OpenAI Expects First Custom AI Chip By End Of 2025

OpenAI plans to manufacture its first specialized chips for artificial intelligence models by the end of this year.

According to an exclusive report by Reuters, sources familiar with the matter said that the AI company is nearing the end of the design process and will soon begin “taping out”—the stage where a chip’s design is finalized and then sent to the semiconductor manufacturing facility for fabrication—within the next few months.

The anonymous sources explained that Taiwan Semiconductor Manufacturing Co (TSMC) will be in charge of the manufacturing process, as revealed in October last year. OpenAI and TSMC declined to comment on the topic.

This update confirms OpenAI’s intentions to reduce its dependence on Nvidia chips and move forward with its ambitious plans of reaching mass production by 2026.

The process of building AI chips takes approximately six months. However, OpenAI still needs to test the product, and tape-out process costs can increase if issues require diagnosis and a remake, or if they choose to accelerate production.

Richard Ho—Head of Hardware at OpenAI, previously at Google’s Tensor Processing Unit—, in collaboration with Broadcom, is in charge of the team designing the AI chips for AI, which has significantly increased its members in the past few months.

Producing AI chips in-house is seen as a strategic move to gain leverage in negotiations with other chipmakers, but it’s no easy task. Other tech giants, such as Meta and Microsoft, have attempted this but failed.

OpenAI, along with Oracle, Softbank, and the government of the United States recently announced a $500 billion joint venture to develop the Stargate Project in the American nation.