Storage sizing becomes part of agentic migration
AWS Transform for migrations now recommends a storage type and configuration for each server based on observed performance and capacity. The service presents expected cost before migration and can apply an approved configuration when the target launches. Recommendations cover Amazon Elastic Block Store and Amazon FSx for NetApp ONTAP, including file-system layout and sizing for the latter. Customers can choose an emphasis on cost or performance, and AWS says the recommendation remains consistent from assessment through launch. The update addresses a common source of both cloud waste and post-cutover performance problems.
Observed behaviour is better than allocated capacity
On-premises servers are often provisioned for historical peaks or organisational standards rather than current use. Copying allocated disk size and performance directly into the cloud can lock in unnecessary cost. Telemetry allows a recommendation to reflect actual throughput, IOPS, latency and capacity. The sample window must represent the workload. A month-end close, seasonal campaign or backup job may not appear in a short assessment. Teams should inspect the period, missing data and percentile assumptions before accepting a smaller target that looks efficient during ordinary days.
Cost and performance are not a simple slider
Choosing cost optimisation should not mean accepting an unspecified risk of latency. Storage has several dimensions, including capacity, throughput, IOPS, durability, snapshots and network constraints. An application may be sensitive to one dimension only during recovery or compaction. AWS Transform should explain which observations drove the recommendation and how close the proposed configuration sits to service limits. Decision-makers need the expected monthly cost and headroom, not just a product name. A recommendation becomes actionable when its assumptions can be challenged by the application owner.
EBS and FSx solve different problems
EBS provides block storage for individual compute instances, while FSx for NetApp ONTAP supports shared file-system capabilities and data-management features. Choosing between them can affect architecture, not merely price. An existing application may rely on shared namespaces, snapshots, protocols or replication behaviour that cannot be inferred from throughput counters alone. Migration discovery should include access patterns and operational requirements. Applying a technically efficient EBS configuration would be wrong if several servers expect shared file semantics; an elaborate FSx layout may be unnecessary for a simple local volume.
Automatic application needs approval and evidence
AWS Transform can apply approved settings at launch, reducing manual configuration drift across a migration wave. The approved recommendation should be versioned with the telemetry and cost estimate that supported it. If assessment data changes after approval, the system should show the difference rather than silently applying a new configuration. Infrastructure-as-code output or an equivalent export helps teams review and reproduce the target. Change records also make it possible to explain why one server received a different storage class from a similar neighbour.
Post-launch measurement closes the loop
A recommendation remains a hypothesis until the workload runs on AWS. Teams should compare latency, queue depth, throughput, errors and cost after cutover with the assessment forecast. If demand exceeds the selected configuration, an expansion path and rollback plan should already exist. If utilisation remains low, optimisation can continue after stability is established. The service should avoid aggressive immediate downsizing based on a quiet launch period. Historical context, business calendars and application owner input are necessary to interpret early metrics correctly.
Governance matters at fleet scale
Bulk recommendations can produce meaningful savings across thousands of servers, but a systematic assumption error can also affect an entire estate. Migration programmes should sample and independently review each workload class before applying changes widely. Policies can define minimum headroom, approved regions, encryption keys and backup requirements. Sensitive workloads may require customer-managed keys or specific replication regardless of the lowest price. A central exception process is preferable to hidden manual edits that make later audits and support difficult.
A practical use of AI-assisted infrastructure planning
Storage selection is a strong candidate for machine-assisted analysis because it combines large telemetry sets with a finite catalogue of cloud configurations. AWS Transform can shorten assessment and keep the approved choice attached to the migration plan. It should support, not replace, architectural judgement. The best result is a transparent recommendation that helps an owner make a faster decision and can be revised safely when reality differs. Used with representative data and post-launch verification, the capability can reduce both overprovisioning and the operational risk of undersized storage.
Teams should also examine how recommendation errors are surfaced. A storage choice that later hits a throughput ceiling needs an alert tied back to the original forecast, not an isolated performance incident. Recording actual versus predicted demand builds evidence for future migrations and can reveal a workload class that the model consistently underestimates. AWS can strengthen trust by making this feedback loop visible and by allowing customers to retain their own evaluation history when recommendations or service algorithms change.