What happened
BNP Paribas named Arista Networks, Seagate Technology, and Astera Labs as its top picks for exposure to AI infrastructure demand through 2027. The bank raised AMD's price target to $600 from $460 and Astera Labs' target to $475 from $260. Analyst Karl Ackerman favors companies exposed to agentic AI CPUs, custom processors, 200G networking chips, and 1.6T optical transceivers.
Why it matters
Data-center networking and storage are critical bottlenecks as AI deployment scales. BNP expects data-center switch sales to grow at a 44% compound annual rate, reaching $114.7 billion（約18兆円） by 2028 from $26.9 billion（約4.3兆円） in 2024—a sign that infrastructure vendors are positioned to capture substantial value from the AI buildout. These picks offer investors concentrated exposure to sectors the bank sees as cornerstones of AI scaling.
What to watch
The most aggressive forecast is for backend switching, which BNP expects to expand nearly 8-fold to $78.6 billion（約13兆円）. Optical transceiver sales may reach $53.1 billion（約8.5兆円）. Ackerman notes the next major demand catalyst may not arrive for another quarter, so near-term catalysts may be limited.
Summaries like this, in your inbox every morning.
BNP Paribas' upgrade of its AI infrastructure outlook reflects confidence that the buildout of large-scale AI systems will continue to drive demand for specialized hardware well into 2027. The analyst's emphasis on agentic AI CPUs (processors designed for autonomous systems), custom silicon, and high-speed networking and optical components suggests the bank sees the competitive advantage shifting toward companies that can supply the most demanding, specialized parts of the data center rather than commodity infrastructure.
The forecasted growth rates are striking: a 44% compound annual rate for data-center switches and an 8-fold expansion for backend switching imply that networking infrastructure is becoming a larger fraction of total data-center spending. Optical transceivers—the devices that move data between racks and across continents at ultra-high speeds—are projected to reach $53.1 billion（約8.5兆円）, a scale comparable to entire traditional semiconductor markets. These projections suggest BNP believes AI data centers will require not just more compute but proportionally much more networking and connectivity capacity to move data between chips and systems.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
