Google Reportedly in Talks with Marvell to Develop Custom AI Chips, Including Memory Processing Unit
Google taps Marvell for custom TPU and memory processing unit to diversify chip supply.
Google is reportedly negotiating with Marvell Technology to develop new custom AI chips, according to The Information. The partnership would involve creating a next-generation Tensor Processing Unit (TPU) and a dedicated memory processing unit designed to support Google's TPU infrastructure. The memory processing unit is expected to complete design next year and enter test production, marking a strategic shift in Google's chip supply chain.
This collaboration aims to diversify Google's AI chip sourcing beyond its long-time partner Broadcom, potentially improving performance and cost efficiency for large-scale AI workloads. By working with Marvell, Google could accelerate the development of specialized hardware tailored to its growing AI needs, including training and inference for models like Gemini. The move also reflects the broader industry trend of tech giants investing in custom silicon to gain a competitive edge in AI.
- Google is in talks with Marvell Technology to co-develop custom AI chips, including a new TPU and a memory processing unit.
- The memory processing unit is slated for design completion next year, followed by test production.
- This partnership aims to diversify Google's AI chip supply chain beyond Broadcom, potentially improving performance and cost efficiency.
Why It Matters
Custom AI chips give Google more control over performance and cost for massive AI workloads.