Google is reportedly in discussions with Marvell Technology to develop two new custom AI chips, a move that signals a strategic broadening of its hardware supply chain. The potential partnership focuses on a memory processing unit and a Tensor Processing Unit (TPU) optimized for inference—the phase where trained AI models execute tasks and respond to queries.
If finalized, the deal would make Marvell the third major design partner in Google’s custom silicon stable, joining Broadcom and MediaTek. While Broadcom has historically been the primary architect for Google’s high-end TPUs, the search giant appears increasingly interested in diversifying its dependencies. This shift comes as the industry moves away from general-purpose computing toward highly specialized hardware designed to handle the massive data throughput required by generative AI.
The negotiations highlight the intensifying silicon arms race among hyperscalers. By bringing Marvell into the fold, Google gains not only technical redundancy but also greater leverage in a market where specialized manufacturing capacity is at a premium. While no contract has been signed yet, the direction is clear: the future of AI will be defined as much by the physical architecture of the data center as by the algorithms that run within it.
With reporting from The Next Web.
Source · The Next Web
