OpenAI teams up with Broadcom for massive 10GW AI infrastructure push

OpenAI has entered into a significant multiyear partnership with Broadcom, aiming to develop custom chips and networking technologies as part of a massive 10-gigawatt AI infrastructure initiative. This collaboration will see OpenAI design advanced AI accelerators that will be integrated into data centers in cooperation with Broadcom. The deployment of these AI accelerator racks is scheduled to commence in the latter half of 2026, with a target completion date set for the end of 2029. These new systems, which will utilize Broadcom's Ethernet and connectivity solutions, are intended to address the surging global demand for AI technologies. OpenAI's approach of creating its own hardware is designed to incorporate insights gained from its development of cutting-edge AI models and products, ultimately enhancing capabilities and intelligence levels. Sam Altman, CEO of OpenAI, emphasized the importance of this partnership, stating that it is a crucial move towards building the infrastructure necessary to unlock the full potential of AI for the benefit of individuals and businesses alike. Broadcom’s President and CEO, Hock Tan, expressed enthusiasm about the partnership, highlighting OpenAI’s leadership in the AI landscape since the rise of ChatGPT. The collaboration aims to co-develop and deploy innovative AI accelerators and network systems, setting the stage for the future of artificial intelligence. OpenAI, which has seen its user base soar to over 800 million active users weekly, has been actively securing several computing deals to support its growing AI needs. Recently, Nvidia announced a substantial investment of up to $100 billion in OpenAI for AI data center infrastructure, which includes the deployment of Nvidia’s forthcoming Vera Rubin chips starting in 2026. Additionally, OpenAI has forged a multi-year agreement with AMD to implement billions of dollars' worth of AMD GPUs, with the first phase involving six gigawatts of AMD technology expected to roll out in the latter half of 2026.