Alphabet's shares increased by 3% after The Information reported on the company's development of a new server chip, known as 'Frozen v2.' This chip is designed to optimize the performance of Gemini models by embedding parts of their architecture directly into the silicon, which could significantly reduce the computational demands for processing queries.
Alphabet stated that it is continuously exploring innovations to enhance performance and efficiency, although not all projects may reach production. The new chip is projected to handle six to ten times more tokens per unit of power compared to Alphabet's existing tensor processing units (TPUs).
However, it will not replace the TPUs but will serve as a specialized addition to Alphabet's custom-chip portfolio, with a target deployment date set for 2028. This initiative aims to alleviate a significant internal compute shortage that has reportedly led Google Cloud to decline external business opportunities.
Recently, Google committed nearly $1 billion per month to SpaceX to help meet its computing obligations. The development of Frozen v2 comes at a time when Google faces increased competition in the AI space, particularly from Chinese models that now represent 45% of token usage among U.S. companies.
Additionally, Google is experiencing delays with its next Gemini Pro release and has lost key researchers to competitors. The competitive landscape is intensifying, highlighted by recent advancements from companies like Moonshot AI and Alibaba.
Furthermore, Google DeepMind's chief, Demis Hassabis, is advocating for a regulatory framework for AI on Capitol Hill, emphasizing the need for oversight of advanced AI models to mitigate national security risks