OpenAI is progressing with its strategy to reduce its dependence on Nvidia for chip supply by creating its initial generation of in-house artificial intelligence silicon. The developer of ChatGPT is completing the design of its initial in-house chip within the upcoming months and intends to send it for production at Taiwan Semiconductor Manufacturing Co.

The update indicates that OpenAI is poised to achieve its ambitious aim of large-scale production at TSMC by 2026. An average tape-out expenses tens of millions of dollars and takes about six months to create a finished chip, unless OpenAI invests significantly more for quicker production.

Within OpenAI, the chip designed for training is considered a strategic tool to enhance OpenAI's negotiating leverage with various chip manufacturers. Following the first chip, OpenAI's engineers intend to create progressively more sophisticated processors that offer wider functionalities with each subsequent version.

If the first tape out is successful, it will allow the ChatGPT creator to mass-manufacture its first in-house AI chip and possibly experiment with an alternative to Nvidia's chips later this year. According to industry sources familiar with chip design budgets, a fresh chip design for an extensive program might require $500 million for just one version of a chip. These expenses might increase double to develop the essential software and related hardware.

OpenAI's in-house AI chip, although designed for both training and executing AI models, will initially deployed on a small scale, mainly for running AI models. The chip will play a restricted part within the firm’s infrastructure.