Home
Google launches frozen v2 chip embedding Gemini into silicon, boosting token output per watt six to tenfold

By integrating the model directly into the silicon, Google is pursuing a new server chip architecture codenamed Frozen v2 that significantly reduces computational overhead and data transfer bottlenecks. According to multiple media reports, this specialized chip is designed to run the Gemini model with greater efficiency. Reports suggest that a portion of the Gemini architecture will be hardwired into the chip, minimizing the computation and data movement required to answer user queries. Google engineers project that, compared to its latest generation general-purpose AI chip, the TPU (Tensor Processing Unit), the Frozen chip will deliver 6 to 10 times more token processing capability per unit of power.
Scheduled for release in 2028, Google has clarified that this chip will not replace general-purpose TPUs but will instead serve as a specialized branch within its custom silicon portfolio. Following the announcement, the market responded swiftly: Google's Class A shares (GOOGL) rose nearly 3.7% in early trading, while Class C shares (GOOG) peaked with an increase of approximately 3.9%.
In response to media inquiries, Google stated that its team continues to research and test various innovative technologies to deliver optimal performance and efficiency for users and customers. The company emphasized that while not all projects will reach mass production, this rigorous exploration is a core component of its full-stack strategy. Google further explained that by co-designing hardware and software at the fundamental level, the entire system achieves deep integration and high optimization for real-world workloads. However, Google currently classifies Frozen v2 as an experimental project with no plans for mass production akin to the TPU.
Analysts suggest that this project reflects the increasingly constrained computing power situation within Google. Previously, insufficient computational resources intensified internal competition and reportedly forced Google Cloud to decline some external customer business. Just last month, Google agreed to pay nearly $1 billion per month to SpaceX to bridge the computing power gap and meet its commitments to enterprise clients. Frozen v2 appears to be a strategic move to alleviate this pressure.
Related article
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
Related Special Topic Recommendations
Comments (0)
0/500

By integrating the model directly into the silicon, Google is pursuing a new server chip architecture codenamed Frozen v2 that significantly reduces computational overhead and data transfer bottlenecks. According to multiple media reports, this specialized chip is designed to run the Gemini model with greater efficiency. Reports suggest that a portion of the Gemini architecture will be hardwired into the chip, minimizing the computation and data movement required to answer user queries. Google engineers project that, compared to its latest generation general-purpose AI chip, the TPU (Tensor Processing Unit), the Frozen chip will deliver 6 to 10 times more token processing capability per unit of power.
Scheduled for release in 2028, Google has clarified that this chip will not replace general-purpose TPUs but will instead serve as a specialized branch within its custom silicon portfolio. Following the announcement, the market responded swiftly: Google's Class A shares (GOOGL) rose nearly 3.7% in early trading, while Class C shares (GOOG) peaked with an increase of approximately 3.9%.
In response to media inquiries, Google stated that its team continues to research and test various innovative technologies to deliver optimal performance and efficiency for users and customers. The company emphasized that while not all projects will reach mass production, this rigorous exploration is a core component of its full-stack strategy. Google further explained that by co-designing hardware and software at the fundamental level, the entire system achieves deep integration and high optimization for real-world workloads. However, Google currently classifies Frozen v2 as an experimental project with no plans for mass production akin to the TPU.
Analysts suggest that this project reflects the increasingly constrained computing power situation within Google. Previously, insufficient computational resources intensified internal competition and reportedly forced Google Cloud to decline some external customer business. Just last month, Google agreed to pay nearly $1 billion per month to SpaceX to bridge the computing power gap and meet its commitments to enterprise clients. Frozen v2 appears to be a strategic move to alleviate this pressure.
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber











