Memory Interface Chip Primer: Riding on the agentic AI boom

Research evidence excerpt

Memory Interface Chip Primer: Riding on the agentic AI boom

9 July 2026

China & Japan Semiconductors

We initiated coverage on Montage in April (initiation report and deck) where we discuss Qingyuan Lin, Ph.D.

+852 2123 2654 the opportunity for memory interface chip vendors driven by agentic AI. In this primer, we

qingyuan.lin@bernsteinsg.com provide a comprehensive overview of the industry's growth drivers, competitive dynamics,

and investment implications to help investors navigate the sector. Updated models can be

David Dai, CFA downloaded here: Memory Interface Chip Industry Model and Montage Model. Reiterate

+852 2918 5704

david.dai@bernsteinsg.com Outperform on Montage, PT revised up from HKD 320 to 520 in H-share and CNY 220 to

400 in A-share. Reiterate Outperform on Renesas.

Francis Ma

+852 2123 2626 Memory interface chips are a direct and amplified beneficiary of the server CPU

francis.ma@bernsteinsg.com Renaissance driven by Agentic AI. As the industry shifts from GPU-centric training

toward inference-heavy agentic AI deployments—where the CPU reclaims its role as

Kai Zhang orchestrator of multistep, sequential tasks—server CPU volumes are poised to grow +852 2123 2665

kai.zhang@bernsteinsg.com structurally faster than expected (our CPU report), and the role of memory interface chip

becomes increasingly important when the DRAM capacity and bandwidth per CPU grows.

Jack Lin We project the global memory interface chips TAM to reach USD 20Bn by 2030 (3x of our

+852 2123 2683 previous projection) at 65% CAGR (vs 32%), driven by three reinforcing tailwinds: rising jack.lin@bernsteinsg.com

server CPU volume, increasing DRAM package count per CPU, and higher interface chip

content per package through the MRDIMM upgrade. Juho Hwang

+81 3 6777 6980