Full story
Frozen v2 for Gemini
Alphabet is developing a new server chip internally dubbed "Frozen v2" to run Gemini models more efficiently, with The Information reporting the design would hardwire parts of Gemini's architecture directly into the silicon.
The chip is described as potentially "six to 10 times more efficient" than Google’s latest Tensor Processing Units (TPUs) when measured by the number of AI tokens processed per unit of power, and it is reportedly being designed specifically for Gemini rather than as a general-purpose AI chip.
The Information’s report also says Google engineers estimate Frozen v2 would reduce computation and data movement needed to generate AI responses, and that Google views Frozen v2 as a new family of AI chips rather than a replacement for TPUs.
CNBC reported Alphabet shares closed 1.51% higher on Monday after the report, and it said the company is targeting 2028 for deployment.
The Decoder similarly frames Frozen v2 as embedding Gemini’s architecture directly into silicon for efficiency gains, while noting Google plans to deploy it starting in 2028 and sees it as a test run for specialized chips.
Co-design and trade-offs
Google told CNBC in a statement that its teams are "constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers," while adding that "while not every project moves into production, this rigorous exploration is central to our full stack approach."
The Information’s reporting, as relayed by CNBC, says Frozen v2 would permanently embed parts of Gemini's architecture directly into the silicon, reducing the number of calculations and amount of data movement required to answer queries.
TechCrunch reported that in a response to its inquiry, Google did not directly confirm the report but said, "Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers," and that "While not every project moves into production, this rigorous exploration is central to our full stack approach."
TechCrunch also said the chip is slated to be released sometime in 2028 and could be between six and 10 times more efficient than Google’s existing AI chips, measured by the number of tokens generated per unit of power.
The Decoder adds that Frozen v2 takes a more flexible path by embedding the model architecture instead of weights, with new weights still loadable onto the chip, even as it remains tied to the same underlying architecture.
Compute crunch and next steps
CNBC said the project is aimed at easing a major internal compute shortage that has fueled tensions and reportedly forced Google Cloud to turn away outside business, and it tied the effort to broader pressure on AI infrastructure.
“Finimize Pro The Information says the “Frozen v2” design could be 6-10 times more power-efficient than Google’s latest custom AI chips, with deployment penciled in for 2028”
CNBC also reported that just last month, Google agreed to pay SpaceX nearly $1 billion a month to help bridge the gap and meet its enterprise compute commitments.
The Decoder describes Frozen v2 as a test run for specialized chips with a smaller production volume than the TPU line, and it says Google viewed Frozen v2 as a new family of AI chips rather than a replacement for TPUs.
TechCrunch reported that after The Information’s report, the company’s stock climbed some 3% on Monday morning ahead of its earnings report later this week.
Across the coverage, the chip’s trade-off is that it works with future Gemini models only if Google sticks with the same underlying architecture, and The Information’s reporting as relayed by CNBC says Google reportedly currently views Frozen v2 partly as a trial run and does not plan to produce it at the same scale as its TPUs.
