Powering AI: The Semiconductor Ecosystem at the Foundation of Data Centers (Semiconductor Industry Association)
SaveSummary
- A single AI server rack contains over 4,500 packaged chips, comprised of approximately 20,000 individual dies.p.3
- Semiconductors account for more than 95% of a leading AI server rack's content value and more than 50% of the total capital expenditures for building and operating an AI data center.p.3
- Government and industry will invest over $4 trillion in new data center infrastructure through 2028, of which up to $2.8 trillion will be spent on semiconductors.p.3
- A leading data center can accommodate over 45,000,000 chips, assuming 10,000 AI compute racks.p.5
- AI accelerators have a unit price of about $10,000 - $40,000, and CPUs about $7,500 - $15,000.p.11
- AI Accelerators can handle 10-100x more calculations per second than CPUs.p.16
Key insights
- A typical AI compute board combines 1–2 AI Accelerators with 1–2 CPUs.p.12
- High-bandwidth memory stacks up to 16 dies each, with 6–12 stacks (96–192 dies) per integrated AI GPU.p.14
- A typical data center NVMe SSD includes 256 NAND flash dies.p.17
- The power tray is estimated at $50,000 - $290,000 per server rack.p.23
- The network and IPMI tray is estimated at $17,000 - $25,000 per server rack.p.25
- Active copper cable chips enable short-distance communication of up to 5 meters across trays or racks.p.22
Topics and entities
Summary and figures generated from the document and checked automatically: every number shown appears on the cited page. The full document is available from the publisher. Publishers can request removal.