alphabet Shares rose 3% on Monday after The Information reported that the company is internally developing a new server chip, dubbed “Frozen v2,” designed to run the Gemini model more efficiently.
The chip will permanently embed parts of Gemini’s architecture directly into silicon, reducing the number of computations and data movement needed to answer queries, the report said.
Alphabet told CNBC in a statement that its teams are “constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers.” “While not every project will make it to production, this rigorous exploration is central to our full-stack approach,” Alphabet told CNBC in a statement.
“Co-designing the hardware and software from the ground up ensures that our systems are integrated and highly optimized for real-world workloads,” the statement continued.
Google engineers predict that it will be able to deliver six to 10 times more tokens per unit of power than the company’s latest AI chips, called TPUs (tensor processing units), The Information reported. Rather than replacing general-purpose TPUs, Frozen will become a more specialized branch of Google’s custom chip portfolio.
According to the report, the company is targeting implementation in 2028. The project is aimed at alleviating a significant internal computing shortage that has reportedly fueled tensions and forced Google Cloud to turn away outside business.
Just last month, Google agreed to pay space x Nearly $1 billion is disbursed every month to close the gap and meet enterprise computing commitments.
The tradeoff is flexibility. According to The Information, the chip will only work in future Gemini models if Google sticks to the same underlying architecture. Google reportedly currently considers Frozen v2 to be partially experimental and has no plans to produce it at the same scale as the TPU.
Google’s AI efforts face more pressing challenges.
The release of the next Gemini Pro has been delayed and Google has lost several senior researchers to rivals as Chinese models establish themselves in American companies. These models currently account for 45% of US corporate token usage.
With recent new releases from Moonshot AI, the competitive pressure will only increase. alibaba The ability gap will close this weekend.
Demis Hassabis, the head of Google DeepMind, is on Capitol Hill this week pitching lawmakers on a FINRA-style AI oversight body overseen by the federal government, largely funded by industry, and built to test cutting-edge models against national security risks before release.
Read the full text of The Information here.
Google stock price chart.
