普通外文研报
Daily Views on Tech Headlines | AVGO, OpenAI, CBRS, QCOM, China, IBM
研报英文原文证据摘录
Daily Views on Tech Headlines | AVGO, OpenAI, CBRS, QCOM, China, IBM
Quick Note
June 25, 2026 Hardware
Daily Views on Tech Headlines | AVGO, OpenAI, CBRS, QCOM,
Analysts China, IBM
Matt Bryson
(617) 695-6302
matthew.bryson@wedbush.com
Antoine Legault, CFA, CPA AVGO, NVDA, CBRS: OpenAI and Broadcom Unveil Jalapeño custom AI ASIC
(626) 497-1619
antoine.legault@wedbush.com View: OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom chip, a
purpose-built LLM inference accelerator manufactured by Broadcom on TSMC and
integrated by Celestica, taped out in roughly nine months with help from OpenAI's
own models. The companies claim substantially better performance per watt than
current state-of-the-art (Bloomberg cites ~50% lower inference cost; Hock Tan calls
it comparable to NVIDIA Blackwell and Google TPUs), with initial gigawatt-scale
deployment targeted for late 2026 under the October 10GW agreement. OpenAI
says it will keep buying NVIDIA and AMD for other workloads.
History suggests that successful ASIC development for compute is a multi-
generation process spanning many years. And Tom's Hardware does a good job
noting some of the questions around Jalapeño in light of a dearth of detail as of yet
(as well as some possible technical takeaways around the chip's design from the
photos and details that AVGO and OpenAI have shared).
Net, we see this development as a probable positive for AVGO (and OpenAI
assuming the silicon eventually works as presented), but also wouldn't be surprised
if initial shipments are more modest in scale and broader adoption requires a 2nd
or 3rd (or 4th) iteration of the design. OpenAI Tom's Hardware, CNBC
QCOM, AVGO, China: Qualcomm ups forward data center sales outlook, with
China's market creating potential for further upside.
本摘录由系统从所标注的 PDF 证据页直接提取并保留英文原文,不做批量翻译;登录后在阅读器切换中文时才按需翻译。
打开研报阅读器