For the last decade, hyper-converged infrastructures (HCIs), a software-defined IT infrastructure model that combines compute, storage, and networking resources into a single, integrated system, have been the main route to accelerating performance and scale in data centers. But with the boom of AI, a change is happening: the new technology, especially generative AI, is supercharging demand for AI-ready data center capacities. By 2030, advanced AI workloads are projected to represent 70 percent of all data center demand in the world.

AI revolves around unstructured data that forms the foundation for every step in the AI data cycle, from ingesting and data preparation (including vectorization, labelling) to training, inference, content generation, and monitoring. With these new workloads and demands, it is unsurprising that the annual volume of data generated is expected to more than double to 527.5 zettabytes (ZB) in 2029, according to IDC.

For enterprises embracing AI and AI workloads, dealing with these huge amounts of information becomes a challenge when using HCIs, as they can lead to bottlenecks when accessing data across different servers. To counter these agility and performance inefficiencies, a viable alternative to HCIs for the AI enterprise is emerging in the form of disaggregated storage.

Flexibility isn’t just a feature – it’s foundational

For years, large businesses using HCIs had to scale storage and compute together, adding entire new servers when only one element needed expansion. Comparable to buying a seven-seater car for a family of two, IT decision makers had to pay for and run much larger operations than necessary.

As AI workloads surge and accelerated computing takes the center stage, data center architectures and storage systems must keep pace with the increasing demand for memory and compute. Yet, the fast and ever-evolving high-performance computing (HPC) and AI systems have different requirements for the various IT infrastructure hardware components. While they require Central Processing Unit (CPU) and Graphic Processing Unit (GPU) nodes to be refreshed every couple of years to keep up with the AI workload demands, storage solutions like high-capacity HDDs come with longer warranties (up to five years), are therefore built to last several years longer, and don’t need to be refreshed as often. Based on this, more and more organizations are moving storage out of the server and embracing disaggregated infrastructures to avoid wasting resources. Forecasts indicate that the market for disaggregated storage will more than double by 2033, with the model offering unprecedented flexibility, efficiency, and cost savings.

Why disaggregation is worth it: Understanding the business value

In the AI era and ZB age, IT leaders need more from their storage systems. They are looking for scalable, low-risk solutions that can evolve with them, delivering an optimized cost per Terabyte ($/TB), better energy-efficiency per TB (kW/TB), improved storage density, high-quality, and trust to perform at scale.

Disaggregated storage can be a solution that offers precisely this flexibility of demand-driven scaling to meet the individual requirements of data center workloads and business needs. By decoupling storage and computing power and dynamically combining them as needed, IT managers can achieve the following benefits:

1. Independent scalability of compute, storage, and networking resources to match workloads
2. Cost efficiency through the reduction of overprovisioning and optimization of infrastructure for better total cost of ownership (TCO)
3. Flexible adaptation to fast-changing demands depending on business priorities
4. Performance optimization by leveraging technologies like Non-Volatile Memory Express over Fabric (NVMe-oFTM) for high-speed, efficient data flow
5. Future-proof IT infrastructure that supports the transition to accelerated computing. This is achieved by decoupling storage from the server to make room for more CPU, GPU, and data processing resources (DPU) to meet the performance requirements of modern AI workloads.

For large organizations, disaggregated infrastructures help to control the cost, boost performance, and maximize IT efficiency.

The future of storage is composable and disaggregated

With disaggregated storage, enterprises can embrace AI and HPC while no longer being tethered to HCI architectures. For those businesses, there’s a clear case for decoupling storage and compute in a disaggregated approach, particularly through an efficiency lens. As more enterprises become AI-driven, greater flexibility within IT infrastructure will become a necessity for success.