Alphabet Inc. (NASDAQ: GOOGL) shares rose 1.15% Monday morning following a report that the company is taking an audacious new approach to custom hardware: etching its flagship AI model directly onto its next-generation chips.

According to The Information, Google’s experimental project—dubbed "Frozen v2"—is designed to bypass the traditional processing overhead of general-purpose GPUs and TPUs. By burning Gemini’s blueprint into the silicon itself, the chip slashes processing steps and data movement.

Key Highlights

• 
  

  Projected to be 6 to 10 times more power-efficient per token than current TPUs.

  
  


• 
  

  Built to solve compute shortages that currently bottleneck internal teams and force Google Cloud to turn down external deals.

  
  


• 
  

  Planned for deployment as early as 2028 as a parallel branch to the TPU product line.

  
  


• 
  

  The architecture locks the hardware into Google’s current AI design framework, though engineers can still update parameters via new model weights.

  
  

In response, a Google spokesperson noted that while teams actively experiment with high-efficiency innovations, not every lab project reaches commercial production.