Broadcom Ships Industry-First 2nm 3.5D AI Chip

Edgen Stock·Feb 26 2026, 14:30

Share to

Share to

Copy link

Key Takeaways

Broadcom has begun shipping a breakthrough 2-nanometer custom AI chip, leveraging a novel 3.5D packaging technology to meet the intense demands of modern AI data centers. The move solidifies its leadership in the high-value custom accelerator market and sets a new benchmark for performance and efficiency in the semiconductor industry.

Industry-First Technology: On February 26, Broadcom announced shipments of the first 2-nanometer custom computing SoC built on its 3.5D eXtreme Dimension System in Package (XDSiP) platform.
Advanced Packaging: The new XDSiP platform combines 2.5D and face-to-face 3D stacking, enabling superior signal density and energy efficiency for next-generation AI processors (XPUs).
Targeting AI Clusters: This technology is designed to power large, gigawatt-scale AI clusters, strengthening Broadcom's competitive position in the lucrative custom AI hardware market.

Broadcom Ships First 2-Nanometer 3.5D AI Chip

Broadcom announced on February 26 it has started shipping the semiconductor industry's first 2-nanometer (2nm) custom computing System-on-a-Chip (SoC). This development is built on the company's new 3.5D eXtreme Dimension System in Package (XDSiP) platform, a modular, multi-dimensional stacked chip architecture. The technology represents a significant leap in semiconductor packaging by integrating established 2.5D techniques with advanced face-to-face (F2F) 3D-IC integration. This hybrid approach forms the foundation for what Broadcom terms the next generation of XPUs, or advanced processors.

New Platform Targets Gigawatt-Scale AI Clusters

The 3.5D XDSiP platform is engineered to address the massive computational requirements of modern artificial intelligence infrastructure, specifically targeting gigawatt-scale AI clusters. Broadcom stated the technology allows its consumer AI clients to deploy state-of-the-art XPUs with unmatched signal density, superior energy efficiency, and low latency. The modular design of the XDSiP platform enables compute, memory, and network I/O components to be scaled independently within a compact physical footprint. This capability is critical for building high-efficiency, low-power systems for large-scale computation, strengthening Broadcom's strategic position in the competitive custom AI accelerator market and challenging rivals to match its performance benchmarks.