Global semiconductor sales are rebounding as AI infrastructure demand drives growth in memory chips, advanced packaging, and specialized components across the U.S., Europe, and Asia.
Amkor Technology, the largest U.S.-headquartered outsourced semiconductor assembly provider, leads the advanced packaging boom. These technologies integrate high-bandwidth memory with AI processors—critical for data centers from Silicon Valley to Singapore. Analog Devices reports strong demand from industrial and data center customers worldwide.
Memory markets show cyclical recovery. DRAM demand from AI training systems is tightening global supply after manufacturers across South Korea, Taiwan, and the U.S. spent years cautiously managing capacity following pandemic overbuilding. Prices depressed through 2023 are now stabilizing.
Cisco's new Silicon One G300 networking chip targets global AI infrastructure. "AI at scale demands open, standards-based networking that customers can deploy with confidence across diverse environments," said Yousuf Khan. AI systems require networking and interconnect technologies beyond compute chips alone.
European supplier ams OSRAM AG secured over EUR 500 million in design wins for Digital Light technology, highlighting growth in photonics and optical components. These enable high-speed chip-to-chip communication essential for distributed AI training across continents. The Austrian-German company expects modest FY26 revenue softening from divestments and currency headwinds but maintains technology investments.
Recovery patterns vary by segment. AI-focused areas—high-bandwidth memory, advanced packaging, networking ASICs, and photonics—see robust growth globally. Traditional chip segments face ongoing cyclical pressure.
Manufacturers worldwide now plan capacity cautiously. Pandemic-era oversupply taught hard lessons about overexpansion. Current AI demand is substantial but concentrated in specific technology nodes and package types. Companies invest in 2.5D and 3D packaging, CoWoS (chip-on-wafer-on-substrate) technologies, and silicon photonics rather than broad expansion.
AI hardware requirements drive innovation beyond traditional Moore's Law scaling. Training large language models requires thousands of GPUs with high-bandwidth memory and low-latency interconnects. Inference at scale demands energy-efficient ASICs. These needs push semiconductor development toward heterogeneous integration and specialized architectures across global supply chains.