Google Teams with Marvell on Two Inference Chips to Rival Nvidia
Google is developing two custom AI inference chips in partnership with Marvell Technology, expected to debut at this week's Google Cloud Next conference in Las Vegas. This initiative shifts its tensor processing units toward inference workloads, diversifies its supplier base beyond Broadcom, and positions the company to challenge Nvidia directly.
1. Marvell Partnership on Inference Chips
Google is in discussions with Marvell Technology to co-develop two custom AI inference chips aimed at improving model processing efficiency, with plans to unveil these alongside a new generation of tensor processing units at Google Cloud Next in Las Vegas this week.
2. Supplier Diversification Strategy
This collaboration marks a strategic effort to broaden Alphabet's TPU supply chain beyond its longtime partner Broadcom, reducing dependency risks and fostering innovation through multiple manufacturing sources.
3. Inference Market Competition
By focusing on inference-optimized hardware, Google aims to directly challenge Nvidia’s dominance in AI deployment, catering to growing customer demand for specialized inference solutions and strengthening its position in the evolving AI infrastructure market.