Google Reportedly Developing 'Frozen v2' Chip to Hardwire Gemini AI Model

Google is developing a new generation of server chips that integrate parts of its Gemini artificial intelligence model directly onto the hardware, a move aimed at boosting efficiency and addressing internal computing constraints, according to a report from The Information citing people familiar with the matter. The project, informally known as "Frozen v2," underscores the tech giant's push for vertical integration in the high-stakes AI race.

Project Aims for Major Efficiency Gains

The initiative seeks to create a new line of homegrown chips separate from, but not intended to replace, the company's existing Tensor Processing Units (TPUs). By hardwiring elements of the Gemini model into the silicon, Google aims to serve its AI models to users more efficiently.

Key details from the report include:

• Deployment Target: Google plans to deploy the chip as soon as 2028, though engineers are still finalizing the design.
• Performance Goal: The chip could be six to 10 times more efficient than Google’s latest custom AI chips, based on the number of AI tokens served per unit of power.
• Strategic Role: The 'Frozen' project represents a new approach to AI acceleration, complementing the more general-purpose TPUs.

Addressing Internal Capacity Constraints

Ad

The development is reportedly driven by an internal "AI computing capacity crunch" at Google. This shortage of processing power has fueled internal tensions and prompted Google Cloud to decline some deals with outside customers due to a lack of resources, The Information's sources said. The new chip architecture is designed to alleviate these pressures by making AI inference significantly more power-efficient.

In response to the report, a Google Cloud spokesperson said, "Our teams are constantly researching and experimenting with new innovations... By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized," without commenting directly on the 'Frozen' project.

Market Reaction and Context

Investors reacted positively to the news, with shares of parent company Alphabet (GOOGL) rising 3.3% in early trading on Monday. The move to create more specialized hardware highlights the immense cost and competitive pressure to optimize AI workloads as the technology becomes more widespread.

The report follows a separate story from Bloomberg News last week, which stated that Google had delayed the launch of its latest Gemini AI model after it failed to meet internal performance goals. Together, these developments illustrate the significant technical and resource challenges Google faces as it competes in the rapidly evolving AI landscape.