OpenAI, Broadcom Partner on Chip

OpenAI is collaborating with Broadcom to develop its own AI chip, slated for internal use in 2026. This move aims to optimise the performance of models like ChatGPT and reduce reliance on Nvidia. Broadcom will design and produce the AI accelerator. OpenAI will initially use the chip for its own operations.

Broadcom's CEO indicated over $10 billion in orders from a new customer, identified as OpenAI, for custom accelerator business. These accelerators are crucial for AI development across major tech firms. Broadcom's shares saw a surge, while Nvidia's experienced a dip amid concerns over increased competition.

The chip is expected to utilise a systolic array architecture, similar to other AI accelerators. Broadcom is set to deliver custom AI accelerators and networking chips, alongside reference AI rack platforms. This partnership signifies a strategic shift as tech giants seek diverse options for AI infrastructure.