XCENA Launches MX1 Production Lineup to Accelerate Expansion in Hyperscale AI Infrastructure
SANTA CLARA, Calif. — XCENA, a company specializing in memory-centric computing solutions for artificial intelligence (AI) infrastructure, has officially introduced its MX1 production product lineup, a new family of Compute Express Link (CXL)-based memory solutions designed to overcome some of the most pressing challenges facing modern AI data centers. The announcement marks an important milestone for the company as it transitions from prototype development to commercial-ready products aimed at hyperscalers, cloud service providers, and enterprise AI deployments.
The MX1 lineup is being showcased at FMS 2026: The Future of Memory and Storage, taking place from August 4–6 at the Santa Clara Convention Center. During the event, XCENA is demonstrating how its latest innovations can help organizations address growing memory bottlenecks, reduce infrastructure costs, and improve AI inference performance through intelligent memory expansion and near-data computing.
Addressing the AI Memory Challenge
Artificial intelligence continues to evolve at an extraordinary pace. The rapid adoption of large language models (LLMs), generative AI applications, multimodal systems, and AI-powered enterprise workloads has significantly increased the demand for computing resources. While processors and accelerators have become more powerful, memory has emerged as one of the biggest constraints limiting AI performance.
Modern AI models require increasingly larger context windows and expanded Key-Value (KV) caches to deliver faster and more accurate inference. As these memory requirements grow, conventional server architectures struggle to keep pace.
High-Bandwidth Memory (HBM), while extremely fast, remains expensive and limited in capacity. Expanding AI infrastructure by simply adding more servers often leads to underutilized DRAM resources, higher power consumption, and significantly increased total cost of ownership (TCO).
XCENA believes the future of AI infrastructure depends on rethinking how memory is deployed, shared, and utilized rather than continuously adding compute resources.
Introducing the MX1 Production Lineup
The company’s new MX1 production portfolio has been designed around the open Compute Express Link (CXL) standard, enabling organizations to scale memory resources independently from processors.
Rather than forcing operators to purchase additional servers whenever memory demand increases, MX1 allows memory capacity to grow more efficiently alongside compute resources.
This architecture enables AI infrastructure operators to maximize resource utilization while reducing unnecessary hardware investments and improving overall system efficiency.
XCENA developed the MX1 lineup in collaboration with leading ecosystem partners and demonstrated compatibility with Intel Xeon 6 server platforms, highlighting the growing maturity of the CXL ecosystem.
Together, the technologies illustrate how next-generation memory expansion can address the increasingly memory-intensive requirements of AI inference.
Bringing Compute Closer to Data
XCENA Chief Executive Officer Jin Kim emphasized that memory—not computing power—is becoming the defining limitation for modern AI systems.
According to Kim, today’s AI infrastructure increasingly spends valuable time moving massive volumes of data between processors and memory rather than performing computation.
The MX1 platform addresses this inefficiency by moving compute capabilities closer to where data resides.
Instead of transferring large datasets back and forth across the system, MX1 enables data-intensive operations to occur directly adjacent to memory, reducing latency, lowering power consumption, and improving overall inference efficiency.
Kim described this approach as the natural evolution of memory architecture for the AI era, providing operators with a practical way to reduce infrastructure complexity while improving scalability.
From Prototype to Commercial Deployment
The production-ready MX1 lineup builds upon MX1P, XCENA’s prototype platform introduced the previous year.
Since its introduction, MX1P has been deployed in numerous proof-of-concept projects and technical evaluation programs with customers around the world.
These collaborations have allowed hyperscalers, enterprise organizations, cloud providers, and technology partners to validate the benefits of memory-centric computing across real-world AI workloads.
With the introduction of the production platform, XCENA aims to move beyond early evaluations toward commercial deployment discussions and production-scale testing with global customers.
The company believes the availability of production-ready hardware will accelerate adoption of CXL-based memory infrastructure across the rapidly expanding AI market.
MX1 Compute: Near-Data Processing for AI
One of the two flagship products within the MX1 family is MX1 Compute, a solution combining CXL-based memory expansion with advanced near-data processing capabilities.
The platform integrates an impressive 2,048 RISC-V processing cores, allowing computation to occur directly alongside memory rather than relying entirely on central processors.
By reducing unnecessary movement of large datasets between CPUs and memory subsystems, MX1 Compute improves AI inference efficiency while minimizing latency and energy consumption.
This architecture is particularly valuable for inference workloads involving large language models, recommendation engines, search systems, and other memory-intensive AI applications where enormous datasets must be accessed repeatedly.
The result is higher overall system performance while reducing infrastructure costs associated with excessive data movement.
MX1 Expand: Flexible Memory Scaling
Complementing MX1 Compute is MX1 Expand, a dedicated memory expansion solution designed to provide hyperscale operators with a flexible approach to increasing memory capacity.
The platform incorporates eight DDR memory slots, enabling organizations to reuse existing DRAM investments while scaling server memory more economically.
Rather than replacing entire server platforms, operators can expand available memory resources incrementally, reducing capital expenditures and improving infrastructure utilization.
This flexibility makes MX1 Expand particularly attractive for hyperscale cloud providers managing rapidly growing AI workloads that require continuous memory scaling.
The solution also supports evolving memory architectures such as pooled memory, shared memory resources, and disaggregated computing environments.
Demonstrating Large-Scale CXL Memory Pools
At FMS 2026, visitors to XCENA’s booth are experiencing several live demonstrations showcasing the capabilities of the MX1 product family.
Among the highlights is a CXL memory pooling demonstration supporting up to 20 terabytes of shared memory.
The demonstration illustrates how organizations can build flexible memory pools that dynamically allocate resources across multiple AI systems, improving overall efficiency while reducing wasted memory capacity.
The company is also showcasing KV cache sharing, demonstrating how AI inference servers can access shared memory resources more effectively.
Because KV cache requirements continue expanding alongside larger AI models, efficient sharing of memory resources represents an increasingly important capability for future AI infrastructure.
Collaboration with Intel
XCENA is also participating in demonstrations at Intel’s FMS booth, where the companies are showcasing a CXL-enabled memory architecture running on Intel Xeon 6 platforms.
The demonstration illustrates how MX1 can offload large KV caches from traditional server memory into CXL-connected memory resources.
This approach significantly improves memory scalability while increasing utilization efficiency across AI servers.
Debendra Das Sharma, Senior Fellow and Chief I/O Architect at Intel, noted that expanding AI inference workloads are making efficient memory utilization increasingly important for hyperscale infrastructure.
According to Sharma, XCENA’s MX1 demonstrates how CXL-based memory expansion can effectively support KV cache offloading on Intel Xeon 6 platforms while helping customers address demanding AI workloads.
Sharing Technical Expertise
In addition to product demonstrations, XCENA is contributing technical presentations during FMS 2026.
Chief Product Officer Harry Kim is presenting a session exploring KV Cache Sharing through cost-effective DRAM and NAND converged CXL memory technologies.
His presentation examines implementation strategies for next-generation memory systems capable of supporting rapidly expanding AI infrastructure.
Senior Marketing Manager Jay Yeon is delivering a separate technical session focused on validating the MX1 platform.
The presentation introduces early validation results while outlining a standardized methodology for evaluating the performance, interoperability, and reliability of CXL memory devices.
Together, these sessions reflect XCENA’s commitment to advancing both technology innovation and broader industry adoption of CXL-based memory architectures.
Industry Recognition
XCENA has rapidly established itself as one of the most innovative companies within the emerging CXL ecosystem.
The company received the Most Innovative Startup award during FMS 2024, followed by the Most Innovative Memory Technology award at FMS 2025.
Receiving recognition at the industry’s premier memory technology conference in consecutive years underscores XCENA’s growing influence within next-generation AI infrastructure.
Building on this momentum, the company continues expanding proof-of-concept programs with global technology leaders while accelerating commercialization efforts across hyperscale and enterprise markets.
Redefining AI Infrastructure
Founded by semiconductor veterans from Samsung Electronics and SK hynix, XCENA combines decades of semiconductor engineering expertise with advanced software capabilities to redefine how memory is utilized within AI systems.
Its comprehensive Software Development Kit (SDK) enables customers to deploy, optimize, and manage memory-centric AI infrastructure at scale.
Headquartered in Seongnam, South Korea, the company focuses on helping hyperscalers, cloud providers, and enterprises improve AI performance while reducing latency, power consumption, and total ownership costs.
With the launch of the production-ready MX1 lineup, XCENA is taking a significant step toward commercializing memory-centric computing for the AI era. As demand for larger AI models and more efficient inference infrastructure continues to grow, the company’s CXL-based solutions provide organizations with a scalable alternative to traditional server architectures. By enabling intelligent memory expansion, near-data processing, and efficient resource sharing, XCENA aims to help shape the next generation of hyperscale AI infrastructure while driving broader adoption of CXL technologies across the global data center ecosystem.
Source link: https://www.businesswire.com/