š° Full Story
Alphabet's Google is in discussions with Marvell Technology to design two custom chips for running AI models, multiple reports said on April 19, 2026.
One chip would be a memory processing unit (MPU) intended to work alongside Google's Tensor Processing Units (TPUs); the other would be a TPU optimized for inference workloads.
The talks have not produced a signed contract and were not immediately confirmed by the companies.
According to reporting, Google and Marvell aim to finalise the MPU design potentially as soon as 2027 before moving to test production.
The move would add Marvell as a design partner alongside existing suppliers such as Broadcom, MediaTek and TSMC, reflecting Google's strategy to diversify its custom silicon supply chain and to halve serving costs as inference becomes the dominant compute expense.
The partnership, if concluded, would bolster Marvellās position in cloud custom silicon while advancing Googleās efforts to offer TPUs as an alternative to Nvidia GPUs.
š Based On
š¤ Social Media Insights
Social Summary
Key takeaways: Marvell would most likely be a design partner (not a fabricator), leveraging its interconnect and memory IP to build MPU/coāprocessor elements for Googleās TPU ecosystem. The aim is supplyāchain diversification to reduce Nvidia reliance; if realized, this could affect GPU market dynamics.







š¬ Commentary