Google is working on a new AI chip designed to make Gemini more efficient
Source Entity
Lucas Ropek
Google is reportedly developing a new AI server chip, codenamed 'Frozen v2', aimed at significantly boosting the efficiency of its Gemini models. Targeted for a 2028 release, the chip promises a six to ten-fold increase in token generation per unit of power.
The Future of AI Infrastructure: Google's 'Frozen v2' Strategy
Alphabet, the parent company of Google, is reportedly embarking on an ambitious hardware initiative to refine the computational efficiency of its flagship AI models, Gemini. According to recent reports, the company is developing a new server chip internally referred to as “Frozen v2.” This development highlights a broader industry shift where major tech firms are moving away from total reliance on third-party silicon providers to create custom-tailored hardware that aligns perfectly with their specific software architectures.
The Quest for Computational Efficiency
At the core of this initiative is the critical metric of efficiency. The report suggests that 'Frozen v2' is engineered to be between six and 10 times more efficient than Google’s current generation of AI chips. By measuring this efficiency through the number of tokens generated per unit of power, Google is addressing the most significant bottleneck in modern artificial intelligence: the immense energy cost required to run large language models at scale. As models like Gemini become more sophisticated, the power demands for inference and training grow exponentially, making hardware optimization a primary business imperative.
Strategic Hardware Verticalization
Google’s move to design its own silicon is a continuation of its long-standing strategy to control the full stack of its technological ecosystem. By building custom chips, the company can optimize its hardware specifically for the transformer architecture that powers Gemini, effectively bypassing the generalized limitations of off-the-shelf hardware. This vertical integration is essential for maintaining a competitive edge in the high-stakes AI market, where latency and cost-per-query are the defining factors for long-term profitability and user experience.
Long-Term Roadmap and Industry Trends
With an expected release window of 2028, the 'Frozen v2' project underscores the long-term planning required to stay ahead in the semiconductor space. The lead time for developing high-performance AI silicon is substantial, often spanning several years from design to deployment. This timeline indicates that Google is positioning itself for a future where AI integration is pervasive across all consumer and enterprise products, necessitating a hardware backbone that is vastly more sustainable and cost-effective than current solutions.
Corporate Response and Future Outlook
While Google has neither confirmed nor denied the specifics of the 'Frozen v2' project, their official statement emphasizes a culture of continuous research and rigorous exploration. By noting that not every project moves into production, the company maintains a level of strategic ambiguity common in the hardware sector. However, the pursuit of such high-efficiency gains suggests that the industry is entering a new phase of maturity, focusing less on raw power and more on the optimization of energy consumption as a key performance indicator.