Open AI to partner with TSMC to make custom chips?
Introduction
OpenAI is making significant strides in developing its first custom AI chip, with plans to finalize the design in the coming months. This ambitious project aims to reduce the company’s reliance on Nvidia, which currently dominates the AI chip market with an 80% share and also compete with Nvidia
Design and Manufacturing Partnership
The ChatGPT maker is collaborating with Taiwan Semiconductor Manufacturing Company (TSMC) to fabricate its inaugural in-house chip. TSMC will manufacture the chip using its advanced 3-nanometer process technology, ensuring cutting-edge performance and efficiency.
Key aspects of the chip design include
High-bandwidth memory (HBM)
Systolic array architecture optimized for AI tasks
Extensive networking capabilities
The chip development is led by Richard Ho, a former Google engineering lead who now heads OpenAI’s 40-person in-house team.
Timeline and Production Goals
OpenAI’s timeline for chip development is ambitious yet structured:
Early 2025
Finalization of chip design
Mid-2025
Tape-out process at TSMC
2026
Target for mass production
The initial phase, known as “taping out,” is a critical and costly step, typically requiring tens of millions of dollars and about six months to produce a prototype.
Strategic Implications
This move into custom chip design carries significant strategic importance for OpenAI:
Reduced dependence on Nvidia
Enhanced negotiating power with chip suppliers
Greater control over AI infrastructure and costs
Initially, the chip will have a limited role within OpenAI’s infrastructure, primarily focused on running AI models. However, future iterations are expected to have broader capabilities, including AI model training.
Challenges and Investments
Developing a custom AI chip is a complex and expensive endeavor. The total cost, including software and peripherals, could approach $1 billion. OpenAI faces several challenges:
Ensuring the chip functions correctly on the first tape-out
Expanding the engineering team to support large-scale production
Building a reliable software ecosystem to complement the hardware
Conclusion
Despite these challenges, OpenAI’s investment in custom chip development underscores its commitment to technological independence and innovation in AI hardware.