Google is reportedly working on a new custom server chip that could embed certain parts of its Gemini AI model directly into hardware, a move aimed at making AI processing faster and more energy efficient.
According to a report by The Information, the chip, internally codenamed “Frozen v2,” is being developed to overcome AI computing capacity constraints that have limited Google Cloud’s ability to take on more external customer workloads.
The report states that Google is targeting a rollout around 2028, although engineers are still refining the chip’s design and deciding how much of the AI model should be built directly into the hardware. If successful, Frozen v2 could deliver six to ten times greater efficiency than Google’s current AI chips by serving more AI tokens while consuming less power.
Responding to the report, a Google Cloud spokesperson said the company is continuously exploring new technologies and emphasized that designing hardware and software together allows Google to build highly integrated and optimized AI systems.
Rather than replacing Google’s existing Tensor Processing Units (TPUs), the Frozen project is expected to introduce a separate family of AI chips focused on specific inference workloads. TPUs will continue to support the company’s broader AI infrastructure and cloud services.
Google’s reported plans come as the race to build advanced AI hardware intensifies across the technology industry. Heavy investment in AI infrastructure has fueled strong growth in semiconductor companies, while analysts continue to caution that soaring expectations for AI demand will require companies to prove the long-term value of their investments.
If launched, the Frozen chips would expand Google’s growing portfolio of custom AI hardware by complementing its TPUs and improving the efficiency of AI inference tasks, helping the company deliver faster and more cost-effective AI services.
Also Read: Amazon and Google Chip Away at Nvidia’s AI Chip Supremacy
