Google Reportedly Developing Gemini-Specific Chip Frozen v2

According to The Information, Google is developing a custom inference chip codenamed Frozen v2 to alleviate compute shortages and serve Gemini more efficiently. The news remains a rumor, with Google yet to publicly confirm project details.

Key Details

Frozen v2 is not a direct replacement for general-purpose TPUs; instead, it hardcodes parts of Gemini's model logic and architecture directly into the chip. According to @ns123abc, @mark_k, and others, the team expects the new chip to serve 6 to 10 times more tokens per unit of power compared to Google's latest generation TPUs. The chip is expected to launch as early as 2028. @ns123abc also noted that this chip line stems from a more aggressive design direction proposed earlier by Jeff Dean.

Background and Impact

Reports indicate that due to severe compute constraints, Google Cloud has even started turning down some orders. @Afinetheorem believes this reflects a clear trend in the AI industry: companies are leaning towards burning more model logic into silicon, optimizing efficiency through deep synergy between software and custom hardware, which could alter the current market structure and competitive landscape.

2026-07-20 ~ 2026-07-21 · 9 related posts

• Google is reportedly building a Gemini-specific chip — zephyr_z9 · 2026-07-20
• Google’s rumored Frozen v2 chip could deeply tie hardware to Gemini — kimmonismus · 2026-07-20
• Google’s rumored Frozen chip could be 6–10x more efficient — Afinetheorem · 2026-07-20

6 near-duplicate retellings: ns123abc · ns123abc · toptickcrypto · mark_k · atShruti · 新智元