Google in Talks with Marvell Technology to Build New AI Inference Chips Alongside Broadcom TPU Programme
Google is exploring the development of custom AI inference chips, diversifying its partnerships beyond Broadcom. According to The Information, Google is in discussions with Marvell Technology to create two new AI chips: a memory processing unit and an inference-optimised Tensor Processing Unit (TPU).
Background
This move comes days after Broadcom secured a long-term agreement to design and supply TPUs through 2031. Google aims to diversify its custom silicon supply chain, which already includes Broadcom for high-performance chips, MediaTek for cost-efficient variants, and TSMC for fabrication.
The Shift to Inference
Google’s recent Ironwood TPU is designed for the age of inference, where serving AI models becomes the primary demand driver. While training a model requires substantial compute power, inference operates continuously, serving user queries and scaling with demand. Purpose-built inference silicon offers a competitive advantage over general-purpose GPUs in terms of cost and efficiency.
Marvell’s Role
Marvell would contribute design services for Google’s new chips, similar to MediaTek’s involvement on the Ironwood TPU. While discussions are ongoing, no formal contract has been signed yet. The collaboration could result in chips targeting different workload profiles or cost points, supplementing the existing Ironwood offerings.