US Semiconductors and Semi Equipment "SemiBytes: Ultra-Fast Inference, Q..."

研报英文原文证据摘录

US Semiconductors and Semi Equipment "SemiBytes: Ultra-Fast Inference, Q..."

Global Research

22 June 2026ab

US Semiconductors and Semi Equipment Equities

AmericasSemiBytes: Ultra-Fast Inference, QCOM ADay,

WFE Upside, AMD-MEXT Semiconductors

Timothy Arcuri

Analyst

timothy.arcuri@ubs.com

UBS Chip Chats - A Discussion on Ultra-Fast Inference with a Hardware Expert +1-415-352 5676

We will be hosting an expert call on fast inference and disaggregated architectures on Natalia Winkler, CFA

Tuesday, June 23 from 11:00am to 12:00pm ET. Please register here. Analyst

natalia.winkler@ubs.com

QCOM Analyst Day Preview - $20+ EPS by FY31 +1-415-352 4626

We expect QCOM Analyst Day this Wednesday to lean more heavily into a long-term Alex Kivali

data center narrative—which was largely absent back in 2024 and not part of QCOM's Analyst

standing financial model. Qualcomm has already begun to showcase this buildout, alex.kivali@ubs.com

including a rack-scale AI inference system around AI200 with a 43TB GDDR memory +1-212-713 3945

rack (a 350B-parameter model fitting on a single card) optimized for decode, with initial Gianmarco Vella

deployments this year. Beyond announced accelerators for decode applications, we Associate Analyst

expect incremental updates on its server CPU efforts and potential inorganic options. On gianmarco.vella@ubs.com

the latter, the Tenstorrent angle (per press reports) makes sense to us as a stack +1-415-352 4555

extension, moving QCOM from a largely decode-optimized system provider toward a Aaryan Wadhwa

more balanced inference architecture (CPU + accelerator + system), adding exposure to Associate Analyst

higher-value prefill / latency-sensitive workloads with Tenstorrent hardware (also based aaryan.wadhwa@ubs.com

on GDDR memory, as is QCOM's AI line).