Google develops Frozen v2 server chip to embed Gemini AI
Google is working on a next-generation server chip called Frozen v2, which will pack some of its Gemini AI smarts right into the hardware.
This project comes as Google faces a shortage of computing power, so much so that Google Cloud has had to turn away some customers.
The news got investors excited, pushing Alphabet's stock up 3% in early trading.
Frozen v2 expected 6-10x efficiency
Frozen v2 is expected to be six to 10 times more efficient than Google's current AI chips, meaning it'll handle way more AI tasks using less energy.
Engineers are figuring out how much of the Gemini model can fit directly onto the chip, and it'll work alongside Google's existing TPUs rather than replace them.
The chip should be ready by 2028, helping Google keep up with growing demand for powerful AI tools, even as its latest Gemini model faces delays and tweaks behind the scenes.