In a strategic move to optimize its artificial intelligence infrastructure, Google is reportedly designing a new server chip that integrates core elements of its Gemini model directly into the hardware architecture. Known internally as "Frozen v2," this custom silicon is intended to alleviate the current AI computing bottleneck that has forced the company’s cloud division to turn away potential external clients. By hardwiring aspects of the model, Google aims to streamline AI service delivery, potentially achieving efficiency gains six to ten times greater than its current custom hardware when measured by tokens processed per unit of power.

Scheduled for a potential rollout by 2028, the project represents a significant shift in how Google approaches its data center capabilities. While the company is still refining the design, these chips are intended to complement, rather than replace, Google’s existing Tensor Processing Units (TPUs). This development follows recent internal challenges regarding model performance, including reports of delays in launching newer iterations of Gemini as the engineering team works to enhance the model's coding proficiencies. As competition in the AI space intensifies, this proprietary hardware strategy highlights Google's ongoing commitment to building a more scalable and efficient ecosystem for its generative AI tools.