The 'Frozen v2' Chip: Google's Bet on Multiplying Gemini Model Efficiency

The Race to Break Free from Nvidia's Dominance

Major artificial intelligence companies are racing to design their own custom processing chips, pursuing two intertwined goals: reducing dependence on a single vendor that dominates the market, and achieving higher performance at lower energy costs. In this context, a report from The Information revealed that Alphabet, Google's parent company, is developing a new server chip internally codenamed Frozen v2, slated for release in 2028, with the aim of running its internal Gemini models with unprecedented efficiency.

Compelling Numbers

According to the same report, the new chip could achieve efficiency between six and ten times greater than Google's current generation of AI chips, measured by the number of tokens generated per unit of energy consumed. This specific metric — tokens per watt — directly reflects the operational cost of running large language models at scale.

When TechCrunch reached out to Google for comment on this information, the company neither confirmed nor denied the report, instead pointing to its general development approach:

• Integrated hardware and software design built from the ground up.
• Continuous system optimization tailored to real-world workloads.
• An ongoing pursuit of innovations that maximize both performance and efficiency.

Why Now? The Dual Pressure on Big Tech

This move cannot be understood in isolation from the broader context in which AI companies currently operate. On one hand, investor anxiety is mounting over massive infrastructure spending on artificial intelligence without a clear return. On the other, demand for AI chips continues to far outpace supply, driving every company to seek alternative sources.

Google announced earlier this year capital expenditure plans ranging between $180 and $190 billion. This staggering figure puts direct pressure on leadership to prove that every dollar poured into these investments will deliver tangible long-term value.

Competitors Aren't Standing Still

Google is not the only company taking this path. OpenAI announced its first dedicated inference chip last June, codenamed "Jalapeño," while reports indicate that Anthropic is in talks with Samsung to establish a chip manufacturing partnership. This collective push to reduce dependence on Nvidia — historically the dominant supplier of AI chips — reflects a growing conviction that whoever controls their hardware controls their competitive future.

Market Reaction: A Telling Signal

The leak of the Frozen v2 news triggered an immediate positive reaction in financial markets, with Alphabet's stock rising approximately 3% during the Monday session following the report's publication — coming just ahead of the company's quarterly earnings announcement. This uptick signals that investors view favorably any move that enhances spending efficiency and reduces reliance on external vendors.

The Bottom Line

If the leaks are accurate, Frozen v2 would represent a qualitative leap in Google's AI infrastructure. However, the road to 2028 is long and filled with countless technical and competitive variables. What is certain now is that the battle for supremacy in the world of artificial intelligence is no longer decided solely by model quality — computational efficiency and hardware sovereignty have become indispensable pillars in this intensifying race.

✦ بقلم فريق دروب أيديا