Key facts
- Google has released Gemini 3.6 Flash, an AI model focused on efficiency and coding.
- The new model offers reduced token usage and API costs.
- Google is reportedly developing a custom AI chip, codenamed Frozen v2, to run Gemini models.
- This chip is projected to be six to ten times more efficient than current TPUs.
- The Frozen v2 chip is targeted for deployment around 2028.
- Alphabet shares saw a roughly 3% increase following the news of the chip development.
Google has released Gemini 3.6 Flash, an updated AI model focused on efficiency and coding, with user feedback cited for improvements. This new iteration offers reduced token usage and lower API costs. Alongside Gemini 3.6 Flash, Google also introduced Gemini 3.5 Flash Lite and Gemini 3.5 Flash Cyber. The flagship Gemini 3.5 Pro model remains in internal testing.
In parallel, Google is reportedly developing a custom server chip, codenamed Frozen v2, specifically designed to run its Gemini AI models. This chip aims to achieve six to ten times greater efficiency compared to current Tensor Processing Units (TPUs) by hardwiring parts of Gemini's architecture directly into silicon. Engineers project a 2028 deployment target for Frozen v2. This initiative is a strategic move to address Google's growing AI infrastructure capacity constraints and reduce its significant reliance on Nvidia's GPUs, a problem that has led the company to turn away potential customers.
Alphabet shares experienced a roughly 3% climb following the news of the custom chip development. The company is investing heavily in AI infrastructure, with reports indicating it is renting a substantial number of Nvidia GPUs from xAI's data centers as a temporary solution. Competitors like Meta, Amazon, Microsoft, and OpenAI are also pursuing custom silicon programs to gain efficiency and reduce dependence on Nvidia.