OpenAI Wants to Make Its Own AI Chips By 2026, With Help From AMD, Broadcom, Or TSMC

According to media reports, OpenAI is undergoing a hardware strategy adjustment aimed at optimizing computing resources and reducing costs: On the one hand, besides NVIDIA's GPUs, OpenAI will also give AMD's MI300 series chips a try. On the other hand, OpenAI is collaborating with Broadcom Inc. and TSMC to plan the production of custom AI chips starting in 2026.

The report mentioned that OpenAI has assembled a chip development team of about 20 engineers, including senior engineers who participated in Google's Tensor processor project. However, according to OpenAI's current timetable, the actual production of custom chips is not expected to be realized until 2026.

Previously, OpenAI mainly relied on NVIDIA's GPUs for model training and inference. Currently, NVIDIA's GPUs hold over 80% of the market share, but issues such as chip shortages, insufficient supply of NVIDIA AI compute cards, and rising costs have led OpenAI to seek alternative solutions.

By introducing AMD's MI300 series chips, OpenAI can not only ensure high-performance computing but also diversify supply risks.

In addition to bringing in AMD chips, OpenAI is working with Broadcom to develop new custom chips designed to handle the large loads used for AI inference and is collaborating with TSMC to ensure chip manufacturing capabilities.

OpenAI originally planned to establish a network of chip manufacturing factories, but due to cost and time constraints, this plan has been temporarily shelved. Currently, OpenAI will focus on internal chip design, partnering with industry players like Broadcom and TSMC to ensure the stability of chip supply.

Training AI models like ChatGPT is costly. According to previous reports, OpenAI is expected to lose $5 billion this year, with revenues of $3.7 billion. Computational costs, which include the hardware, electricity, and cloud services needed to process large datasets and develop models, are the company's largest expense.

NVIDIA's GPUs currently hold over 80% of the hardware market share. The shortage and rising prices of chips have prompted OpenAI to explore internal development or external alternatives for AI hardware. This strategy is similar to that of tech giants like Amazon, Meta, Google, and Microsoft, which aim to reduce costs and ensure access to AI hardware through custom chips. However, Google, Microsoft, and Amazon are several stages ahead in this area, and OpenAI may need substantial funding to become truly competitive.

In October of this year, OpenAI just completed a new round of financing of $6.6 billion, led by Thrive Capital, with participation from Microsoft, NVIDIA, Altimeter Capital, Fidelity Funds, SoftBank, and the Abu Dhabi state-backed MGX investment company, valuing the company at $157 billion post-investment. OpenAI hopes to use these funds to strengthen computational resources and continue expanding its capabilities in AI tools and technological research.