DDN, Nebul, NVIDIA Advance AI Inference with KV Cache Acceleration

DDN, the global leader in AI and data intelligence solutions, today announced continued progress in its collaboration with Nebul, a European leader in providing sovereign-hybrid cloud solutions, to optimize large-scale AI inference performance through advanced KV Cache acceleration and high-performance data infrastructure.

The collaboration brings together Nebul’s AI inference platform, DDN’s Infinia data intelligence architecture, and NVIDIA accelerated computing technologies to address one of the most important challenges facing production AI environments: maximizing the economic return on AI infrastructure investments through higher GPU utilization, faster token generation, and lower cost-per-token.

Also Read: Infrastructure as Code Security: Why Policy as Code Is Becoming Essential

As AI adoption moves from experimentation to production, organizations are confronting a new reality: the challenge is no longer whether AI works, but whether the economics work at scale.

Across the industry, enterprises, cloud providers, and sovereign AI initiatives are increasingly measuring success through business outcomes such as GPU utilization, cost-per-token, tokens-per-watt, and time-to-production. In this environment, data infrastructure has emerged as a critical determinant of AI profitability.

“The AI conversation has fundamentally changed,” said Alex Bouzari, CEO and Co-Founder at DDN. “For years, the industry focused on acquiring GPUs. Today, the question is how efficiently those GPUs generate value. Inference has become the economic engine of AI, and reducing the cost of every token produced is now one of the most important challenges facing the industry.”

As part of an ongoing proof-of-concept engagement, DDN and Nebul are validating next-generation KV Cache acceleration capabilities designed to support NVIDIA DSX-based AI factory deployments by improving inference efficiency, reducing latency, and increasing utilization of AI infrastructure.

“For years, the industry focused on building larger models. Today, the challenge is making those models economically viable in production,” said Arnold Juffer, CEO at Nebul. “Every organization is looking for ways to generate more value from its AI infrastructure investments. Through our collaboration with DDN and NVIDIA, we are demonstrating how KV Cache optimization and high-performance data architectures can improve inference efficiency, reduce latency, and help unlock the next phase of AI adoption.”

“AI infrastructure is increasingly defined by efficiency at scale,” said Rod Evans, Vice President of Cloud Infrastructure at NVIDIA. “As organizations deploy larger models and agentic AI workloads into production, technologies that improve GPU utilization, reduce latency, and accelerate token generation become critical. DDN continues to be an important collaborator in advancing the data and infrastructure capabilities needed to support the next generation of AI factories.”

Also Read: Terraform vs OpenTofu: Which Infrastructure as Code Platform Should Enterprises Choose?

Recent benchmarking efforts have demonstrated promising early results, including measurable improvements in Time-to-First Token (TTFT) performance with KV Cache enabled. The collaboration has successfully completed RoCE-based infrastructure validation and continues to expand benchmarking activities across larger inference sequence lengths while identifying additional optimization opportunities within DDN’s Infinia platform.

The project has also expanded to include collaboration with NVIDIA around benchmarking methodologies, scalability validation, and future technical publications.

The Economics of AI Have Shifted

Industry attention has historically centered on model performance and GPU availability. However, as organizations deploy AI into production, a new bottleneck has emerged: data movement and inference efficiency.

Agentic AI workloads, retrieval-augmented generation (RAG), and large-scale inference environments place unprecedented demands on infrastructure, including networking and storage. Every millisecond of latency and every percentage point of GPU idle time directly impact profitability.

“Training builds the asset. Inference is where it earns,” said Bouzari. “The next generation of AI applications will be powered by agents that perform meaningful business transactions and decisions. The economics of those systems depend on infrastructure that can deliver data at the speed AI operates.”

DDN’s Infinia platform was purpose-built to address these challenges through distributed KV Cache services, GPU-native data movement, intelligent data orchestration, and high-performance storage architectures that maximize accelerator efficiency.

Accelerating the AI Factory Era

The Nebul collaboration reinforces DDN’s broader vision that AI infrastructure must evolve beyond traditional storage architectures and become an active participant in AI execution.

The collaboration supports the broader NVIDIA DSX platform approach to AI factories, where compute, networking, storage, software, and operations are designed together to improve tokens-per-watt, cost-per-token, and time-to-production.

As organizations seek to operationalize AI at scale, DDN believes the defining metrics of success will increasingly become:

• GPU utilization
• Cost-per-token
• Tokens-per-watt
• Time-to-first-token
• Time-to-production

Organizations that optimize these metrics will achieve sustainable AI economics. Those who do not risk deploying infrastructure that remains underutilized despite significant investment.

“We don’t sell storage,” Bouzari added. “We help organizations maximize the economic return on every GPU, every token, and every watt. That’s the foundation of the AI economy.”