According to media reports, insiders revealed that OpenAI is collaborating with $Broadcom (AVGO.US)$ and $Taiwan Semiconductor (TSM.US)$ to create its first "in-house" chip to support its artificial intelligence system.

The report mentioned that OpenAI has explored a range of options to diversify chip supply and reduce costs. Insiders mentioned that the company also plans to use Nvidia chips alongside $Advanced Micro Devices (AMD.US)$ Chips to meet its increasing demand for computing power.

Previously, OpenAI even considered establishing a chip foundry. However, according to sources, considering cost and time constraints, OpenAI has abandoned the foundry plan. As an alternative, the company plans to focus on in-house chip design work.

Affected by the news, during Tuesday's regular trading hours, the stock prices of Taiwan Semiconductor, Broadcom, and AMD all saw significant gains.

"In-house chip" refers to chips designed and developed internally by a company, rather than relying on external chip suppliers or third-party design companies. This practice often appears in some large technology companies with strong technical capabilities, such as Apple's M-series chips.

The purpose of developing in-house chips is to improve performance, optimize power consumption, reduce costs, or create differentiated products according to specific enterprise needs. However, there are also some challenges, such as the substantial financial and technological investment required for chip design and manufacturing. Chip development typically entails multiple years of design and testing cycles.

Sources revealed that to develop the first dedicated inference AI chip, OpenAI has been collaborating with Broadcom for several months. While there is currently a higher demand for training chips, analysts predict that with more AI applications being deployed, the demand for inference chips may exceed that of training chips.

Broadcom has cooperated with technology giants such as Google and Meta. With its deep accumulation in the field of custom chips, Broadcom's recent revenue has surged significantly. Some analysts believe that after Nvidia, Broadcom has quietly become another big winner in this field.

Reports suggest that OpenAI has formed a chip team of about 20 people, led by top engineers who have previously built tensor processing units (TPUs) at Google, including Thomas Norrie and Richard Ho.

Sources say that OpenAI has confirmed manufacturing capabilities through Broadcom and Taiwan Semiconductor, with plans to manufacture its first custom chip in 2026. However, they added that the timeline may change.

Two sources indicated that OpenAI is still deciding whether to develop or acquire other elements of chip design, and may introduce more partners.

Currently, Nvidia's GPUs hold over 80% of the market share, but supply shortages and high costs have led major customers like Microsoft and Meta to start exploring alternative solutions. OpenAI also plans to use AMD chips through Microsoft's Azure cloud computing service.

Insiders also added that OpenAI has been cautious about poaching talent from Nvidia because the company wishes to maintain a good relationship with Nvidia, especially as the Blackwell chip is about to be shipped.

Editor/new