US Semiconductors: Revisiting AMD and ARM Estimates Amid Agentic AI Acceleration

研报英文原文证据摘录

US Semiconductors: Revisiting AMD and ARM Estimates Amid Agentic AI Acceleration

Global Research

24 June 2026ab

US Semiconductors Equities

AmericasRevisiting AMD and ARM Estimates Amid

Agentic AI Acceleration Semiconductors

Timothy Arcuri

Analyst

timothy.arcuri@ubs.com

Summary +1-415-352 5676

Since publishing our CPU TAM model in May, we sense investors have extrapolated Natalia Winkler, CFA

strong near-term demand into more aggressive C2030 TAM outcomes (in some cases Analyst

we have heard ~$300B). We have pushed back against these narratives, instead steering natalia.winkler@ubs.com

such bullish outcomes to a long thesis on MU because we believe DRAM will remain too +1-415-352 4626

constrained to enable a TAM that large by 2030. All of that said, we do see accelerating Alex Kivali

traction in standalone CPU racks with the architectural split we outlined playing out: 1) Analyst

latency-sensitive head nodes favoring ARM (NVDA Grace/Vera (which we think is alex.kivali@ubs.com

gaining traction due to combination of low latency+multithreading with NVDA offering +1-212-713 3945

memory as part of the system), Axion, Graviton), and 2) standalone agentic AI racks Gianmarco Vella

being more core and throughput-driven — an area where we think AMD has a distinct Associate Analyst

advantage. We think INTC retains optionality with x86, but mid-term upside hinges on gianmarco.vella@ubs.com

improving its high core-count/performance roadmap. While ARM’s Gen1 CPU is better +1-415-352 4555

suited for head nodes today, its roadmap (200–500 cores) could become more Aaryan Wadhwa

competitive in standalone use cases over time. Net, we raise estimates for AMD and Associate Analyst

ARM as key beneficiaries of standalone CPU and head node deployments, and move PTs aaryan.wadhwa@ubs.com