Google Develops Frozen v2 Chip to Accelerate Gemini AI Inference

Google is currently designing a new custom server chip codenamed Frozen v2, aimed at optimizing its Gemini artificial intelligence models. This initiative reflects the company's broader effort to integrate AI hardware and software from the ground up, ensuring higher computational efficiency. Reports indicate the chip is designed to serve six to 10 times more AI tokens per unit of power compared to the company's existing hardware, directly addressing current limitations in AI computing capacity that have previously constrained Google Cloud operations.

The development of Frozen v2 comes as high demand for AI infrastructure forces technology giants to rethink their hardware roadmaps. Google has historically relied on its Tensor Processing Units (TPUs) to power both internal workloads and external cloud services, but rising computational needs have necessitated more specialized solutions. By creating a separate family of AI chips specifically for inference tasks, Google intends to complement rather than replace its existing TPU portfolio, allowing for better scaling of its most advanced generative models.

Engineers are still finalizing the architectural design and hardware specifications, with a planned deployment timeline targeting 2028. This move signifies an intensification of the semiconductor race, as companies face mounting pressure to justify heavy investments in AI infrastructure by demonstrating clear performance gains. For India, which serves as a major hub for Google's cloud engineering and research operations, the advancement in localized inference hardware could eventually lower the operational costs and latency for AI-driven services delivered to the local market.