OpenAI's First In-House Chip Will Be Developed By TSMC On Its A16 Angstrom Process
TSMC has not yet started mass production of its A16 Angstrom process, but there is already buzz about its capabilities, especially since Apple has reportedly placed orders with the foundry. In an interesting turn of events, OpenAI is also working on a custom chip using the same technology to enhance Sora’s video-generation abilities.
Custom Chip Development in the AI Race
The competition in the AI sector has pushed major tech players to explore custom solutions to stay ahead in the game. For example, ByteDance, the parent company of TikTok, recently announced a partnership with U.S. firm Broadcom to develop an in-house AI chip on TSMC’s 5nm process. Similarly, OpenAI is aiming for the stars, with reports from the Economic Daily News suggesting discussions with TSMC for a new fabrication plant dedicated to this project.
Collaboration Speculations and Production Timeline
While speculations suggest a close collaboration between OpenAI and Apple, known for its superior chip development, no official confirmation has been made. Nevertheless, it will be some time before the chip goes into mass production, as TSMC’s A16 Angstrom process is projected to start mass manufacturing in 2026.
The wait for the A16 chip seems promising, as it offers speed enhancements of 8-10% over the N2P process at the same voltage levels. Additionally, it boasts a 20% reduction in power consumption with sustained speeds and a remarkable 110% increase in density, making it a suitable choice for server applications.
Source: Economic Daily News
