AI infrastructure spending is lifting semiconductor sector margins after two years of inventory corrections. Analog Devices reported stronger demand from data center and industrial customers as AI chip volumes climb, while memory and packaging suppliers see pricing stabilize.
Amkor Technology, the largest U.S.-headquartered outsourced semiconductor assembly and test (OSAT) provider, is gaining from advanced packaging orders for AI accelerators and high-bandwidth memory stacks. DRAM pricing—pressured through 2024 by pandemic-era capacity additions—is firming as hyperscalers order server memory for large language model training clusters.
Cisco launched the Silicon One G300 network processor for AI data center fabrics, reflecting infrastructure needs beyond GPUs. "AI at scale demands open, standards-based networking that customers can deploy with confidence across diverse environments," said Yousuf Khan, Cisco executive.
ams OSRAM reported Q4 results above consensus but guided FY26 revenue and adjusted EBITDA modestly lower year-over-year due to divestments, dollar weakness, and one-off costs including higher precious-metal prices and stranded expenses. The company highlighted its Digital Light technology, which has secured over €500 million in design wins with a clear growth trajectory in automotive and projection applications.
The sector faces cautious capacity expansion. Foundries and memory makers are adding wafer starts selectively, focused on leading-edge nodes and high-bandwidth memory rather than broad buildouts. Photonics integration and FPGA-based AI inference chips are emerging niches as customers diversify beyond GPU-only architectures.
Trading desks see upside in companies with exposure to advanced packaging, DRAM spot pricing, and AI accelerator supply chains. Margin expansion from better mix and higher utilization rates is offsetting unit pricing pressure in commodity logic. The cycle is in early recovery, with hyperscaler capex commitments extending through 2026 providing visibility for component suppliers.
Risks include smartphone and PC demand remaining soft, potential overcorrection in memory capacity additions if AI server orders slow, and geopolitical supply chain fragmentation raising costs. Near-term trade setups favor OSAT providers and specialty memory over commodity logic foundries.