Google is working on a new type of server chip that would have elements of its Gemini artificial intelligence model built directly into the hardware itself, according to a report from the Information published Monday, which cited sources with knowledge of the project.
The chip, which has been informally nicknamed “Frozen v2” inside the company, is being developed by Google’s parent company, Alphabet. According to the report, the chip is intended to help solve a serious AI computing capacity shortage — a problem that has caused internal tensions and led Google Cloud to turn away potential deals with outside customers.
Alphabet’s stock climbed 3% in early trading following the report.
Here are the key details from the report:
Google is targeting a deployment date as early as 2028 for the new chip, though engineers are still working out the final design and determining how much model data will be permanently embedded into the hardware.
The chip could deliver between six and ten times the efficiency of Google’s most current custom AI chips, measured by the number of AI tokens processed per unit of energy consumed.
The “Frozen” project is focused on building an entirely new line of in-house chips that would exist alongside Google’s existing tensor processing units, known as TPUs — not replace them.
Google had not responded to a request for comment at the time of the report.
Separately, Bloomberg News reported last week that Google pushed back the launch of its newest Gemini AI model after it failed to meet the company’s internal performance targets. The company is said to be working on improvements, with a particular focus on the model’s coding abilities.
