Marvell Technology announced a definitive agreement to acquire XConn Technologies, a leader in PCIe and CXL switching silicon. The deal, valued at approximately $540 million (60% cash and 40% stock), is a decisive move to bolster the Ultra Accelerator Link (UALink) ecosystem the open-standard alternative to NVIDIA’s proprietary NVLink.
Why XConn is the “Missing Piece” for Marvell
Marvell already leads in optical connectivity and SerDes technology, but XConn brings the industry’s first hybrid switch capable of supporting both CXL and PCIe on a single SoC. This is critical for 2026 AI data center architectures, which are moving from single racks to multi-rack “scale-up” environments.
Key Strategic Gains:
- UALink Leadership: Marvell gains XConn’s elite engineering team to lead the development of UALink scale-up switches, designed to connect hundreds of accelerators into a single, unified “Super-XPU.”
- Memory Disaggregation: By combining Marvell’s CXL controllers with XConn’s switches, the company can now offer a full-stack solution for memory pooling, allowing AI models to access terabytes of shared DRAM with near-native latency.
- First-to-Market Silicon: XConn is already sampling PCIe 6.0 and CXL 3.1 products, positioning Marvell to capture the next refresh cycle in hyperscale infrastructure.
Also Read: VAST Data Transforms AI Inference Architecture for the Agentic Era in Collaboration with NVIDIA
The $100 Million Growth Engine
Marvell expects XConn to become accretive to earnings by the second half of fiscal 2027. Projections suggest the division will ramp to $100 million in revenue by fiscal 2028, fueled by more than 20 existing hyperscale customer engagements.
The Competitive Landscape: Marvell vs. Broadcom
This acquisition, following the December 2025 purchase of Celestial AI, places Marvell in a head-to-head battle with Broadcom for the future of the AI data path. While Broadcom has focused on Ethernet scale-out, Marvell is doubling down on the scale-up fabric, betting that the future of LLM training lies in massive, tightly-coupled accelerator clusters using open standards.
The Bottom Line: Marvell is moving beyond being a “component supplier” to becoming the primary architect of the AI interconnect. For enterprise IT leaders, this means a more flexible, non-proprietary path to scaling their private AI clouds.
