| Mobile Web

Search results for TPU

AI & Enterprise

Google develops Frozen v2 AI inference chip, up to 10 times more efficient than TPU

Google is developing a new server chip that directly integrates the design of its Gemini AI model, The Information reported on July 21, citing two people familiar with internal matters. The chip, called Frozen v2 inside Google, is aimed at addressing an AI computing power shortage that has fueled internal conflict and led Google Cloud to turn down external customer contracts. Employees involved expect Frozen v2 to be 6 to 10 times more efficient than the latest TPU by tokens processed per unit of power. Google plans to deploy it as early as 2028.

AI & Enterprise

Google to take on South Korea AI market with Gemini Enterprise

Google is targeting South Korea\'s AI market by promoting the full-stack structure of Gemini Enterprise and its compatibility with other companies\' models. Ruth Sun (루스 선), CEO of Google Cloud Korea, said Google’s ownership of all layers from chips to models and support for more than 200 models, including Anthropic’s Claude, delivers strong return on investment. She cited Samsung Electronics’ DX division as a user. Google also highlighted security offerings and growing AI use in marketing and sales.

AI & Enterprise

ByteDance targets mass production of in-house AI CPU by 2027 to cut GPU reliance

TikTok owner ByteDance is accelerating development of a next-generation in-house central processing unit as AI services drive a surge in computing demand. It is seeking to reduce reliance on external GPUs and raise the share of its own semiconductors. The company aims to complete the CPU design by early 2027 and begin mass production and internal deployment in the second half of that year. It is also working with Qualcomm and seeking foundry capacity.