OpenAI, a leading AI model developer, has introduced a new inference chip designed by Broadcom and manufactured by TSMC. This development indicates OpenAI's strategy to diversify its hardware supply chain for running AI models, moving away from a sole reliance on existing chip providers. This custom silicon is specifically tailored for the computational demands of AI inference.
This move is significant because it reflects a growing trend among major AI developers to create their own specialized hardware. By doing so, companies like OpenAI aim to optimize performance, reduce costs associated with AI model operations (capex), and potentially gain more control over their technological infrastructure, which is crucial for scaling AI services.
The mechanism involves OpenAI collaborating with chip designers (Broadcom) and fabricators (TSMC) to produce application-specific integrated circuits (ASICs) optimized for AI inference workloads. These custom chips are designed to efficiently process data when an AI model makes predictions or decisions, which is a different task than the training phase that typically requires more general-purpose GPUs.
This initiative could impact established semiconductor companies like Nvidia (NVDA), which currently dominates the AI chip market, by introducing more competition and potentially shifting demand towards custom solutions. It also highlights the growing importance of foundries like TSMC (TSM) and chip design firms like Broadcom (AVGO) in the custom AI silicon ecosystem, as AI model developers seek diverse supply options.
An AI breakdown of exactly what changed and who it moves.