AI capacity, the SIA finds, almost all of that spend flows to chip suppliers and their supply chains.

Why does the full chip stack matter?

An AI rack does not run on accelerators alone. The SIA report stresses that the full range of semiconductor technologies participates in that 95% share. A modern AI system couples:

• High-performance logic processors for training and inference
• High-bandwidth memory sitting close to the compute silicon
• Storage devices holding training corpora and model weights
• Interconnect and networking chips linking processors across the rack and across racks
• Power management and interface silicon supporting the compute load

Advanced packaging ties several of these elements together. Accelerator modules now ship as multi-die constructions that bond logic to memory stacks, which blurs the line between component categories and pushes more system value into the chip domain.

Who benefits from the finding?

The SIA represents U.S. semiconductor companies and publishes research to document the industry's economic weight. A 95% value share for chips in AI racks positions semiconductors as the primary beneficiary of AI capital expenditure rather than one supplier among many. Server assemblers, integrators and data center builders, by this measure, capture a thin slice of the value chain.

The figure also carries policy weight. Policymakers tracking AI infrastructure investment often count data center construction, power and land. The SIA's accounting redirects attention to fabs, packaging lines and chip design as the locus of AI-driven value creation.

What comes next?

The SIA will likely use the finding to argue for sustained investment in domestic chip manufacturing capacity and for trade policies that keep semiconductor supply chains open. As AI rack shipments scale, the 95% share implies that semiconductor suppliers — not system builders — will remain the principal recipients of AI infrastructure budgets, and that any constraint on chip supply will directly cap AI capacity growth.