OpenAI is collaborating with Broadcom and Taiwan Semiconductor Manufacturing Company (TSMC) to develop its first in-house chip for supporting its artificial intelligence (AI) systems. This move is part of OpenAI's strategy to diversify chip supply and reduce costs. The company is also planning to incorporate AMD chips alongside its existing NVIDIA chips to meet the rising demand for computational power.
Initially considering building its own chip factory, OpenAI has shifted focus to in-house chip design due to cost and time constraints. This decision has positively impacted the stock prices of TSMC, Broadcom, and AMD in the U.S. market.
In-house chips, designed and developed internally by a company, aim to enhance performance, optimize power consumption, and reduce costs by creating tailored solutions to meet specific business needs. However, the process involves significant funding and technical investment, requiring years of design and testing.
OpenAI has been collaborating with Broadcom for several months on an AI chip focused on inference. While training chips are currently in demand, analysts predict a growing need for inference chips as AI applications expand.
Broadcom, with its extensive experience in custom chip solutions and partnerships with tech giants like Google and Meta, has seen its revenues rise significantly. Some analysts suggest Broadcom is emerging as a key player in the custom chip market, alongside NVIDIA.
OpenAI has assembled a team of about 20 experts, including former Google engineers who built Tensor Processing Units (TPUs), to spearhead the chip development. The team is led by engineers like Thomas Norrie and Richard Ho.
OpenAI has reportedly secured manufacturing capabilities with TSMC through its collaboration with Broadcom, aiming for their first custom chips to be produced by 2026, although the timeline may adjust. The company is also considering further partnerships to enhance or acquire additional chip design elements.
NVIDIA's GPUs dominate over 80% of the market share, but high costs and supply shortages are prompting major clients like Microsoft and Meta to explore alternatives. OpenAI plans to use AMD chips through Microsoft’s Azure cloud services.
OpenAI is cautious about recruiting talent from NVIDIA, as it seeks to maintain a good relationship with the chipmaker, especially with NVIDIA’s Blackwell chips soon to be released.
