As AI drives an unprecedented explosion in data volumes, Alex Segeda, Platforms Business Director, EMEAI at WD, explains why storage strategy will be critical to turning that data into lasting value.

AI is driving an unprecedented surge in the amount of data being generated worldwide; IDC forecasts that global annual data generation will increase from 218 ZB in 2025 to 718 ZB by 2030, representing a 26.9% CAGR. Yet organisations still store only a small proportion of the data they produce, with an even smaller share being analysed.

This gap sits at the heart of the AI-era challenge. Data is increasingly one of an organisation’s most valuable assets, but only when it can be accessed, trusted and managed cost-effectively. The limiting factor is no longer compute alone. It’s the ability to store, move and manage growing volumes of data efficiently and economically over time.

AI is, fundamentally, a data system. Every training run, inference process and interaction creates new data that persists over time. Compute is temporary and can be reused, while data is persistent and continues to accumulate. As volumes increase, the choices organisations make around storage directly influence both cost and capability.

At scale, this requires a different approach to infrastructure design. Treating every bit of data in the same way is no longer practical. Instead, value comes from placing data in the right location, on the right media and at the right point in its lifecycle, while maintaining the appropriate cost profile. Flash provides the performance needed for demanding workloads, while HDDs offer a cost-effective foundation for storing large volumes of information. In hyperscale environments, economics and a strategy built for scale, rather than preference, ultimately determine the right approach.

For organisations seeking to turn growing data volumes into lasting value, how this strategy is implemented will depend on the environment, workload and specific business requirements.

Designing infrastructure around data growth

Take Enzo, the head architect at a global enterprise. His challenge is to design and operate a storage architecture that can scale reliably and cost-effectively across regions, platforms and regulatory frameworks. He is accountable for performance, availability and security, while managing constant data growth and increasing system complexity.

Enzo approaches AI infrastructure as a coordinated data system. He defines how storage tiers support different workloads and reserves flash for performance-critical use cases (e.g. active AI pipelines), while HDDs hold datasets, backups and long-term records that must scale economically.

A key part of Enzo’s role is reducing fragmentation and helping ensure future scalability. Data that sits across disconnected systems limits utilisation and increases operational overhead. By consolidating data into workload-specific systems and enforcing consistent policies for movement between tiers, he can help ensure that data is both accessible and stored efficiently. This enables the organisation to support AI workloads without compromising resilience or compliance. His success is measured by delivering predictable performance and uptime while maintaining control over cost and complexity at scale.

Bringing data closer to the core of the platform

Sofia, a senior systems director at a neocloud provider, operates with a different set of priorities. She is responsible for building infrastructure that delivers high performance to customers while maintaining strict control over cost, risk and efficiency. Her decisions are shaped by return on investment, security and competitive differentiation.

Some neocloud providers rely on hyperscalers for large-scale data storage, while focusing on delivering GPU-as-a-service (GPUaaS) to AI and HPC customers. This model allows them to scale quickly and compete on performance without investing heavily in data infrastructure.

However, it can also create a structural challenge further down the line. By storing data outside their own infrastructure, providers may have less control over its use. Data generated through training and inference may remain external, potentially restricting their ability to optimise systems over time or develop differentiated services based on that data.

For Sofia, the challenge is to balance short-term efficiency with long-term control. Performance tiers remain essential to support demanding workloads, but introducing a cost-efficient capacity layer allows her to retain more data within her own platform. This capacity layer can also support future scalability while maintaining economic viability. It can improve data locality, reduce dependency on external providers and create opportunities to generate additional value from accumulated data.

Her success depends on maintaining performance and reliability while steadily increasing control over the data that underpins her services.

Moving from AI experimentation to production at scale

Adam, an infrastructure engineer at an AI specialist, focuses on enabling efficient operations. His priority is ensuring that data moves through systems fast enough to keep compute resources fully utilised. In his environment, any delay in data access translates directly into wasted GPU capacity.

For many AI teams, the most practical starting point is to rely on neocloud platforms for compute resources. This allows rapid experimentation and avoids upfront investment. From a cost perspective, it often makes sense in the early stages.

As workloads mature, however, trade-offs might appear. Running training and inference on separate external systems can limit control over how data is retained and reused. The value of AI systems increasingly lies not just in models, but in the data they generate and refine over time.

As his models move from experimentation to production, Adam should consider designing infrastructure that supports both immediate performance and long-term scalability. Flash is used where low latency directly affects workload efficiency, particularly in active pipelines. At the same time, the continuous growth of training data, checkpoints and outputs requires scalable and cost-effective storage, where HDDs can play an important role.

Building more of the data infrastructure internally through a tiered strategy can improve control, reduce inefficiencies and allow the organisation to extract greater value from its data over time. Adam’s role is to navigate this transition, ensuring that systems remain efficient while supporting future growth.

Building the infrastructure behind long-term data value

There is no single blueprint for turning growing data volumes into meaningful value. The requirements of a global enterprise, a neocloud provider and an AI specialist are fundamentally different. What they share, however, is the need to view storage not simply as a supporting layer, but as the foundation of the wider data system.

Across these environments, the same core principles apply. Data needs to be positioned according to the requirements of each workload, with performance resources deployed where they deliver the greatest benefit. At the same time, scalable capacity provides the economic foundation for continued growth, creating a sustainable balance between price and performance. The right architecture ultimately depends on the organisation’s specific environment and priorities, including its requirements for control, latency, compliance and scalability.

As AI continues to expand, data will remain the constant. Organisations that build their infrastructure around how data is generated, stored and ultimately used will be better placed to turn increasing data volumes into long-term value.