Archive — CN — technode
OpenAI’s first AI chip to tape out at TSMC in first half of the year: report
OpenAI plans to tape out its first in-house AI chip at TSMC in H1 2024 to reduce reliance on NVIDIA. This training-focused chip aims to enhance bargaining power, with mass production targeted by 2026. The tape-out costs millions and takes six months, but production success is not guaranteed, requiring potential rework if issues arise.
Key points
- OpenAI is developing its first in-house AI chip.
- The tape-out is expected at TSMC in the first half of 2024.
- The chip aims to lessen dependence on NVIDIA and other suppliers.
- OpenAI views the chip as a strategic tool for better negotiations with suppliers.
- Mass production is planned by 2026 after successful tape-out.
- Tape-out costs tens of millions and takes approximately six months.
- If the first chip attempt fails, it will need a re-diagnosis and restart.
Why it matters
- As of 2023, the AI chip market is competitive, with major companies investing heavily in AI hardware.
- OpenAI's initiative is part of a broader trend where companies are seeking independence from dominant suppliers like NVIDIA.
- By 2025, trends in the semiconductor industry show a push towards custom AI chips to enhance computational efficiency and performance.
OpenAI's move to develop its own AI chip signifies a critical step towards technological independence and competitiveness in the evolving AI landscape, aiming for better control over its hardware resources.