OpenAI, the company behind ChatGPT, has unveiled a custom AI chip design developed in partnership with semiconductor manufacturer Broadcom, according to The New York Times. The announcement marks a significant move by OpenAI to develop its own specialized hardware for artificial intelligence workloads.
According to the report, OpenAI plans to deploy enough of these custom chips to consume approximately 10 gigawatts of electricity. The New York Times notes this amount of power could be sufficient to supply millions of households, highlighting the massive scale of computing infrastructure OpenAI is targeting for its AI operations.
The collaboration with Broadcom represents OpenAI’s effort to secure dedicated chip supply and potentially reduce reliance on existing GPU manufacturers. Custom chip designs can be optimized specifically for the training and inference tasks required by large language models like ChatGPT, though the timeline for production and deployment was not specified in the available reporting.