Google is in talks with Samsung to build next-gen 'Icefish' TPU on 1.4nm and 2nm process split - TweakTown

Google is reportedly in talks with Samsung to manufacture part of its next-generation AI processor, the 10th-gen Tensor Processing Unit codenamed 'Icefish'. The chip's production is expected to be spl... Read original →

Google’s split of the Icefish TPU between TSMC’s 1.4nm compute die and Samsung’s 2nm I/O die signals a definitive shift toward heterogeneous integration in AI chip design. This move intensifies co-evolution pressure on HBM and advanced packaging, escalating competition for CoWoS and FOPLP capacity. From a compliance standpoint, overreliance on Taiwan, China as a single manufacturing node now constitutes strategic vulnerability, with U.S. CHIPS Act incentives pushing hyperscalers toward decentralized supply chains. In response, NVIDIA may accelerate its custom SoC development tightly coupled with TSMC’s 3DFabric, while Intel leverages IFS and EMIB to court AI clients. Over the next 18 months, Samsung—bolstered by its memory-logic integration edge—could capture over 15% of the custom AI chip foundry market, though 2nm yield ramp remains the critical bottleneck.