Marvell’s Acquisition of XConn Technologies

Marvell Technology, Inc. announced an agreement to acquire privately held XConn Technologies for approximately $540 million, in a deal expected to close in early 2026. The consideration will be a mix of cash and stock, with the stock portion expected to represent around 2.5 million shares of Marvell common stock (based on a 20-day VWAP).

Marvell’s stated objective with the acquisition is to expand its PCIe and CXL switching portfolio and to expand its UALink scale-up switch engineering team with experienced talent to play a major role in AI-driven data center interconnects. This comes at a time when data centers are evolving from single-rack deployments to rack-scale and multi-rack configurations to support large AI models and accelerator clusters. These environments place a premium on high-bandwidth, ultra-low-latency fabrics that connect thousands of XPUs, memory pools, storage, and accelerators.

Key Technologies at the Heart of the Deal

PCIe and CXL Switching (What XConn Brings)

• • PCI Express (PCIe) switching enables flexible connectivity between CPUs, GPUs, DPUs, storage, and accelerator devices, a foundational building block in modern servers and AI systems.
  • Compute Express Link (CXL) extends PCIe’s physical layer to include cache-coherent memory access and memory expansion, enabling pooling and sharing of memory across devices — a critical feature for AI workloads struggling with the “memory wall.”

XConn’s product readiness ladder:

• • In production: PCIe 5 & CXL 2.0 switches
  • Sampling: PCIe 6 & CXL 3.1 switches demonstrating forward compatibility with next-gen standards.

Industry reports highlight XConn’s Apollo switch series, including hybrid PCIe/CXL silicon with high port counts (e.g., up to 64–260 lanes) suitable for large systems, and upcoming Apollo 2 chips integrating PCIe Gen6.2 and CXL 3.1.

UALink: Emerging Open Connectivity Standard

Marvell repeatedly links this acquisition to UALink, which it frames as an open industry standard optimized for scale-up fabrics, networks that efficiently connect many accelerators into a coherent system. UALink is designed to deliver high throughput, low latency, and extended reach, positioning it as a more vendor-neutral alternative to proprietary interconnects like Nvidia’s NVLink.

The UALink Consortium (formed late in 2024) has been expanding its membership, including major players like Alibaba, Apple, and Synopsys, highlighting the industry’s interest in an open accelerator-to-accelerator fabric standard. Early specifications aim to support up to hundreds of gigabits per second per lane and connect large numbers of devices in AI pods.

Marvell already offers a custom UALink scale-up IP portfolio that includes controller and fabric IP, high-speed SerDes, and packaging options — intended to enable rack-scale AI deployments.

Investor and Market Reaction

Investor attention was evident as Marvell shares moved higher on the announcement, with analysts like Raymond James reiterating Strong Buy ratings and higher price targets, viewing the XConn deal as a catalyst for Marvell’s growth in AI connectivity.

Analyst firms like Zacks also noted that integrating XConn’s advanced switching tech could help Marvell build “the industry’s most comprehensive CXL portfolio,” further enhancing its strategic positioning.

According to industry reporting, Marvell expects XConn’s products to begin contributing revenue in the second half of fiscal 2027, with potential incremental revenues approaching $100 million by fiscal 2028 as demand for high-performance connectivity solutions grows.

Marvell’s acquisition of XConn is a strategic bet on the future of AI data center architecture, anchoring its position in switching silicon, memory-centric interconnects, and scale-up fabrics. By combining technical capabilities, industry standards participation, and cross-company engineering talent, Marvell aims to be a central player in the next era of AI infrastructure connectivity.