OpenAI to launch its own AI chip in 2025, reducing reliance on Nvidia

OpenAI is accelerating efforts to reduce dependence on Nvidia's AI accelerators. The company plans to finalize chip development and begin production at TSMC using advanced manufacturing processes in the coming months. Analysts predict mass production by 2026, with the tape-out phase—the transition from design to manufacturing—expected soon. This stage typically costs tens of millions and can take up to six months.

The new AI accelerator is seen as a strategic tool to strengthen OpenAI's bargaining power with chip suppliers. If successful, OpenAI could offer an alternative to Nvidia, which currently holds over 80% of the AI accelerator market. Future iterations will bring more advanced processors, aligning with OpenAI’s participation in the $500 billion Stargate infrastructure program announced last month by U.S. President Donald Trump.

The project is led by Richard Ho, a former Google AI chip specialist, in collaboration with Broadcom. While OpenAI's 40-person chip team remains small compared to Google or Amazon, scaling will require hiring hundreds of engineers. Initial impact on OpenAI’s infrastructure is expected to be limited, but a full-scale AI chip program could cost $500 million, doubling with software and hardware integration. For comparison, Meta plans to invest $60 billion in AI infrastructure in 2025, while Microsoft's annual AI investment is projected at $80 billion.