OpenAI has begun developing its own AI processor in collaboration with Broadcom and TSMC, according to Reuters, citing insider sources.

Exploring Processor Development

OpenAI is diversifying its chip supply chain and evaluating various strategies, including raising capital for costly manufacturing processes or creating its own chips. Due to the significant financial and time commitments involved in building foundries, the company is temporarily focusing on in-house processor design.

Several months ago, OpenAI partnered with Broadcom to design chips optimized for running pre-trained AI models. While the current demand for processors used in training neural networks is higher, journalists note that the need for inference chips will likely grow as new AI applications emerge.

Collaboration and Production Timeline

OpenAI is still considering whether to develop or acquire components for its future chip and may onboard additional partners. The project team, led by former senior Google developers Thomas Norri and Richard Ho, consists of approximately 20 experts.

Through its partnership with Broadcom, OpenAI gains access to TSMC’s manufacturing capabilities. TSMC is expected to produce the first chips for OpenAI by 2026.

Addressing Cost Challenges

The shift to proprietary AI chip development stems from the high costs associated with supporting AI operations, which have become a major expense for OpenAI. In 2024, the company is projected to face a potential loss of $5 billion, with development and training expenses reaching $7 billion. Despite these challenges, OpenAI raised $6.6 billion in October, valuing the company at $157 billion.