SMART Modular Technologies and Pegatron Present CXL Memory Advancements for AI Infrastructure

During a COMPUTEX 2025 sideline event, SMART Modular Technologies, in collaboration with Pegatron, shared insights into the growing role of Compute Express Link (CXL) in shaping the future of AI infrastructure. The session focused on current developments in CXL-based memory solutions, the supporting ecosystem, and the readiness of enterprise hardware for next-generation workloads.

Company Overview: SMART Modular Technologies

SMART Modular Technologies is a global provider of advanced memory and storage solutions, with a strong presence in enterprise, industrial, and data center environments. The company offers a range of DRAM modules, NAND-based storage, and custom-engineered solutions tailored to emerging compute demands, including AI, edge computing, and data-intensive applications.

Understanding CXL: Architecture and Use Cases

Compute Express Link (CXL) is a high-speed, CPU-to-device interconnect standard built on PCIe physical infrastructure. Unlike traditional PCIe, which is designed primarily for I/O, CXL enables coherent memory sharing and low-latency communication across CPUs, memory expanders, and accelerators.

The CXL protocol stack includes:

• CXL.io – for configuration and control, equivalent to standard PCIe
• CXL.cache – allowing accelerators to cache host memory
• CXL.mem – enabling host and devices to access shared memory in a coherent manner

These capabilities position CXL as a key enabler for heterogeneous computing environments, particularly those involving AI/ML workloads.

CXL for AI Workloads

During the session, Morris Yang, APAC Marketing Manager at SMART Modular Technologies, explained how CXL helps address three major challenges in AI compute environments:

1. Memory Scalability – CXL modules expand memory capacity beyond native CPU limits without architectural changes
2. Resource Efficiency – Memory pooling across systems reduces over-provisioning and underutilization
3. Performance Gains – Reduced latency in shared memory access benefits real-time AI inference and training pipelines

The benefits are particularly relevant to large language models, dynamic memory workloads, and high-throughput inference operations that require flexible, low-latency access to large datasets.

SMART’s Product Showcase: CXL-Compatible Modules and AICs

SMART presented a portfolio of CXL-ready products, including:

• CXL Memory Modules (CMMs) – conforming to CXL 2.0 with forward compatibility to CXL 3.1
• E3.S and U.2 Add-in Cards (AICs) – for flexible integration in modern server platforms
• Custom memory expansion solutions – built for data centers and AI-optimized compute infrastructure

These components are designed for integration into server environments adopting CXL-compatible CPUs and chipsets, enabling memory disaggregation and modular scaling.

Pegatron’s Role: Platform Readiness and Ecosystem Support

The event also featured insights from Michelle Wu, Project Director at Pegatron BU18, who shared how Pegatron is working with ecosystem partners to advance the standard and incorporate CXL 3.1 capabilities into its server platforms in the future.

CXL 3.1 introduces features such as:

• Memory Sharing Across Hosts – enabling a more dynamic and distributed architecture
• Switching and Fabric Enhancements – for building scalable memory fabrics
• Improved Reliability and Manageability – critical for enterprise environments

According to Wu, Pegatron's platforms are aligning with these advancements, offering early access to system builders and OEMs preparing for next-generation AI deployments.

Conclusion

This COMPUTEX 2025 sideline event highlighted how CXL is transitioning from concept to implementation, with real products, ecosystem collaboration, and platform readiness. For AI infrastructure architects, the combination of SMART’s memory expansion modules and Pegatron’s CXL-integrated systems reflects the maturing state of the technology and its relevance to scalable, efficient enterprise computing.