Google has entered negotiations with semiconductor firm Marvell to design and manufacture in-house AI accelerator chips. The partnership aims to create custom silicon optimized for Google’s machine learning workloads, potentially lowering costs and improving performance compared with off-the-shelf GPUs.

These discussions represent a direct challenge to Nvidia’s stronghold on the AI accelerator market, where its GPUs power the vast majority of data center AI workloads. If successful, Google’s move could erode Nvidia’s market share and force the company to adjust pricing or accelerate its own chip innovations.

The Google-Marvell collaboration may prompt other hyperscale cloud providers to explore bespoke AI chips, reshaping vendor dynamics in the data center. A shift toward proprietary silicon could lead to greater diversity in AI hardware and intensify competition across the semiconductor landscape.