BNP Paribas 一年看多硬體空軟體

• Data Center Networking Architecture: BNP Paribas emphasizes optical transceiver supply chain as critical to AI infrastructure, enabling high-speed connectivity between servers and storage systems in hyperscaler environments • Memory IC Demand: Sustained growth in AI model training and inference driving demand for high-bandwidth memory (HBM) and DRAM, though rising costs are pressuring vendor margins • Manufacturing Services: Celestica's end-to-end design, manufacturing, and supply chain solutions position it to capture value from OEM and service provider demand for AI-optimized hardware platforms • Server and Storage Ecosystems: Lead times across server, networking, and storage components remain extended through 2026, supporting visibility and pricing power for hardware suppliers • Silicon Innovation: Cisco's Silicon One G300 represents next-generation AI cluster optimization, addressing hyperscaler requirements for custom silicon in data center deployments

BNP Paribas's hardware-over-software positioning reflects a structural shift in enterprise AI spending toward infrastructure buildout rather than application layer monetization. This strategy implies sustained capex cycles from hyperscalers through 2026-2028, benefiting optical, memory, and manufacturing vendors while software companies face extended timelines to convert large free user bases into revenue. The $120 billion data center networking market projection suggests hardware suppliers will capture disproportionate value during the AI infrastructure buildout phase, though margin compression from component cost inflation may limit profitability gains. Enterprise software vendors like Oracle and Adobe remain positioned for long-term benefits as AI spending matures, but near-term investor sentiment favors hardware exposure to infrastructure bottlenecks.