Google is reportedly working with semiconductor company Marvell Technology to design new AI chips aimed at improving the efficiency of running large artificial intelligence models. According to a report cited by Reuters, the discussions involve two specialized chips: a memory processing unit designed to complement Google’s existing Tensor Processing Units (TPUs), and a next-generation TPU built specifically for AI workloads. The goal of the collaboration is to enhance performance and reduce costs in AI computing, particularly as demand for large-scale model training continues to grow. Google has been steadily investing in its TPU ecosystem as an alternative to Nvidia’s dominant GPUs, and these chips are increasingly important for its cloud business and AI services. If finalized, the memory processing unit design could be completed as early as next year before moving into testing and production. The move reflects broader industry competition as major tech companies race to build custom silicon to power increasingly complex AI systems more efficiently.
Read more here
