Google is developing new AI chips, including a memory-focused processor and an upgraded tensor processing unit (TPU), according to The Information. The effort reflects a direct push to reduce reliance on Nvidia’s GPUs in high-performance AI workloads.
The proposed memory processing unit is designed to work alongside existing TPUs, improving efficiency in handling large-scale model operations. Google aims to finalize the chip design by next year before moving into test production. The company is also expanding partnerships with manufacturers such as Intel and Broadcom to support rising infrastructure demand.
Can Google Close The Gap With Nvidia In AI Hardware?
The initiative comes as competition in AI accelerators intensifies across major technology firms. Nvidia continues to dominate the sector with its GPU ecosystem, while advancing new inference chips that integrate emerging architectures. But, Google’s vertically integrated approach could offer cost and performance advantages within its cloud platform.
Adoption of TPUs has already contributed to growth in Google Cloud, where AI infrastructure is becoming a key revenue driver. Compared with traditional GPU deployments, custom silicon allows tighter optimization between hardware and software, particularly for large-scale model training and inference tasks.
Google’s hardware push aligns with its broader AI strategy, including the recent launch of Gemma 4 models. These models are designed for advanced reasoning and agent-based workflows, with features such as structured outputs and API integration. The combination of custom chips and optimized models suggests a coordinated effort to control the full AI stack.
The next catalyst will be Google’s April 29 earnings report, where investors are expected to assess cloud performance and capital allocation toward AI infrastructure as competition with Nvidia accelerates.
