OpenAI is partnering with Broadcom, TSMC and AMD to develop its first custom AI chip, aiming to diversify its chip supply and manage rising infrastructure costs as it navigates high demand for AI computing power. (Source: Image by RR)

OpenAI Prepares for High Demand With Diverse Chip Suppliers and Custom Designs

OpenAI is collaborating with Broadcom and TSMC to develop its first in-house AI chip to meet growing infrastructure demands, while also incorporating AMD chips alongside its primary supplier, Nvidia. Faced with the high costs of building its own manufacturing network, OpenAI has set aside plans for dedicated foundries and instead aims to design custom chips to better manage costs and diversify its chip supply, a strategy similar to larger tech companies like Amazon and Google. Broadcom’s stock rose following this news, as the company assists OpenAI in refining chip designs and securing production capacity through TSMC for 2026, with a focus on inference chips that could eventually surpass demand for training chips as AI applications expand.

OpenAI’s plans to incorporate AMD’s new MI300X chips into its systems via Microsoft’s Azure also indicate its strategy to explore alternative suppliers as Nvidia grapples with high demand, shortages, and increasing costs. As reported in reuters.com, Nvidia continues to hold over 80% of the market share in GPUs, but OpenAI, along with other tech giants, is seeking options to mitigate reliance on Nvidia and manage supply chain challenges. This expansion of chip sources highlights OpenAI’s efforts to optimize costs as it faces significant expenses, including a projected $5 billion loss this year, largely due to the compute costs required for AI training and operations.

OpenAI’s chip design team includes seasoned engineers from Google who previously worked on Tensor Processing Units (TPUs), showing the company’s commitment to building a robust in-house team capable of advancing chip development. While OpenAI is venturing into custom chip design, it remains cautious about maintaining a positive relationship with Nvidia, which continues to provide essential hardware for its AI projects, particularly Nvidia’s upcoming Blackwell chips.

Ultimately, OpenAI’s approach to balancing in-house chip design with external partnerships demonstrates its ambition to control more of its critical infrastructure, reduce dependency on any single supplier, and enhance its ability to scale AI applications effectively. As OpenAI looks to maintain a good relationship with Nvidia for essential hardware while exploring custom and alternative solutions, this multi-faceted chip strategy positions the company to remain agile and competitive in a fast-evolving industry, setting a potential trend for other AI-driven companies aiming to optimize costs and reliability.

read more at reuters.com