AWS re:Invent 2025 has spotlighted innovation in storage with a strong push for cost efficiency. With the addition of features like Intelligent-Tiering for Amazon S3 Tables and FSx Lustre, GPU acceleration for OpenSearch vector databases reduces cost by up to 75% with Database Savings Plans, offering up to 35% savings on services.
These improvements increase the storage capacity of RDS to 256 TiB and improve performance, allowing EFS scale-out to 2.5M IOPS, with a focus on automated tiering and more intelligent resource utilization, balancing innovation with financial foresight. 
Bonus
Download a PDF version of this blog. Access it offline anytime. Bring it to team or client meetings.
Additionally, cloud storage costs are revealed as the “silent villain” in AWS costs, building up over time due to missed snapshots, unattached EBS volumes, and inefficient tiering, despite contributing only 2% to average cloud bills. Recent reports indicate that up to 35% of overall cloud spend is wasted, with storage inefficiencies such as unused resources contributing as data volumes from AI/ML workloads and legacy migrations grow.
In this blog, we will highlight strategies that combine re: Invent 2025’s new services, such as S3 Intelligent Tiering, which can deliver up to 80% savings on tables, with foundational tweaks like rightsizing EBS and using Storage Lens metrics.
New & Enhanced AWS Storage Capabilities from re: Invent 2025
At AWS re: Invent 2025, Amazon boosted storage services for AI, HPC workloads and analytics, with key enhancements – S3 object size up to 50TB (10x increase); EMR Serverless auto-storage slashing spark job cost by 20%; RDS for SQL Server/Oracle to 256 TiB with 4x IOPS; and EFS scale out to 2.5M read IOPS plus intelligent-Tiering for FSx Lustre.
A) Amazon S3 Vectors – Radically Cheaper Vector Storage
Amazon S3 Vectors was released at AWS re: Invent 2025 and allows vector data to be stored and retrieved natively in S3 buckets, with significant cost savings, up to 90 percent, over traditional vector databases when uploading, storing, and querying large datasets.
You Might Also Like
S3 Vectors also supports real-time querying, providing sub-second response times for infrequent queries (<100ms) and ~100ms for frequent queries. Each index within S3 Vectors can contain up to 2 billion individual vectors, and each S3 bucket can contain up to 20 trillion unique vectors; write throughput for S3 Vectors is 1,000 IOPS per second.
Moreover, it is customised for RAG, semantic search, and AI agent memory, seamlessly integrating with Bedrock Knowledge Bases and OpenSearch without data movement, leveraging S3’s elasticity, durability, and zero provisioning.
Features include per-index customer-managed keys for multi-tenancy and metadata filtering up to 50 keys/vector, suited to cost-sensitive, large-scale use, knowledge refreshes, or hybrid setups with fast databases for hot data.
B) Choose Express One Zone for Warm Data
AWS S3 Express One Zone is a high-performance, single AZ storage class for warm data such as log file archives, medical imaging, and real-time analytics dashboards.
Offers up to 10x faster performance than S3 Standard with consistent single-digit millisecond latency through custom-built hardware that reduces network latency by bringing storage and compute resources closer together. Intended for performance-critical workloads such as analytics or interactive applications, where high availability in multiple AZs is not required.
For small businesses and large enterprises, it reduces costs by lowering API costs and scaling compute efficiently to millions of requests per minute without pre-provisioning. Natively supports S3 APIs for caching, big data processing (up to 4x faster Spark read/write times), and retains three internal copies for durability. Re: Invent 2025 sessions showcased its use cases in performance-critical applications.
C) Archive Deep with Purpose
S3 Glacier Instant Retrieval and Deep Archive storage classes are AWS storage options for infrequently accessed data, reducing costs by up to 95% and integrating seamlessly with S3 Lifecycle policies. Announced at re: Invent 2025, they enable automatic tiering of classes such as S3 Standard-IA as data ages, while delivering 99.999999999% (11 9s) durability across Availability Zones.
In addition, Glacier Instant Retrieval provides millisecond access for quarterly requests for medical records or media, matching S3 Standard-IA throughput at lower rates (with high retrieval fees).
Deep Archive offers the lowest costs for backups and compliance data retrieved in hours to days (for instance, 12-hour standard), including free bulk retrievals in Flexible Retrieval mode.
Cost Governance Tools That Reduce Storage Waste
Cost governance tools help organisations reduce cloud storage waste by automating the detection of underutilised resources, enforcing lifecycle policies, and streamlining tiering to lower costs through options such as cold storage. These tools offer visibility into spending patterns, prevent overprovisioning, and integrate governance rules to ensure compliance without manual intervention.
A) S3 Storage Lens & Metrics
S3 Storage Lens and Metrics provide powerful AWS-native solutions to identify and reduce storage waste in Amazon S3 buckets. By providing a centralised dashboard with over 60 metrics on usage patterns, cost drivers, and activity trends, it visually highlights “money drains” such as small objects, unused prefixes, low-access data, and incomplete multipart uploads that inflate bills without value.
Furthermore, it is much like a fridge thermometer reveals expired items before they spoil further; S3 Storage Lens highlights inefficiencies via interactive views sliced by region, prefix-level granularity, and bucket, including 15-month historical trends in advanced mode.
Users can export metrics in CSV or Parquet for deeper analysis, apply actionable recommendations (like lifecycle policies to transition to cheaper classes such as Glacier) and clean unused versions or cold data to realise immediate savings, free for basic features.
B) AWS Compute Optimizer & Cost Explorer
AWS Compute Optimiser and Cost Explorer are powerful AWS-native tools for tackling storage waste, especially by automating the cleanup of unattached EBS volumes and state snapshots.
Compute Optimiser evaluates usage patterns and includes automated EBS optimisation features that identify unattached volumes that have been idle for 32 days, create snapshots before deletion to prevent data loss, and recommend upgrading from older types, such as gp2, to cost-efficient gp3. Cost Explorer complements this by offering granular visibility in storage costs, pinpointing high-spend areas such as underutilized volumes via filters and forecasts.
To incorporate, enable Compute Optimiser in the AWS console, configure automation rules in its EBS section, and set daily, weekly, or monthly schedules to automatically snapshot and delete unattached volumes or upgrade volume types based on your criteria.
Pair with Cost Explorer’s resource optimization reports to export CSV data comparing current vs recommended monthly prices, quantifying savings (20-30% on storage). Regularly review these tools via AWS Organisations for multi-account governance, ensuring continuous trimming of storage bloat without manual intervention.
Real-World Tactics That Work
AWS re: Invent 2025 highlighted practical strategies for optimising storage costs for cloud users. Key tactics involve right-sizing EBS volumes by migrating from gp2 to gp3 (up to 20% savings) and leveraging elastic volumes to adjust performance. Using S3 Intelligent Tiering and lifecycle policies for automatic data tiering to infrequent access classes, yielding 35%+ reductions as seen in real workloads.
A) Audit Data Access Before Tiering
Auditing data access before implementing storage tiering helps ensure cost-effective decisions by identifying frequently accessed vs. infrequently accessed data. Using Amazon S3 Storage Lens gives you better insight into bucket access trends and helps reduce unexpected transfers that could incur data retrieval fees or result in lost savings.
To start, enable S3 Storage Lens metrics and recommendations for your target buckets to generate reports on historical access patterns, geographic trends, and storage class usage.
Then analyse the reports to determine whether your data is hot (often accessed), cool (occasionally accessed), or cold (seldom accessed). Subsequently, manage it according to this distribution using S3 Lifecycle Policies or Transition Unpredictable Data to Intelligent Tiering.
B) Consolidate Logs & Move to Cheaper Classes
Consolidating logs in a central storage system, such as Amazon S3, and transitioning them to cost-effective Glacier tiers reduces costs for high-volume logging in cloud environments. Set retention polices to 7-90 days for hot data in S3 Standard, then automate transitions through lifecycle rules – e.g., move logs older than 30 days to S3 Glacier Flexible Retrieval or Deep Archive, costs under $1/TB/month.
This tactic typically works in practice for services such as CloudTrail or app logs: compress files (for instance, tar.gz), archive them after 180 days to Glacier, and explore them after 1-3 years to avoid indefinite storage bloat. Tools such as AWS Lambda or lifecycle policies automate tasks, saving 75-90% on costs while enabling rare restores for audits.
C) Tag Everything (Yes, Everything)
Tagging each cloud resource allows precise chargeback by assigning cost directly to responsible teams, promoting accountability without stifling innovation. This tactic, rooted in FinOps practices, needs extensive labelling like team names, projects, owners, and environments – applied consistently from resource creation.
Automation enforces mandatory tagging through policies, while regular audits catch inconsistencies such as case sensitivity or misspellings, ensuring accurate cost allocation reports. Teams gain visibility into spending patterns, enabling budgeting, forecasting, and optimization discussions that drive efficiency across departments.
D) Use FSx Compression Where Applicable
Amazon FSx for Lustre now supports LZ4 data compression. This allows FSx for Lustre to automatically compress data written to disk and decompress data read from disk. Thus, FSx for Lustre can achieve a compression ratio of up to 3.75:1 for compressible workloads such as machine learning datasets and logs. This minimizes storage needs by 70%+ in tests, capacity – up to 2x faster reads/writes.
Enable it on new or existing FSx Lustre systems via the AWS Console or CLI; it’s best for HPC video processing or analytics, where data compresses well, and performance cannot lag. Uncompressible data (e.g., encrypted files) sees minimal gains, so test your workload first to assess ROI.
Think Beyond S3 – Optimize Related Storage Costs
Traditional reliance on Amazon S3 Standard results in higher storage costs for infrequently accessed data in cloud engineering workloads. “Think Beyond S3” advocates exploring intelligent tiering, lifecycle policies, and alternative classes such as Glacier or Deep Archive for automatic optimization.
Utilizing this methodology adheres to enterprise-wide recognized best practice methods and allows you to save as much as 95% on infrequently accessed data without any loss in performance.
A) Database Savings Plans
Database Savings Plans allow AWS users to save up to 35% on managed databases such as RDS, DynamoDB, Aurora, and ElastiCache by committing to consistent hourly spend over one year – no upfront payment is required.
Unlike rigid Reserved Instances, they flexibly apply discounts across engines, instance types, models (provisioned or serverless), and regions, adapting to changing workloads. While not focused on storage, they cut costs like snapshots and backups stored in S3.
To implement, use AWS Billing Console’s Recommendations View or Purchase Analyzer to simulate commitments from historical data, aiming for high coverage without overcommitment. Monitor via Cost Explorer and adjust as required; excess usage reverts to on-demand pricing. This pairs well with S3 strategies to streamline hybrid database storage costs for small businesses.
B) Serverless Analytics (EMR Serverless)
EMR Serverless transforms Apache Spark jobs by eliminating the need for manual local storage provisioning, which requires guessing disk sizes for shuffle, spill, and caching operations.
It offers fully managed, elastic serverless storage that automatically scales for intermediate data processing, decoupling storage from the compute. This allows Spark workers to scale down instantly when idle, avoiding overprovisioning waste and disk-related failures.
The benefits involve up to 20% lower data processing costs – paying only for compute and memory, not idle disk capacity – plus improved reliability via even data distribution that minimizes skew and node failures. Encrypting data at rest and in transit ensures its security. Available in EMR releases 7.12+ and above, it excels at dynamic serverless analytics on AWS.
Cost Monitoring + Guardrails
AWS Budgets enables precise cost monitoring by letting you set custom budgets and alerts directly in AWS Billing, tailored to services, accounts, or tags for granular control.
Pair it with AWS Cost Explorer’s forecasting features to predict future spend based on historical trends, usage patterns, and ML-based insights, helping prevent surprises from seasonal demand or unexpected resource growth.
The above configurations serve as a guardrail by sending notifications to users via email, SNS, or automated IAM actions when the cost of accessed items approaches certain thresholds, giving users an opportunity for proactive intervention.
Furthermore, aligning with FinOps best practices also helps develop this practice into a team discipline by promoting monthly cadences, reviewing the actual vs projected expenditures, and adjusting budgets dynamically (i.e., 5% growth each month) and performing collaborative anomaly searches using Cost Explorer visualizations.
Integrate guardrails, such as service quotas and automated scaling policies, to enforce limits and promote accountability across finance, leadership, and engineering for sustainable cloud economics.
Conclusion
AWS re: Invent 2025 makes one thing clear – storage cost optimization is no longer about cutting corners; it is about using smarter controls. Begin with a storage health audit this week and turn those insights into well-defined tiering and lifecycle policies that automatically move data to the most cost-efficient storage classes.
Blend this approach with database savings plans and EMR Serverless, where applicable, to minimize idle capacity cost while maintaining scalability and performance.
Moreover, re: Invent 2025 provides tools you can use to tighten your bill without hurting overall performance. Organizations that act early will see immediate savings and long-term efficiency gains across this cloud stack.
As a leading web development and cloud solutions company, we focus on helping businesses evaluate, streamline, and implement cost-efficient AWS architectures so you can save more while building smarter and faster.
