Alphabet, the parent company of Google, is reportedly designing a new server chip called “Frozen v2” to boost the efficiency of its in-house Gemini models. The chip, expected to debut sometime in 2028, promises to be six to ten times more efficient than current AI chips, measured by tokens generated per unit of power.
The move comes as a growing number of AI companies seek to produce their own custom hardware for greater efficiency and independence from Nvidia’s dominance. Google’s response to TechCrunch was noncommittal, focusing on the company's commitment to innovation and integrated system optimization.
Such investments have been necessary amid concerns about AI spending, dampening market enthusiasm in the tech industry. With plans to spend between $180 billion and $190 billion, Google’s future depends on proving that these investments will yield tangible benefits. The announcement of Frozen v2 appeared to placate investors, pushing Google's stock up by 3%.
Meanwhile, other AI firms are making their own moves; OpenAI recently unveiled its first custom chip and Anthropic is reportedly partnering with Samsung. As the tech race intensifies, one can only wonder what secrets lie within these frozen cores.
