Google Develops 'Frozen v2' AI Chip to Supercharge Gemini Efficiency
Mountain View, CA — Alphabet, Google's parent company, is designing a new server chip internally codenamed "Frozen v2" that promises to dramatically boost the efficiency of its Gemini AI models, according to a report from The Information.
The chip, slated for release sometime in 2028, could be between six and ten times more efficient than Google's existing AI processors, measured by the number of tokens generated per unit of power consumed.
Fierce Competition in Custom Silicon
The move places Google in the growing ranks of AI companies designing their own chips. OpenAI announced its first custom inference processor — codenamed "Jalapeño" — just last month. Reports also surfaced this month that Anthropic is discussing a chipmaking partnership with Samsung.
The trend reflects a broader industry effort to reduce dependence on Nvidia, which has historically dominated the AI chip market and left major AI companies reliant on its hardware.
Investor Confidence Boost
News of the Frozen v2 chip appeared to reassure investors who had been concerned about Alphabet's massive planned expenditures. Earlier this year, Google said it plans to spend between $180 billion and $190 billion to build out its AI strategy. Following the Frozen v2 report, Alphabet's stock climbed about 3% on Monday.
"Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers," Google told TechCrunch in a statement. "By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized for real-world workloads."
Efficiency Becomes a Key Battleground
AI chip efficiency has become a central selling point as concerns about AI spending have dampened some of the market euphoria that previously characterized the industry. The ability to generate more tokens per watt of power directly translates to lower operating costs for AI workloads.
Google already produces its own Tensor Processing Units (TPUs), but Frozen v2 represents a significant leap: the 6-10x efficiency gain would be a generational improvement over its current custom silicon.
What's Next
With a 2028 target, Frozen v2 won't arrive for at least two more years. But its development signals that the custom AI chip race is only intensifying. As more companies design their own silicon, the landscape of AI hardware — long dominated by a single player — is becoming increasingly competitive.
