Prateek Tanmay
2 min readNew DelhiJul 21, 2026 06:54 AM IST
Search giant Google, is reportedly making a new server chip that may bring some elements of its Gemini model directly into hardware. With this the company aims to make its AI models more efficient for users, according to a report in the Information.
According to the report, the chip codenamed ‘Frozen v2’ is expected to address the limitations in AI computing capacity which has pushed Google Cloud to refrain from deals with outside customers.
The company reportedly plans to deploy the chip as soon as 2028. However, engineers are still finalising its design, and the amount of information that is likely to be hardwired. The chip is said to be about six to 10 times more efficient than Google’s latest custom AI chips based on AI tokens served per unit of power.
“Our teams are constantly researching and experimenting with new innovations… By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized,” a Google Cloud spokesperson said.
The report says the Frozen project is intended to create a separate family of Google-designed AI chips alongside the company’s Tensor Processing Units (TPUs), rather than replace them.
Google’s reported effort comes at a time when AI infrastructure spending continues to reshape the semiconductor industry. Investor push around AI chips has driven sharp gains in chip stocks this year, although analysts have also warned that expectations for AI-related demand remain exceptionally high and companies may face increasing pressure to justify their massive investments in AI hardware.
It needs to be noted that Google has been expanding its portfolio of custom AI hardware in recent years through its TPUs, which power both its internal AI workloads and Google Cloud services. According to the report, the proposed Frozen chips would complement, rather than replace, the company’s TPUs by handling specific AI inference tasks more efficiently.
