Summary:
-
New Chip Development
OpenAI is collaborating with Broadcom and TSMC to design its first in-house AI chip, shifting focus from ambitious foundry plans due to cost and time constraints. -
Focus on Inference
The new chip is primarily aimed at inference tasks, responding to a growing demand as more AI applications are deployed, despite current higher demand for training chips. -
Broadcomβs Role
Broadcom will assist in fine-tuning chip designs and provide components to facilitate rapid data transfer on and off the chips, crucial for efficient AI operations. -
Chip Team Formation
OpenAI has formed a dedicated chip team of about 20 engineers, led by former developers of Googleβs Tensor Processing Units (TPUs), to drive the project forward. -
Manufacturing Timeline
OpenAI has secured manufacturing capacity with TSMC for the custom chip, expected to be ready by 2026, although the timeline may adjust based on development progress.
Read more at: Reuters
!