Advanced Micro Devices Inc. (AMD): Key takeaways from Advancing AI 2026

研报英文原文证据摘录

Advanced Micro Devices Inc. (AMD): Key takeaways from Advancing AI 2026

Goldman Sachs Advanced Micro Devices Inc. (AMD)

change in computing demand, requiring both accelerators and CPUs to orchestrate

increasingly complex workflows. As a result, the AI infrastructure market is

expanding rapidly, with the datacenter AI accelerator TAM projected to grow from

$200 bn to $1.4 tn in 2030 (40% CAGR), datacenter CPU TAM from $26 bn to $220

bn (50% CAGR), and AMD’s total compute TAM from $365 bn to $2 tn by 2030.

AMD also expects to capture 50% of the datacenter CPU market by 2030.

n Helios launch: AMD unveiled its next-generation Helios AI rackscale platform, built

on CDNA 5 architecture with GPUs featuring up to 40 PFLOPS (FP4), 20 PFLOPS

(FP8), 432 GB of HBM4 memory, and 23.3 TB/s memory bandwidth. The system

combines 75 GPUs connected via UALink over Ethernet, a 96-core EPYC CPU, Salina

DPU, Volcano 800G AI NIC, and six networking trays to deliver higher-scale AI

performance. AMD stated that Helios is already in full production, with shipments

beginning at the end of 3Q, and volume ramp in 4Q.

n Partnership with Cerebras: AMD and Cerebras have partnered to combine AMD’s

Helios systems with Cerebras’ Wafer-Scale Engine, creating a high-performance AI

inference solution that delivers ultra-low latency, higher efficiency, and up to 5x

better tokens-per-watt. The solution is expected to launch via Cerebras Cloud in

2H26.

n ROCm.ai: AMD launched ROCm.ai, which integrates with tools such as Cursor,

Claude, Codex and Gemini to provide developers with an AI-driven software

development experience. AMD also introduced Hyperloom, an optimization layer

that automatically tunes system configurations and optimizes end-to-end AI

workloads.