How AMD Makes Money

Advanced Micro Devices runs a fabless semiconductor model. It designs CPUs, GPUs, AI accelerators, FPGAs, and adaptive SoCs, and contracts the wafer manufacturing to outside foundries, primarily TSMC. That structure keeps leading-edge fabrication off AMD's balance sheet and makes engineering its largest investment: research and development cost $8.1B in FY2025, about 23% of revenue. Revenue comes from three reportable segments. Data Center, $16.6B in FY2025, covers EPYC server CPUs, Instinct AI accelerators, Pensando DPUs, AI network interface cards, and adaptive SoCs sold to cloud providers, enterprises, and HPC customers. Client and Gaming, $14.6B, covers Ryzen desktop and notebook processors, chipsets, Radeon discrete graphics, and the semi-custom SoCs inside PlayStation and Xbox consoles; within it Client was $10.6B and Gaming $3.9B. Embedded, $3.5B, covers the Xilinx-derived FPGA, system-on-module, and adaptive SoC lines sold into telecom, aerospace, defense, automotive, and industrial equipment. The design lever is chiplets: AMD splits a processor into several smaller dies joined by Infinity Fabric, which raises usable yield and lets it mix process nodes inside one package.

AMD's growth plan centers on data center AI, funded by $8.1B of research and development in FY2025. It has bought capability rather than waiting to build it: Xilinx in February 2022 for FPGAs and adaptive computing, a deal announced in October 2020 at about $35B in stock and worth roughly $49B at close; Pensando in May 2022 for DPUs at about $1.9B; Silo AI in August 2024 for AI software engineering at about $665M; and ZT Systems in March 2025 for rack-scale systems design. The aim is to sell complete AI racks that combine EPYC CPUs, Instinct accelerators, AI networking, and ROCm software instead of individual chips, while Ryzen and Radeon keep funding the effort.