OpenAI Partners with Broadcom and TSMC to Develop Custom AI Chip

The AI leader is scaling back plans for its own chip factories, opting instead to design in-house chips with industry partners to manage costs and supply challenges.

Overview

OpenAI is collaborating with Broadcom and TSMC to create its first in-house AI chip, focusing on inference tasks rather than model training.
The company has decided against building its own chip manufacturing foundries due to high costs and long timelines, choosing to partner with established manufacturers instead.
OpenAI plans to introduce AMD chips into its infrastructure via Microsoft's Azure, diversifying its reliance on Nvidia's GPUs.
With a team of about 20 engineers, OpenAI aims to launch its custom AI chip by 2026, though this timeline remains flexible.
The strategic move to develop custom silicon is driven by the need to reduce operational costs and secure a stable chip supply amid growing AI demands.