CXL 2.0: Penguin Revives DDR4 for AI Data Centers

Penguin Solutions launches a CXL 2.0 card capable of aggregating up to eight RDIMM DDR4 modules to extend the memory of AI servers. The offering comes with a qualification process and a recovery program to secure the reuse of existing modules, as demand for AI-related DRAM surges and drives up deployment costs.

Availability and Reliability Safeguards Announced by Penguin

The SMART CXL-8F2W-D4 module is offered for qualified customer opportunities on CXL 2.0 compatible platforms. To frame the deployment of reused memory, Penguin pairs this product with its Zefr high-reliability memory selection process. In this context, DDR4 belonging to customers is inspected, tested, and qualified before installation. Each populated card is also validated in a compatible host system before shipment. According to Penguin Solutions, Zefr simulates demanding usage conditions to attest to the validity of DDR4 modules in a production environment.

A Memory Market Under Pressure from AI Workloads

According to TechInsights, AI data centers will account for 34% of global DRAM demand by 2026, and this share could reach nearly 50% by 2029. This trend is accompanied by rising memory costs and increased pressure on infrastructure budgets. For organizations deploying new servers, during renewals or new constructions, these costs present a significant challenge.

Operation: Up to Eight RDIMM DDR4 Exposed as Additional Capacity

The SMART CXL-8F2W-D4 is an eight-slot RDIMM DDR4 memory card based on the Marvell StructeraX CXL controller. Once installed, the module makes this memory available to compatible servers as additional system capacity. According to Penguin Solutions, this approach particularly targets AI inference computing scenarios, in-memory databases, and other memory-intensive workloads.

DDR4 Reuse: Recovery Program and Potential Cost Impact

Penguin Solutions highlights the availability of DDR4 from previous-generation server farms and has launched a program to recover these modules during renewal cycles or to ensure their supply through resellers. This system allows companies to reuse their stock of DDR4 to meet immediate needs. According to the presentation, this practice could help reduce expenses related to increasing memory capacity, extend the lifespan of existing components, and optimize previous investments while meeting the reliability and compatibility requirements associated with re-circulated modules. The company promotes DDR4 as a replacement solution for DDR5, emphasizing the announced advantages in performance, availability, and cost for AI infrastructures.

Publisher Positioning and Executive Message

Satya Iyer, Senior Vice-President and General Manager, Embedded Memory at Penguin Solutions, emphasizes that the acceleration of inference workloads increases the challenges related to rising memory costs for building and expanding AI data centers. He argues that many organizations have already invested significantly in DDR4, which can be mobilized to meet growing demand. According to him, the combination of CXL and Penguin's expertise in selection and validation helps customers maximize the value of their memory investments, reduce infrastructure costs, and meet performance and reliability requirements in production. Penguin Solutions also positions itself as a designer, builder, and operator of next-generation data centers for enterprises, sovereign AI initiatives, and cloud-native players, through its Full-Stack AI Factory platform that combines infrastructure software, advanced memory, computing systems, services, and partner technologies. Additional information is available online. The announcement was made in Fremont, California, on November 13, 2026.