OpenAI Designs Custom Chips

OpenAI is moving towards designing its own AI chips to reduce reliance on Nvidia and control costs. The company has partnered with Broadcom to develop custom AI accelerators and network systems, with deployment planned for the second half of next year. This collaboration aims to build a 10-gigawatt data centre to support OpenAI's AI models.

CEO Sam Altman says that developing custom chips is crucial for building the infrastructure needed to unlock AI's potential. The company will design the chips, and Broadcom will handle development, fabrication, and deployment. This move follows OpenAI's agreement with AMD for AI chip supply, signalling a broader industry trend to diversify AI chip sources.

Analysts suggest that securing 1 GW of AI computing capacity requires approximately $35 billion in chip costs. The OpenAI initiative also includes investments from Softbank, Oracle, Nvidia, AMD, MGX, Coreweave, Microsoft and ARM. These infrastructure investments could significantly impact the stock market as AI technology potentially revolutionises various business sectors.