The company has formed a team of around 20 engineers, spearheaded by former Google TPU engineers, who will collaborate with Broadcom and TSMC on an AI inference chip. OpenAI's chip will initially focus on inference, running pre-trained machine learning models on new data. While training AI demands immense computing resources, analysts believe the market for inference chips could surpass training chips in the coming years as more AI applications become mainstream.

Custom silicon production is not expected until 2026. Until then, OpenAI will rely on AMD's MI300X chips, housed in systems on Microsoft Azure. AMD aims to capture market share from Nvidia, which currently dominates the AI chip market with over 80 per cent share. However, shortages and rising costs have driven major customers, including Microsoft, Meta, and now OpenAI, to explore in-house or custom silicon solutions.

As demand for AI capabilities surges across industries, even the largest tech firms find it challenging to build sufficient resources internally. Through its partnership with Broadcom, OpenAI aims to access chip design and manufacturing expertise while reducing costs compared to independent development.

Other prominent AI players like Google, Microsoft, and Amazon have already designed multiple generations of customised chips, using their early forays into chip design. However, OpenAI believes that aligning with established entities like Broadcom will expedite scaling hardware tailored to its workloads. The company is likely to onboard more partners and is still contemplating whether to design or outsource other chip components.

This collaboration represents OpenAI's strategic maneuver to enhance its AI capabilities while navigating the complexities of custom silicon production.

In a statement OpenAI said that it will be adding more partners and is still considering whether it will design or outsource other chip components.