OpenAI's first in-house AI chip is set to advance with exciting new developments. Reports indicate that TSMC will be creating a state-of-the-art A16 chip using a cutting-edge 3nm process technology tailored specifically for OpenAI's Sora video model. This chip aims to significantly enhance Sora's video generation capabilities.
The A16 represents TSMC's most advanced process node disclosed to date and marks their debut into the 3nm process technology. Mass production of the A16 is projected to begin in the latter half of 2026.
The A16 will feature next-generation nanosheet transistor technology and employ Super Electrical Rail (SPR) technology—a pioneering solution in the industry. SPR technology, which introduces a novel backside power delivery approach, relocates the power network to the wafer's rear. This innovation frees up more space on the front of the wafer, significantly increasing logic density and performance, making the A16 ideal for high-performance computing (HPC) applications that demand intricate signal routing and dense power networks.
Compared to the N2P process, the A16 offers a 1.10x increase in chip density. It delivers an 8% to 10% speed improvement at the same operating voltage and reduces power consumption by 15% to 20% for the same speed.
While TSMC has yet to commence mass production of the A16, early orders have already been placed. Notably, Apple has secured initial production capacity, and OpenAI has also reserved A16 capacity to support its in-house AI chip development needs.