Tech Desk 
Feb 10, 2025
OpenAI is reportedly developing its first in-house artificial intelligence (AI) chip, planning to reduce reliance on Nvidia. The ChatGPT manufacturer is planning to send it for fabrication at TMSC in the near future, aiming for large-scale production in 2026.
This step marks a major leap forward in OpenAI’s efforts to secure enhanced control over its AI infrastructure.
The first-generation chip is designed mainly for running AI models and is being developed by OpenAI’s in-house team, led by Richard Ho, in partnership with Broadcom.
However, while the team has been expanded to 40 members, it still remains smaller in contrast to the AI chip divisions of Alphabet-owned Google and Amazon. If the initial fabrication procedure, called “taping out,” is successful, OpenAI is likely to start testing its alternative to Nvidia’s chips by the end of this year.
Developing a custom AI chip is an extravagant endeavour. Industry estimates suggested that a single variant of the latest AI chip is likely to cost $500 million, with costs doubling when accounting for software and infrastructure.
However, increasing reliance on Nvidia, which holds an 80% market share, has pushed leading players, including OpenAI, Microsoft, and Meta to explore alternatives.
TSMC is set to manufacture OpenAI’s chip using its advanced 3-nanometre process. This chip will feature a high-bandwidth memory (HBM), systolic array architecture, and more, that closely aligns with Nvidia’s AI chipsets. OpenAI's chip will be launched on a particular scale, marking a pivotal step toward self-sufficiency in AI hardware.
With a rapidly growing demand for AI chips, OpenAI’s in-house chip development represents a strategic shift, reinforcing its bargaining power with suppliers while minimizing long-term costs and dependence on third-party manufacturers, such as Nvidia.
