As enterprises across India and the Middle East push AI workloads from pilot to production scale, a quiet bottleneck is emerging—not in compute, but in storage. In this conversation with CIO&Leader, Owais Mohammed, Sales Director, India, Middle East, and Africa at WD, unpacks why the industry’s fixation on GPUs overlooks a more fundamental constraint: feeding those GPUs with data efficiently. Mohammed makes the case against an all-flash future, arguing that high-capacity HDDs remain structurally essential to AI pipelines at scale, both economically and operationally. He also discusses India’s greenfield advantage, WD’s roadmap toward 100TB+ HAMR drives, and how power efficiency and post-quantum security are becoming non-negotiable design priorities for AI-era infrastructure.
Sales Director, India, Middle East, and Africa,
WD
CIO&Leader: The AI conversation in most boardrooms still defaults to GPUs and compute capacity. At what point does storage infrastructure become the actual bottleneck, and are most enterprises in India and the Middle East approaching that inflection point yet?
Owais Mohammed: It’s really about the data. Data storage can become the bottleneck when AI moves beyond pilots into scaled, production workloads. At that stage, the constraint is no longer how much compute can be added, but whether the IT infrastructure continuously feeds GPUs with the right data at the right speed across ingestion, preparation, training, checkpointing, inference, and retention. When data cannot be stored, accessed, and moved efficiently, even the most powerful GPUs cannot operate at their full potential.
Many enterprises in India and the Middle East are now approaching this inflection point. AI adoption, cloud growth, sovereign AI initiatives, and rapid data center expansion are driving data volumes at a pace that traditional storage models cannot handle. At the same time, power availability, operating costs, and physical rack or floor space are becoming bottlenecks, so operators often cannot simply scale by adding more hardware. The conversation is shifting from capacity-led expansion to efficiency-led, tiered storage infrastructure designs: flash for time-sensitive workloads and high-capacity, power-optimized HDDs for virtually everything else—massive datasets, checkpoints, logs, and retained data that make large-scale AI economically viable. Enterprises that move early toward this architecture will likely have greater certainty about their ability to scale economically.
CIO&Leader: There is a widely held assumption that “hot” data and real-time inference workloads will eventually migrate entirely to flash and SSD. What is your counterargument for why high-capacity HDD remains structurally essential to AI data pipelines at scale?
Owais Mohammed: The assumption that AI data will eventually move to all-flash overlooks a fundamental reality of AI: economics. At AI scale, putting the majority of data on flash is not economically effective; flash can be up to 22x the cost of HDDs. Plus, most data do not remain in hot or active tiers, but needs to be retained, managed, and accessed at scale. AI infrastructure is a system, not a single tier. Different workloads place different demands on storage, and designing around a single performance tier ignores how AI data actually moves through its lifecycle. While SSDs play a critical role in accelerating training and inference, AI pipelines generate enormous volumes of data that spend most of their lifecycle outside high-performance environments. Training datasets, checkpoints, model versions, inference logs, compliance records, and archived data must be stored efficiently over long periods and require massive capacity, reliability, and efficiency. This is where high-capacity HDDs remain structurally essential.
At AI scale, infrastructure design must reflect data realities. The more effective approach is a tiered architecture where flash handles immediacy and performance, while HDDs provide the most economical, scalable capacity foundation that AI depends on.
CIO&Leader: AI model training, inferencing, and data retention are generating radically different storage workload profiles. How should data center architects today think about tiering their storage infrastructure to serve all three without over-provisioning or creating performance gaps?
Owais Mohammed: Architects need to tier storage across the AI lifecycle, rather than default to a single high-performance layer. Training, inference, and retention each impose different storage demands, and treating them as a single workload leads to either over-provisioning or performance gaps.
The right approach is to match each stage to the appropriate storage tier. Active training and inference workloads need low-latency, high-throughput access close to compute. But AI pipelines also generate massive volumes of intermediate outputs, versioned models, operational logs, and retained data that do not require the same performance tier. High-capacity, power-efficient HDDs serve this layer with the storage density, durability, and efficiency that AI-scale data demands.
The goal is a workflow architecture designed for optimal performance, capacity, and low power so data flows seamlessly across tiers without slowing the pipeline.
CIO&Leader: Power efficiency has become one of the most urgent design constraints for hyperscale and enterprise AI data centers globally, and particularly so in markets like India, where energy costs and grid reliability are real concerns. How is WD engineering for performance-per-watt in its current HDD roadmap?
Owais Mohammed: Power efficiency is no longer just a sustainability goal; it is a core requirement of AI infrastructure. As AI data volumes grow, the challenge is not simply storing more data but doing so with the lowest possible power and operational cost. WD’s HDD roadmap is focused on lowering TCO. Higher-capacity drives allow customers to store more data with fewer drives, servers, and rack space, improving overall infrastructure efficiency. Customers are increasingly focused on both sustainability and operating efficiency, making it especially important to increase storage density while significantly lowering power consumption per terabyte. (Watts/ per terabyte).
Similarly, by deploying the highest-capacity HDDs available, organizations can achieve meaningful gains in energy efficiency and storage density. WD’s helium‑sealed HDDs, for example, operate at approximately 0.3 watts per terabyte in idle, compared to nearly 2.85 watts per terabyte in older air‑based designs. In addition, new upcoming power‑optimized HDDs are designed to reduce energy consumption by up to 20 percent while increasing capacity and performance.
CIO&Leader: India is at an interesting inflection point: sovereign AI ambitions, a rapidly expanding data center pipeline, and a largely greenfield enterprise storage landscape. In your assessment, does this allow India to build AI storage infrastructure more intelligently than markets that are now retrofitting legacy environments?
Owais Mohammed: Yes, India’s greenfield enterprise landscape and expanding data center pipeline give it a structural advantage over markets now retrofitting legacy environments. India can skip incremental upgrades and build AI-native tiered infrastructure from day one.
Instead of treating storage as an afterthought, Indian IT system architects can design from the ground up around the reality that AI infrastructure is a holistic data system, not just a computer system. By deploying workload-aligned, tiered architectures from the start, India can scale its AI ambitions with greater certainty and efficiency with the best economics at scale.
CIO&Leader: WD operates in the HDD ecosystem. As your customers in India and the MEA region build out AI infrastructure, how are you advising them to balance investment between Flash and HDD, and where do you see that balance shifting over the next three to five years?
Owais Mohammed: We advise customers to move beyond a binary Flash‑vs‑HDD decision and instead design storage around how data behaves across the AI lifecycle. It’s not Flash vs HDD. It’s Flash AND HDD. AI infrastructure today spans everything from high‑performance training to long‑term data retention and governance, each with distinct storage requirements. A single‑tier all-flash approach may work early but becomes inefficient, costly, and unsustainable as AI environments scale.
Across India and the MEA region, organizations increasingly recognize that AI success depends on how well they balance performance with scalable, cost-efficient data foundations. As AI data volumes compound, customers will need to store more data without proportionally increasing power, rack space, or overall infrastructure costs. Moving to the highest capacity HDDs can help.
CIO&Leader: What are the most significant innovations WD is currently driving in the HDD ecosystem, whether in areal density, HAMR technology, interface architecture, or drive firmware, that are specifically designed to address AI-era data center requirements?
Owais Mohammed: WD’s HDD innovation roadmap for the AI era is focused on four priorities: higher capacity, better performance, improved sustainability, and easier deployment. On capacity, WD has announced the industry’s highest capacity, 40 TB UltraSMRePMRHDD, now in customer qualification, while advancing its HAMR roadmap designed to scale to 100TB+ by 2029. These innovations help customers store far more AI data within the same data center footprint.
In terms of performance, technologies such as High Bandwidth Drive and Dual Pivot Drive are designed to increase bandwidth and sequential I/O for AI workloads while preserving HDD economics. Alongside OptiNAND, UltraSMR firmware, power-optimized designs, and an open API software layer, WD is enabling scalable, energy-efficient deployment of next-generation storage without major infrastructure disruption.
CIO&Leader: As AI workloads scale, data governance, retention compliance, and long-term archival are becoming storage design requirements, not afterthoughts. How CISOs and infrastructure leaders should be thinking about the intersection of data security, sovereignty, and storage architecture, and what role does WD play in that conversation?
Owais Mohammed: As AI workloads scale, governance, security, sovereignty, and retention can no longer be treated as downstream compliance challenges. AI creates a continuous data lifecycle across ingestion, training, inference, checkpointing, and long-term retention. Hence, CISOs and infrastructure leaders need clear control over where data resides, how it is protected, and how reliably it can be retained and accessed over time.
For WD, this means enabling a security‑by‑design approach, with protections embedded across firmware, hardware, and manufacturing to ensure trusted device integrity, strong access controls, and industry‑standard encryption.
At the same time, the threat landscape is evolving. As quantum computing advances, traditional encryption models may be challenged. WD is addressing this by integrating NIST‑approved post‑quantum cryptography (PQC) into its upcoming UltraSMR drives, embedding quantum‑resistant security into features such as secure boot and firmware updates. Together, this positions storage not just as infrastructure, but as a trusted foundation for secure, sovereign, and compliant AI at scale.