According to a report by people familiar with the matter cited by The Information on July 20, Google is developing a new server chip codenamed “Frozen v2,” aiming to significantly improve the operational efficiency of its Gemini series AI models. After the news broke, Alphabet shares briefly rose by about 3% intraday on July 20, closing up 1.52% at $351.37.
Frozen v2 Technical Specifications: 6 to 10 Times Efficiency, Earliest Deployment in 2028
Per people familiar with the matter cited by The Information, Frozen v2’s core technical feature is that it writes part of the Gemini model architecture directly into the chip hardware. This reduces the number of compute steps and data movement, thereby improving query response speed. Measured by the number of AI tokens that can be processed per unit of electricity, Frozen v2’s efficiency could reach 6 to 10 times that of Google’s current TPU.
The trade-off in the design is flexibility: if future Gemini model versions do not use the same underlying architecture, Frozen v2 likely cannot be directly compatible. Google’s engineering team is still finalizing design details, including how much of the model information is written into the hardware, with the earliest deployment targeted for 2028. Google has not officially acknowledged or denied the project.
Tight Compute Capacity Backdrop for Google: Nearly $1 Billion Monthly SpaceX Contract
The backdrop for the rumored Frozen v2 involves multiple compute and competitive challenges Google is currently facing:
Tight compute capacity: Google’s cloud division previously declined some partnership proposals with external clients for this reason
SpaceX compute contract: Google agreed last month to pay SpaceX nearly $1 billion per month to help make up for shortfalls in enterprise compute commitments
Gemini Pro delayed: The release timeline for the next-generation Gemini Pro model has been pushed back, and the company has also recently lost several senior research employees
Competitive pressure increasing: OpenAI, Anthropic, and Grok have launched new models, and Chinese AI models (including Moonshot and Alibaba, etc.) have reached a 45% share of token usage by U.S. enterprises
Guo Ming-chi’s Leaks: Triggerfish by MediaTek—Will Taiwan’s Supplier Chain Benefit Again Become a Market Focus?
In late June 2026, semiconductor industry chain analyst Guo Ming-chi disclosed that Google has deepened its partnership with MediaTek. Based on the TPU v9 architecture, the two sides are jointly developing an upgraded chip codenamed “Triggerfish,” and MediaTek has obtained exclusive access to this new order with a higher unit price. As of the time of the report, the partnership plan had not yet received official formal confirmation.
With the Frozen v2 news surfacing, the market has started to focus on whether Google will continue its collaboration model with Taiwan’s supply chain. Since Google has previously established partnerships with MediaTek and TSMC for the production of TPU-series chips, if Frozen v2 enters mass production, whether Taiwan’s semiconductor supply chain can once again play a key role will be a key focus for ongoing market observation.
FAQ
What are the technical specifications of the Google Frozen v2 chip, and when will it be deployed?
According to people familiar with the matter cited by The Information, Frozen v2 will write the Gemini model architecture directly into the chip hardware, and the efficiency per unit of electricity in processing AI tokens could reach 6 to 10 times that of the existing TPU. The earliest deployment is in 2028, and Google has not officially acknowledged or denied this project.
How does the Frozen v2 news affect Alphabet stock?
According to market data for July 20, 2026, Alphabet shares rose by about 3% at one point during intraday trading and closed up 1.52% at $351.37.
What is the relationship between MediaTek Triggerfish leaked by Guo Ming-chi and Frozen v2?
According to Guo Ming-chi’s leak in late June 2026, MediaTek has obtained exclusive orders for Google’s TPU v9 upgraded chip (codenamed Triggerfish), as an extension of the existing Google–Taiwan supplier collaboration. Frozen v2 is another new in-house chip development path, and whether the Taiwan supplier partnership will continue remains to be observed based on official announcements.
