OpenAI Unveils First Custom Chip Designed to Optimize ChatGPT Performance

OpenAI announced its first custom-designed chip in June 2026, a move aimed at reducing the company's dependence on Nvidia hardware and cutting the cost of running ChatGPT.

The chip, internally called Jalapeno, was developed in partnership with Broadcom and is being fabricated by TSMC. OpenAI said the chip is designed specifically to optimize inference, the process of generating responses from a trained AI model, rather than training new models from scratch.

The announcement follows a broader industry trend. Google has been using its own Tensor Processing Units for years, and Amazon has developed Trainium and Inferentia chips for its cloud customers. Microsoft, which has a major investment in OpenAI, has also been developing custom AI silicon.

OpenAI's move into custom hardware reflects the enormous cost of running large language models at scale. The company processes billions of queries through ChatGPT each day, and even small improvements in chip efficiency translate into significant savings.

Anthropic is also reportedly in discussions with Samsung about developing a custom AI chip, suggesting that the major AI labs are all moving toward proprietary hardware.

The Jalapeno chip is not expected to replace Nvidia GPUs entirely. OpenAI will likely continue using Nvidia hardware for model training, where Nvidia's chips remain the industry standard. The custom chip is targeted at the inference workload, where the economics of proprietary silicon are more favorable.

OpenAI has not disclosed a timeline for when the chip will be in full production use.