News
Broadcom custom XPUs are designed for AI infrastructure
Tuesday, March 10, 2026 at 11:01 AM
Broadcom has highlighted its custom XPU (accelerator) solutions specifically engineered to support large-scale AI infrastructure, focusing on high-performance compute and networking integration.
Context
Broadcom has significantly expanded its role in the AI supply chain by detailing its roadmap for custom XPUs (accelerators) and next-generation infrastructure. A cornerstone of this strategy is a multi-year collaboration with OpenAI to co-develop and deploy 10 gigawatts of custom-designed AI accelerators and network systems. This partnership involves the deployment of full AI racks scaled with Broadcom’s proprietary Ethernet solutions, with volume shipments targeted to begin in the second half of 2026 and reach full scale by 2029.
To power these highly complex systems, Broadcom is leveraging its 3.5D XDSiP packaging technology, which integrates logic, memory, and high-speed SerDes into a single package. This platform reportedly delivers a 10x reduction in power consumption and a 7x increase in signal density compared to traditional methods. By providing both the silicon and the rack-level networking, Broadcom aims to serve as a primary counterbalance to NVIDIA, already supporting five major custom chip partners with OpenAI slated as its sixth in 2027.
Sources (7)
AI Clusters | AI Servers | AI Infrastructure - BroadcomOpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators | OpenAIOpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators - Investor CenterBroadcom says AI companies can't make their own silicon - The RegisterBroadcom says five AI chip partners are ramping, OpenAI slated for 2027Broadcom May Become The Biggest Counterbalance To Nvidia - The Next PlatformBroadcom just fired the starting gun in the race to 2nm AI chips - SDxCentral
Related Companies
Broadcom
AVGO