What Marvell’s acquisition of XConn Technologies reveals about the future of AI data center switching

Marvell Technology Inc. (NASDAQ: MRVL) has announced its intent to acquire XConn Technologies in a $540 million transaction, composed of approximately 60 percent cash and 40 percent stock. The move signals a decisive push to consolidate leadership in PCIe and Compute Express Link (CXL) switching silicon, a foundational layer for modern artificial intelligence infrastructure. The deal also expands Marvell Technology’s Ultra Accelerator Link (UALink) switching team with highly specialized engineering talent at a time when scale-up connectivity is becoming critical for hyperscale AI deployments.

The acquisition brings XConn’s production-grade PCIe 5 and CXL 2.0 switches into Marvell Technology’s portfolio, with PCIe 6 and CXL 3.1 products currently sampling. With more than 20 customers already engaged, the integration is expected to unlock approximately $100 million in revenue by fiscal 2028 and solidify Marvell Technology’s role as a core enabler of UALink-driven AI platform architectures.

Why switching silicon is becoming the central battlefield in next-generation AI data centers

As artificial intelligence workloads grow in scale and complexity, infrastructure design is rapidly evolving beyond traditional single-rack deployments. Training foundation models, running distributed inference workloads, and sharing memory across accelerators require not just computational horsepower but also a low-latency, high-bandwidth communication fabric. This has created an urgent industry need for advanced interconnect technologies capable of linking thousands of XPUs across boards, chassis, and racks.

Unlike standard Ethernet-based leaf-spine network designs that have long dominated cloud data centers, AI training and fine-tuning systems require more deterministic, tightly coupled interconnect. This is where PCIe and CXL switching become critical. PCIe provides high-throughput communication for components within servers, while CXL enables shared memory pooling and cache coherency across different processing elements.

CXL in particular is unlocking new memory disaggregation models, allowing systems to scale more cost-effectively and enabling accelerators to work as a unified memory domain. As generative AI models stretch beyond hundreds of billions of parameters, memory flexibility and latency optimization are becoming just as important as raw compute. This trend explains why switching silicon—once an afterthought—is now a competitive frontier in AI system design.

How the XConn deal advances Marvell Technology’s UALink ambitions and AI platform roadmap

At the center of this strategic acquisition is UALink, a new open standard being developed by a consortium of companies including Marvell Technology, Advanced Micro Devices, Intel Corporation, Hewlett Packard Enterprise, Cisco Systems, and others. UALink is designed specifically for AI and high-performance computing workloads that require scale-up fabric across multiple racks and servers. It provides a standardized, ultra-low latency interconnect for sharing memory and coordinating accelerator activity.

Marvell Technology has been an early contributor to the UALink specification. However, building a production-ready switching solution that meets UALink’s performance, latency, and interoperability requirements requires deep protocol expertise and silicon maturity. This is where XConn’s product portfolio becomes a force multiplier.

XConn is already shipping PCIe 5 and CXL 2.0 switches to more than 20 customers, with next-generation PCIe 6 and CXL 3.1 switches currently in sampling. These parts are designed to support high port counts and dense memory-sharing configurations, making them ideally suited for large-scale accelerator deployments. The hybrid switching architecture, which supports both PCIe and CXL on a single chip, offers flexibility for system integrators looking to optimize bandwidth, cost, and system layout.

In combining forces, Marvell Technology gains not only proven silicon but also a highly experienced engineering team with deep domain expertise in high-speed I/O and memory interconnect. The company’s internal UALink switching initiative now becomes a much broader platform, accelerated by XConn’s IP and customer momentum. Matt Murphy, Chief Executive Officer of Marvell Technology, emphasized that the combination creates a compelling switching platform for AI and cloud infrastructure that prioritizes architectural flexibility and performance tuning.

How this acquisition strengthens Marvell’s competitive position against NVIDIA, Intel, and Broadcom

The AI infrastructure market is becoming increasingly stratified between companies building full-stack systems and those specializing in critical horizontal layers such as interconnect. NVIDIA Corporation has opted for full vertical integration, bundling GPUs with proprietary NVLink and NVSwitch interconnect in its DGX and HGX platforms. Intel Corporation is embedding CXL switching into its Xeon and Agilex portfolios, while Broadcom dominates Ethernet switching with its Tomahawk and Trident families but has yet to make a decisive move in scale-up switching.

Marvell Technology, with its focus on interconnect, SerDes, optical DSPs, and now PCIe and CXL switching, is positioning itself as the neutral enabler for scale-up systems. The company does not compete with NVIDIA or Advanced Micro Devices in compute, which makes it a preferred partner for hyperscalers and original equipment manufacturers seeking modularity and flexibility. The XConn acquisition reinforces this positioning, especially as UALink moves toward ratification and deployment in late 2026.

XConn’s customer engagements also give Marvell Technology an immediate pipeline of design wins across enterprise, cloud, and system OEM customers. The company’s hybrid PCIe-CXL switching chips complement Marvell Technology’s existing memory controller IP, optical interconnect capabilities, and emerging AI platform initiatives, including its pending acquisition of Celestial AI.

Taken together, these moves create a differentiated strategy focused not on compute but on the interconnect fabric that binds compute together. In an AI environment increasingly defined by data movement and memory access patterns, this strategy could prove more scalable than vertically integrated alternatives.

What institutional investors should watch as Marvell moves toward closing and integration

The financial structure of the XConn acquisition is consistent with Marvell Technology’s recent approach to capital allocation. The $540 million price tag will be paid in approximately 60 percent cash and 40 percent stock, with the stock valued based on a 20-day volume weighted average price. The deal is expected to close in early 2026, pending regulatory approvals and customary conditions.

Importantly, Marvell Technology does not expect revenue contribution until the second half of fiscal 2027, when XConn’s CXL and PCIe 6 switches are expected to ramp into production. The acquisition is expected to become accretive to non-GAAP earnings in the same timeframe, eventually reaching $100 million in revenue by fiscal 2028. This places the deal within the strategic investment window for hyperscaler AI infrastructure upgrades targeting 2026 and beyond.

From a risk perspective, investors will be watching closely for signals on post-close integration. Marvell Technology must align internal and acquired product roadmaps, preserve XConn’s customer momentum, and continue contributing to UALink’s ecosystem development. There is also the execution risk associated with standard convergence and vendor adoption. If UALink fails to gain traction, or if proprietary alternatives fragment the market, Marvell Technology’s switching bets could face delays in monetization.

Nonetheless, institutional sentiment around Marvell Technology remains positive. The company’s proactive acquisitions, broad IP portfolio, and neutral ecosystem positioning give it optionality in a highly fluid AI landscape. The XConn deal further reinforces the view that Marvell Technology sees AI connectivity as a structural growth vector, not just a feature layer.

Why this deal signals a turning point in the economics of AI infrastructure deployment

The most important takeaway from Marvell Technology’s acquisition of XConn Technologies may be what it signals about the next phase of AI infrastructure economics. In the early days of generative AI, attention focused primarily on compute throughput, with companies racing to acquire the most graphics processing units. However, bottlenecks in memory access, latency, and data movement are now reshaping where value is captured in system architecture.

Switching silicon, especially for memory-coherent workloads across accelerators, is emerging as a key control point. Companies that can provide flexible, high-performance interconnect have the opportunity to shape how memory disaggregation, multi-modal model orchestration, and real-time inference are delivered at scale.

The Marvell Technology and XConn combination is a clear expression of this shift. It acknowledges that the era of static server architectures is over and that the next generation of data centers will be defined by composability, interconnect topology, and performance per watt of data movement—not just compute density.

With this acquisition, Marvell Technology has made its bet that owning the switching fabric will be just as important as owning the GPU when it comes to building infrastructure for the AI era.

Key takeaways on what this acquisition means for Marvell, its rivals, and the AI infrastructure market

• Marvell Technology is acquiring XConn Technologies for $540 million to strengthen its position in PCIe and CXL switching for AI data centers.

• The deal directly supports Marvell’s UALink scale-up switching roadmap, enabling dense, low-latency interconnect for multi-rack AI systems.

• XConn brings production-ready PCIe 5 and CXL 2.0 switches, with PCIe 6 and CXL 3.1 in sampling, and over 20 engaged customers.

• The acquisition expands Marvell’s TAM and is expected to generate $100 million in revenue by fiscal 2028, becoming accretive by fiscal 2027.

• Marvell gains a deeper bench of engineering talent and IP that complements its existing SerDes and optical interconnect assets.

• The deal positions Marvell as a neutral interconnect provider amid a landscape dominated by vertically integrated players like NVIDIA.

• Execution risks include customer retention, roadmap integration, and ecosystem alignment around UALink as an emerging standard.

• The transaction underscores a growing industry view that switching fabric will become a primary value driver in AI infrastructure design.