Google Plans New ‘Frozen’ Chip to Run Its AI Models Much More Efficiently

2026-07-20 · Source: The Information · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Google is developing a new server chip, informally dubbed "Frozen v2," designed to directly integrate the blueprint of its Gemini AI model. This strategic initiative aims to significantly enhance the efficiency of serving its AI models to users. Internal projections suggest "Frozen v2" could achieve 6 to 10 times greater efficiency than Google's newest existing homegrown AI chips, based on the number of tokens served per unit of power. This development is crucial for Google to address a major internal AI computing capacity shortage, which has fueled tensions and compelled Google Cloud to turn down deals with outside customers, thereby improving overall AI infrastructure performance and availability.

Key takeaway

For AI Architects and Directors of AI/ML grappling with escalating inference costs and capacity constraints, Google's "Frozen v2" initiative underscores the strategic value of custom silicon. You should evaluate how deeply integrated hardware-software solutions can dramatically improve efficiency, potentially by 6-10x, and alleviate critical resource shortages. Consider exploring custom chip development or prioritizing cloud providers offering highly optimized, model-specific hardware to scale your AI services effectively.

Key insights

Google's "Frozen v2" chip directly integrates the Gemini AI model blueprint for 6-10x greater efficiency.

Principles

• Direct hardware-software integration optimizes AI model serving.
• Custom chip design can yield significant efficiency improvements.

In practice

• Integrate model blueprints directly into custom hardware.
• Measure AI chip efficiency by tokens served per unit of power.

Topics

• AI Chips
• Custom Silicon
• Gemini AI Model
• Google Cloud
• AI Inference Efficiency
• Hardware-Software Co-design

Best for: CTO, VP of Engineering/Data, AI Product Manager, AI Architect, Director of AI/ML, AI Hardware Engineer

Related on AIssential

• Google's "Frozen v2" chip reportedly bakes Gemini's architecture directly into silicon for efficiency gains — Google's "Frozen v2" chip hardcodes Gemini's architecture for 6-10x AI inference efficiency, deploying 2028.
• Google is working on a new AI chip designed to make Gemini more efficient — AI companies are developing custom chips to boost model efficiency and reduce reliance on external hardware providers.
• Google reportedly develops Frozen v2 chip for Gemini AI — Google is developing a highly efficient custom AI chip, Frozen v2, reflecting a broader industry shift towards proprietary silicon for AI.
• Google plans nearly two million new AI chips as it turns to Marvell for custom designs — Google is developing custom AI chips with Marvell to enhance inference capabilities and diversify its supply chain.
• Alphabet Stock Jumps After Information Report of New Google Chip — Google is developing "Frozen v2," a new AI chip 6-10x more efficient, integrating Gemini's blueprint.
• Google Cloud Debuts New AI Chips — AI compute capacity and proprietary silicon are critical differentiators in the competitive cloud and AI model landscape.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Information.