OpenAI Gandeng Broadcom And TSMC To Build The First AI Chip

JAKARTA OpenAI, the artificial intelligence technology company behind ChatGPT, is currently designing its first internal chip designed to support its artificial intelligence system.

In developing this chip, OpenAI collaborates with Broadcom and Taiwanese well-known semiconductor producer TSMC. This step also marks OpenAI's efforts to reduce dependence on Nvidia chips that have dominated the AI computing market.

A trusted source quoted by Reuters said that OpenAI had considered building its own network of chip factories or "foundrys", but canceled the plan due to the high cost and time it needed.

Instead, OpenAI is now focusing on internal chip designs supported by Broadcom and TSMC, which are scheduled to start production in 2026. However, this schedule could experience changes according to development.

Broadcom plays an important role in helping OpenAI improve chip designs while securing production capacity through TSMC. OpenAI's first chip will be focused on inference or the application of artificial intelligence in making predictions and decisions based on new data.

Currently, the biggest demand is still on chips for AI model training. However, analysts predict the need for chips for inference will soon increase as AI applications develop.

Broadcom has extensive experience in helping big companies like Google improve chip designs for manufacturing. In a complex AI system, where thousands of chips work together, Broadcom technology helps ensure fast and efficient information flows among these chips.

In addition to developing internal chips, OpenAI is also expanding its chip sources using the planned MI300X AMD chip to be used on Microsoft's Azure platform.

This shows that OpenAI seeks to diversify its chip suppliers, reducing Nvidia's dominance, which currently holds more than 80% of the AI chip market share. The decision was also driven by an increase in chip prices and a limited supply of Nvidia GPU in the market.

According to estimates, AMD's AI chip sales are expected to reach 4.5 billion US dollars by 2024, supported by the launch of MI300X in the last quarter of 2023. On the other hand, OpenAI is projected to suffer a loss of 5 billion US dollars this year with revenue of 3.7 billion US dollars.

Computing costs such as hardware, electricity, and cloud services are the largest spending for this fast-growing company. This prompted OpenAI to optimize the use of resources and diversify chip suppliers to reduce costs.

Strategy For Long-Term

With the diversification of this chip supplier, OpenAI shows its seriousness in building more efficient and low-cost AI infrastructure. However, OpenAI maintains good relations with Nvidia.

The company is avoiding massive signings from Nvidia to avoid conflict, considering they still need access to Nvidia's latest-generation chip technology, such as the Blackwell chip.

This OpenAI move has the potential to shake up the semiconductor industry, especially for other major tech companies such as Amazon, Meta, and Google, which are also considering similar strategies. With Broadcom and TSMC partnerships as well as the use of AMD chips, OpenAI appears to be working on creating a more diverse, flexible, and cost-effective chip ecosystem.

Nvidia, AMD, and TSMC have yet to provide official comments regarding this development, while Broadcom did not immediately respond to requests for comment from the media.