# Google 拟将 Gemini 架构固化到芯片，能效提升 6-10 倍

- 来源：Chubby♨️ (@kimmonismus)
- 发布时间：2026-07-20 21:16
- AIHOT 分数：64
- AIHOT 链接：https://aihot.virxact.com/items/cmrt94pdm1lfkbitlf7l6tsnm
- 原文链接：https://x.com/kimmonismus/status/2079193746944790666

## AI 摘要

Google 正计划将 Gemini 部分模型架构直接固化到专用芯片中，该芯片代号“Frozen v2”，目标能效比（每瓦 token 数）比最新 TPU 提升 6-10 倍，最早 2028 年部署。此举源于 Google AI 算力严重短缺，其云部门已因此拒绝外部客户。但该方案也带来刚性：未来 Gemini 模型若使用 Frozen v2，必须保持底层架构不变。

## 正文

Google may be preparing to freeze parts of Gemini's architecture directly into silicon.

Informally called "Frozen v2，" the chip reportedly targets 6-10× more tokens per watt than Google's newest TPUs. Deployment is planned for as early as 2028.

The motivation is immediate： Google's AI compute shortage has reportedly become severe enough that its Cloud division has turned down outside customers.

The efficiency comes with rigidity. Future Gemini models could use Frozen v2 only while retaining the same underlying architecture.
TPUs reduced Google's dependence on Nvidia. Frozen v2 would go further， tying Gemini's architecture directly to the silicon running it.