[News] Samsung Reportedly in Talks for Google 2nm TPU I/O Die Order; 2028 Mass Production Eyed - TrendForce

According to multiple reports, Google is reportedly considering Samsung to manufacture key components for its next-generation AI processor, specifically a 2nm memory I/O die that connects the processo... Read original →

Google’s move to assign Samsung the 2nm I/O die isn’t just about foundry diversification—it’s a strategic pivot toward memory-logic co-design in AI accelerators. Technically, tight integration between HBM and I/O dies accelerates the shift to heterogeneous integration, pressuring TSMC to fast-track CoWoS-R and SoIC. Samsung leverages its HBM3E lead (half a generation ahead of SK hynix) to unify memory, foundry, and packaging under one roof. Geopolitically, mass production at Taylor, Texas mitigates U.S.-China tech decoupling risks but demands strict compliance with CHIPS Act localization thresholds. TSMC retains the 1.4nm compute die edge, yet faces client-driven counterbalance; Intel may exploit this by pushing Intel 18A with Groq-like innovators. Within 18 months, this deal will cement chiplet-based architectures as the AI hardware norm, transforming Samsung from a memory vendor into a full-stack AI silicon enabler.