AI sustainability depends on data lifecycle control, not model scale

At a glance

What this is: AI sustainability is presented as a data lifecycle problem, with unused data, storage inefficiency, and infrastructure overhead driving much of the environmental burden.

Why it matters: For identity and security practitioners, the message is that data governance choices increasingly shape resilience, cost, and operational footprint, especially where AI workloads expand data estates and access paths.

Context

AI sustainability is not just about how large a model is or how many prompts it serves. The bigger governance problem is the data and infrastructure behind the model, including what gets stored, moved, and retained long after it has any operational value. That matters to security teams because the same sprawl that wastes energy also widens the control surface for governance, access, and recovery.

Commvault’s article frames dark data and storage inefficiency as a lifecycle issue rather than a pure technology issue. In practice, that puts data classification, retention, tiering, and deletion closer to the center of AI governance, because unmanaged data creates both unnecessary environmental load and unnecessary security complexity.

Key questions

Q: How should teams reduce the environmental impact of AI without slowing adoption?

A: Start by reducing unnecessary data growth, because storage, movement, and cooling drive a large share of the footprint. Classify data by business value, remove duplication, tier infrequently used information, and delete what no longer needs to exist. That approach lowers resource use without constraining legitimate AI work.

Q: Why does dark data make AI governance harder?

A: Because unused data still has to be protected, backed up, moved, and audited. That adds cost and complexity while increasing the number of systems, permissions, and recovery points that must be governed. The result is a larger operational footprint and more places where policy drift can occur.

Q: What operational signals show that AI storage is becoming inefficient?

A: Common signals include rapidly growing storage footprints, repeated copies of the same datasets, low retrieval rates, and backups that protect large volumes of information with little active use. If data growth outpaces business demand, the storage model is probably carrying hidden waste.

Q: How do sustainability controls relate to resilience in AI environments?

A: They overlap because data that is organised, minimised, and lifecycle-managed is easier to secure and faster to recover. Reducing redundant data lowers the amount of information that must be protected and restored, which improves continuity while also reducing unnecessary infrastructure load.

Technical breakdown

Why unused data drives AI sustainability costs

AI systems consume energy at several layers, but data volume is one of the least visible drivers. Every stored dataset creates demand for storage, backups, replication, processing, and movement across systems. When large portions of that data are never accessed, the organisation keeps paying operational and environmental costs without receiving value. This is why dark data matters in AI programmes: it turns a data platform into a long-lived overhead rather than an active capability. For practitioners, the technical issue is not only the model. It is the lifecycle of the information feeding the model, and the surrounding storage architecture that keeps expanding unless controlled.

Practical implication: inventory unused AI data and tie retention decisions to business use rather than storage convenience.

How storage design affects AI energy use

The article points to deduplication, tiering, and compression as efficiency levers because storage architecture shapes how much infrastructure must be powered and cooled. Deduplication removes repeated copies, tiering places data on storage aligned to its access frequency, and compression reduces the physical footprint of retained information. These controls do not change what AI is capable of, but they do change how much supporting infrastructure it needs. For security and platform teams, this is important because storage efficiency is now part of operational design, not just housekeeping. Inefficient storage multiplies cost, carbon, and recovery complexity across the whole data estate.

Practical implication: align storage tiers and duplication controls to data access patterns before scaling AI workloads further.

Why sustainability and resilience are linked in data governance

Redundant and poorly organised data is harder to secure, harder to govern, and harder to recover. That connection is more than a slogan: when data sprawl increases, backup scope grows, restoration becomes slower, and policy enforcement becomes less reliable. The article’s core governance point is that resilient systems depend on controlled data lifecycles, not just on more infrastructure. For NHIs and access governance, that intersection matters because broad machine and application access often accumulates around data stores, pipelines, and archives. Cleaner lifecycle control reduces both waste and the number of places where access and recovery controls can drift.

Practical implication: treat data lifecycle control as a resilience control, not only a sustainability measure.

NHI Mgmt Group analysis

Data lifecycle control has become a governance control for AI, not an operational afterthought. The article is right to separate AI value from AI overhead, because the largest hidden cost often sits in retention, duplication, and movement rather than in model execution itself. Once data is treated as a lifecycle-managed asset, sustainability and security start to converge. Practitioners should manage data estates as part of AI governance, not as a storage cleanup exercise.

Dark data creates governance debt that compounds across resilience, compliance, and infrastructure planning. If half of stored data is never accessed, then retention policy is not a clerical detail, it is a material control boundary. Unused data still needs protection, backup, and oversight, which means every unnecessary dataset expands the attack surface and the recovery burden. The practical conclusion is that lifecycle discipline should be measured as a programme control, not an infrastructure preference.

AI sustainability exposes the same control weaknesses that create identity sprawl in other domains. The more data estates expand, the more service accounts, pipelines, and delegated access paths accumulate around them. That is where identity and data governance intersect: uncontrolled storage growth tends to bring uncontrolled access growth with it. Teams should treat data minimisation as part of broader identity and access containment.

Efficiency measures only matter when they are tied to policy enforcement and usage telemetry. Deduplication, tiering, and compression are useful only if the organisation can prove that retained data still has a defined purpose and an accountable owner. Without that, the estate simply becomes smaller at the edges while still remaining poorly governed at the centre. The practitioner takeaway is to connect storage optimisation to retention review and ownership controls.

AI programmes now need a sustainability lens that is compatible with security architecture. The field is moving toward a world where carbon, cost, and governance are no longer separable concerns. That means the same decisions that reduce environmental waste can also reduce audit burden, recovery time, and control sprawl. Practitioners should build AI governance around data lifecycle discipline because the payoffs now extend across multiple risk domains.

From our research library:

• Organisations that describe themselves as confident in their AI deployment actually experience a 72% security incident rate, compared to 33% for those who remain cautious, according to the 2026 Infrastructure Identity Survey.
• Only 13% of organisations feel extremely prepared for the reality of agentic AI despite the majority racing toward autonomous adoption, according to the 2026 Infrastructure Identity Survey.
• Read next: Shadow AI and AI Agent Discovery Guide

What this signals

Dark data management is becoming a first-order AI governance issue. Organisations that let data accumulate without lifecycle control are also the ones that expand storage, backup, and recovery overhead the fastest. For practitioners, the useful shift is to treat retention policy as a design input for AI architecture, not a legal afterthought.

Data minimisation now has a direct resilience dividend. When teams cut redundant copies, shorten retention where justified, and tier infrequently used data, they reduce the amount of infrastructure that must be powered, protected, and recovered. That creates a cleaner operating model for AI and a smaller control surface for identity-heavy data platforms.

For practitioners

• Inventory dark data in AI pipelines Identify stored datasets that have no current business, training, or recovery purpose and classify them for retention review or deletion.
• Apply tiering to access frequency Move infrequently used data to lower-cost, lower-energy storage tiers and keep active AI training or inference data on appropriately sized platforms.
• Reduce redundant copies and backups Use deduplication and backup rationalisation to cut repeated storage of the same content across environments and recovery layers.

Key takeaways

• AI sustainability is increasingly driven by data lifecycle control, not by model scale alone.
• Dark data adds avoidable cost because it must still be stored, protected, and recovered even when it is never used.
• Teams should tie AI growth to retention, tiering, and deletion decisions so that efficiency and resilience improve together.

Key terms

• Dark Data: Data that an organisation has stored but no longer actively inventories, classifies, or governs. It may contain regulated information, credentials, or operational records, but the organisation lacks reliable visibility into where it lives, who can access it, and how long it should remain retained.
• Data Lifecycle: The data lifecycle is the sequence of stages data passes through, typically from acquisition to storage, use, transfer, retention, and disposal. Governance is only effective when controls and ownership are defined at each stage, because risk changes as data moves and is reused.
• Channel Tiering: A partner model that groups resellers or service partners into levels based on capability, investment, or performance. Tiering is used to shape incentives, set expectations, and grant access to benefits such as training, technical resources, or business planning support.
• Deduplication: Deduplication is the process of identifying repeated applicants or identities across programmes so the same person or entity is not approved multiple times without detection. It is a fraud and governance control that helps expose synthetic identity patterns, reuse, and hidden overlap across customer populations.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, workload identity, and secrets management. It is built for practitioners who need to connect identity control to wider security and operational decisions.