OpenAI is already in the final stages of developing its own chip for artificial intelligence and plans to start production with TSMC in the next few months. This was reported by Reuters.
The company plans to start mass production in 2026 and could hand over the final design to TSMC in the coming months. The first test production will last approximately six months and cost tens of millions of dollars, unless OpenAI decides to invest significantly more to speed up the process.
There is no guarantee that the first test chips will work properly, so the company will probably have to eliminate the defects and restart test production.
OpenAI sees its own chip as a strategic tool to strengthen its position in negotiations with chip suppliers. It will also help the company reduce its dependence on NVIDIA.
The chip is being developed by an internal team of about 40 people led by Richard Ho, head of the company's hardware division, in collaboration with Broadcom. Ho previously worked at Google, where he was also involved in the development of its own AI chip.
OpenAI plans to use its own chip for both training and running language models, but the main focus is on their execution. The chip will be deployed on a small scale and will play only a minor role in the company's infrastructure.
According to sources, the production of even one version of the chip could cost OpenAI more than $500 million. If we take into account the costs of software and peripherals, the total amount could double.