The Semiconductor Industry Association and Deloitte have put a hard number on the AI infrastructure boom: in a leading AI server rack, more than 95% of the value comes from chips. Their June 1 report virtually disassembled an advanced rack and counted more than 4,500 packaged chips inside.
The report estimates that chips account for more than half of total data-center capital spending. By 2028, annual revenue from AI data-center chips could reach $1.2 trillion, nearly ten times the level across four years, while global data-center infrastructure spending could total about $4 trillion, with as much as $2.8 trillion going to semiconductors.
The argument is aimed at a common shorthand: AI chips are not just GPUs. Logic, memory, networking, power management and advanced packaging all shape whether a data center can deliver enough compute. SIA CEO John Neuffer said the future of AI will be determined by the full array of semiconductor technologies that power it.
That reframes the AI build-out as a full-stack semiconductor contest. GPUs remain the most visible part of the story, but HBM, interconnect, packaging and power components decide how much of the AI rack actually works at scale.
Sources: Semiconductor Industry Association, InfotechLead, CocoLoop.