Google develops new AI chip to improve Gemini efficiency

The custom chip, internally codenamed Frozen v2, is expected to be released in 2028.

Google is developing a new server chip aimed at improving the efficiency of its Gemini artificial intelligence (AI) models, according to a TechCrunch report citing The Information.The custom chip, internally codenamed Frozen v2, is expected to be released in 2028.

According to the report, the processor could be six to 10 times more efficient than Google’s current AI chips, based on the number of tokens generated per unit of power.Responding to TechCrunch, Google neither confirmed nor denied the report.“While not every project moves into production, this rigorous exploration is central to our full stack approach.

By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimised for real-world workloads,” the company told TechCrunch.AI firms invest in custom chipsTechnology companies are increasingly developing proprietary AI chips to improve the performance and efficiency of their models while reducing dependence on third-party hardware suppliers.

The move also comes as companies seek to address constraints in global AI computing capacity and optimise the cost of deploying large AI models.The industry is also attempting to reduce its reliance on Nvidia, whose dominance in AI hardware has made major AI developers dependent on its chips.Alphabet has faced investor scrutiny over its spending on AI infrastructure.

Earlier this year, Google said it plans to invest between $180 billion and $190 billion as part of its AI strategy, making improvements in performance and efficiency critical to demonstrating returns on those investments.