Google is working on a new server chip built specifically to run its Gemini AI models more efficiently. The chip is codenamed ‘Frozen v2’ internally, and according to TechCrunch, it could be six to ten times more efficient than Google’s existing AI chips, measured by how many tokens the chip generates per unit of power. The target release window is sometime in 2028.
The Information first broke the story, citing anonymous sources. Google didn’t confirm it outright, but didn’t deny it either. In a statement to TechCrunch, the company said: ‘Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers. While not every project moves into production, this rigorous exploration is central to our full stack approach.’
That careful non-denial is notable. Google has been under real pressure to show that its massive AI spending will actually produce results. Earlier this year, the company said it plans to spend between $180 billion and $190 billion building out its AI infrastructure. Investors have been watching closely, and news of a more efficient chip appears to have given them some reassurance. Google’s stock climbed around 3% on Monday morning after The Information’s report came out, ahead of the company’s earnings report later this week.
This is part of a much bigger shift happening across the AI industry. Companies that once relied entirely on Nvidia chips are now racing to build their own. Nvidia has long dominated the AI chip market, which has left major AI players heavily dependent on one supplier. That’s a risky position, and efficiency concerns have made it even more urgent to find alternatives.
The push for custom chips is also a direct response to cost pressures. Running large AI models is expensive, and the market’s early enthusiasm for AI spending has cooled considerably. Efficiency is now a major selling point, both for internal use and for customers buying AI services.
Google is not alone in this effort. Other major AI companies are moving in the same direction:
- OpenAI announced its first custom chip in June, an inference processor called Jalapeño.
- Anthropic is reportedly in talks with Samsung about a new chipmaking partnership.
Google has its own chip history to build on. The company has been developing its Tensor Processing Units, known as TPUs, for years. Those chips already power a significant portion of Google’s AI workloads. A new chip that’s dramatically more efficient would give Google more control over its costs and reduce how much it depends on outside suppliers.
If Frozen v2 delivers on the reported efficiency gains, it could meaningfully change the economics of running Gemini at scale. That matters not just for Google’s bottom line, but for how competitively it can price AI services against rivals like Microsoft and Amazon. The chip isn’t expected until 2028, so there’s still a long road ahead, but the early signal from investors suggests the market likes what it’s hearing.
