OpenAI and semiconductor giant Broadcom have announced a multi-year partnership to co-develop and deploy custom AI accelerators and networking systems. The ambitious deal, targeting the deployment of 10 gigawatts (GW) of custom-built hardware between 2026 and 2029, marks a strategic shift for OpenAI to secure its own computing power and reduce dependence on third-party suppliers like NVIDIA.
The collaboration outlines a clear division of labor: OpenAI will design the AI accelerators and system architecture, leveraging its deep understanding of model requirements to create optimized hardware. In turn, Broadcom will handle the development, manufacturing, and deployment of the custom chips and provide its extensive suite of networking solutions, including Ethernet, PCIe, and optical connectivity.
This vertical integration is a growing trend among major technology firms, who seek to gain more control over their AI infrastructure, reduce supply chain risks, and optimize performance for their specific workloads. For OpenAI, the move is a direct response to the supply bottlenecks and escalating costs associated with relying solely on standard, off-the-shelf GPUs.
The sheer scale of the 10 GW deployment is a powerful statement of OpenAI’s intent to scale its computing power to industrial levels. To put this in perspective, 10 GW is equivalent to the power consumption of millions of US households, rivaling the energy needs of a major city. This monumental commitment highlights the massive computational demands required to train and deploy the next generation of advanced AI models.
For Broadcom, the deal provides a major entry into the high-stakes custom AI chip market. The deal puts the chipmaker closer to the core of AI infrastructure than its traditional role as a networking component supplier.
Broadcom’s stock surged[1] significantly on the news, reflecting strong investor optimism about the partnership’s revenue potential.
References
- ^ Broadcom’s stock surged (www.investopedia.com)