Google Plans Custom Gemini Hardware to Ease AI Computing Constraints

Google is reportedly working on a next-generation server chip designed specifically to improve the performance of its Gemini artificial intelligence models. According to a report by The Information, the new chip—internally referred to as “Frozen v2”—would integrate selected components of Gemini directly into the hardware, enabling the AI system to operate more efficiently and consume less power.

The project is part of Google’s broader effort to strengthen its AI infrastructure as demand for generative AI services continues to rise across cloud, enterprise, and consumer applications.

The report suggests that Google has been facing increasing pressure on its AI computing resources. The growing demand for cloud-based AI services has reportedly created capacity constraints, making it difficult for Google Cloud to accommodate all external customer requirements. By developing a more efficient custom chip, the company aims to increase the number of AI requests that can be processed while reducing energy consumption.

Engineers are still refining the design of the chip and deciding how much of the Gemini model’s information should be embedded directly into the hardware. This approach could reduce the need for repeated data transfers between memory and processors, resulting in faster response times and improved overall performance.

Google is expected to deploy the chip as early as 2028, although the timeline could change as development progresses. If completed successfully, Frozen v2 could represent a significant advancement in AI hardware architecture.

One of the most notable claims in the report is that the new chip may be six to ten times more efficient than Google’s current custom AI processors when measured by the number of AI tokens generated per unit of power. Higher efficiency would not only lower operational costs but also help address concerns about the increasing energy demands of large-scale AI systems.

The development of Frozen v2 highlights the growing competition among major technology companies to build specialised AI chips tailored to their own models and services. As AI applications become more sophisticated, hardware designed specifically for those models is expected to play a crucial role in improving speed, scalability, and energy efficiency.