OpenAI Develops First In-House Chips to Reduce Dependence on Nvidia

OpenAI is advancing its efforts to lessen its reliance on Nvidia by creating its own line of artificial intelligence (AI) chips. This marks a significant move for the ChatGPT developer as it aims to establish more control over its computing resources.

The company is in the process of finalizing the design for its initial custom chip, which is scheduled to be completed in the coming months. Following this step, OpenAI intends to send the design for production to the Taiwan Semiconductor Manufacturing Company (TSMC). This procedure, known in the industry as “taping out,” is a crucial phase where the chip design is officially submitted for manufacture.

While OpenAI and TSMC did not provide comments on this development, sources indicated to Reuters that the technology company is on track to achieve its ambitious goal of mass production at TSMC by 2026. Typically, the tape-out process can be expensive, costing tens of millions of dollars, and the initial chip may take around six months to manufacture. OpenAI, however, has the option to expedite this process at a greater expense.

There remains an element of uncertainty as well; there is no guarantee that the silicon chip will perform as expected during its first manufacturing round. If there are issues, OpenAI will need to troubleshoot and repeat the tape-out process, which can further prolong production.

Within OpenAI, the custom chip is seen as a strategic asset that can help improve the company's negotiating position with existing chip suppliers. The plan includes developing increasingly advanced processors over time, aiming to expand their capabilities with each new version.

OpenAI, Nvidia, TSMC