Facing increasing pressure from the shortage of AI computing power, Google is secretly developing a new specialized server chip codenamed Frozen v2. The core breakthrough of this chip lies in directly embedding the underlying computing logic of the Gemini large model into the hardware, thereby significantly improving the chip's computational efficiency.

Revamping the Architecture to Reshape the Computing Ecosystem

According to the research team's calculations, the processing performance per unit of power after the deployment of Frozen v2 will reach 6 to 10 times that of Google's latest self-developed chip. This design, which directly etches software algorithms into silicon, can significantly reduce data movement and logical judgment steps, achieving a qualitative improvement in energy efficiency.

Different from general-purpose chips such as NVIDIA GPUs or Google TPUs, this specialized chip is specifically designed for the Gemini architecture as an integrated hardware-software solution. It not only greatly shortens AI response time but also supports more complex new application scenarios. It is expected to be deployed as early as 2028.

Positioned as an Experimental Platform, Not a Replacement for TPU

Although it demonstrates strong performance, Google does not plan to fully replace existing TPU chips with Frozen chips. Instead, the two will complement each other in the future. Due to the embedded algorithm logic, this chip requires the Gemini model to continue using a specific basic architecture, thus having a certain level of technical specialization.

Currently, Google positions it as a technology testing platform, used to accumulate engineering experience for the next generation of more specialized AI chips. Due to the limited initial production scale, the chip will mainly serve internal specific needs and will not be widely launched on the market at this stage.