Kove:SDM™ at FMS 2025 — Breaking the Memory Wall for AI, HPC, and Data Centers

Kove Ideas

John Overton CEO & Founder, KOVE

At The Future of Memory and Storage Event (FMS) 2025, Kove CEO John Overton announced a paradigm shift: memory constraints are no longer a hardware limitation, they’re a design choice. With Kove:SDM™ (Software-Defined Memory), enterprises can finally scale AI, HPC, and enterprise workloads without being bound by DIMM slots, stranded DRAM, or escalating power demands.

Santa Clara, CA — August 5, 2025

Watch the FMS 2025 Kove Keynote

Why Memory is the New Bottleneck

Modern workloads — from AI inference and training to financial risk modeling and real-time fraud detection — are not compute-bound. They’re memory-bound. That makes Kove:SDM™ a key solution to rapidly evolving tech challenges. Consider:

  • Up to 70% of DRAM sits idle even as critical jobs starve for capacity.
  • Buying more CPUs/GPUs only inflates cost, power, and cooling.
  • Stranded memory = wasted potential and capital.

Kove:SDM™ — The Software-Defined Memory Breakthrough

Kove:SDM™ virtualizes DRAM across servers into a single elastic memory pool, instantly available to any workload at local-memory latency or better.

  • Real-time allocation across the data center.
  • Outperforms in-box CXL alternatives.
  • Runs on standard x86 servers — even 10+ years old.
  • No app rewrites, no kernel mods, or new hardware required.

Validated Performance Gains

Unlike hardware-dependent approaches, Kove:SDM™ is commercially available today and in production in some of the highest stakes environments across financial services, AI research, and high-performance computing. Backed by Red Hat, Supermicro, Oracle Cloud and enterprise customers, Kove:SDM™ delivers:

5x

larger KV cache for
AI inference

60x

faster AI/ML training

100x

container density

200ms

recovery for high availability

Why It Matters for AI, HPC, and Data Centers

AI/ML models, in-memory databases, and large-scale simulations demand agility. Kove:SDM™ delivers.

  • Elastic memory on demand: Scale workloads instantly with 64 PiB of addressable memory per process, across the data center.
  • Lower TCO & sustainability gains: Get 200% ROI and up to 54% energy savings.
  • Future-ready design: Works now on x86, integrates with CXL to supercharge it’s performance and deliver what CXL cannot, performance beyond the box.

By unlocking memory as a shared, flexible layer, enterprises can scale the third dimension — memory — just like compute and storage.

The Future of Memory: Limitless, Efficient, Smart

With Kove:SDM™, memory is no longer constrained by DIMM slots. It’s only constrained by your imagination and data center floor plan.

  • Run bigger, faster AI workloads without new hardware
  • Extend infrastructure lifespan instead of costly refreshes
  • Build smarter, not bulkier

This is the new memory equation for the AI era: limitless, efficient, smart.

Get Started with Unlimited Memory

Kove:SDM™ is available as software-only or bundled with Kove:SDM™ Memory Towers. It is instantly compatible with Red Hat OpenShift®, RHEL, and x86 environments.

Start your free trial of Kove:SDM™ today and experience the unheard of performance gains already reshaping AI and HPC.