At re:Invent 2025, AWS announced a series of major updates to its core storage service, Amazon S3, for the AI and big data era. Among these updates is support for native vector storage and querying.Amazon S3 VectorsThe official release further increases the maximum storage size of a single Amazon S3 object to 50TB, thereby helping businesses build generative AI applications and data lakes more efficiently and cost-effectively.
Amazon S3 Vectors officially launched: capable of indexing 20 billion vector sets, with costs reduced by 90%.
To enable AI systems to be stored directly in the native Amazon S3 environment and to perform natural semantic searches by querying vector data, AWS announced the official release of Amazon S3 Vectors.
Compared to the previously released preview version, the official version represents a significant leap in scale, including the ability to scale a single index to 20 billion vector data (a 40-fold increase in capacity), and a single bucket can support up to 20 trillion vector data.
AWS emphasizes that Amazon S3 Vectors significantly improves performance by 2-3 times for frequent queries, helps customers reduce costs by up to 90% compared to other alternatives, and eliminates the complexity of managing dedicated vector infrastructure.
This feature also deeply integrates with Amazon Bedrock Knowledge Bases and Amazon OpenSearch Service, enabling businesses to easily build AI Agents and RAG (Retrieval Augmentation) systems. Currently, customers including the BMW Group, Japanese social networking site mixi, and voice communication technology provider Twilio have already adopted it.
The maximum size of a single item has been increased to 50TB, meeting the needs of AI training and high-definition audio-visual content.
As data volumes grow exponentially, AWS announced that it will increase the maximum storage size of a single object in Amazon S3 from 5TB to 50TB, a tenfold increase, allowing users to store larger single data objects.
This update brings greater advantages to industries that handle massive archives, meaning that there is no need to cut files due to storage format limitations. It allows for the direct storage of ultra-high resolution video, massive seismic exploration data, and even astonishing AI training datasets in their original formats. Furthermore, it simplifies workflows while retaining all the storage categories and functions of Amazon S3.
S3 Batch Operations improves processing efficiency by 10 times, and S3 Tables supports intelligent tiering.
For large-scale data processing, Amazon S3 Batch Operations has also received performance upgrades, with speeds up to 10 times faster at executing large batch jobs (such as cross-region copying, tag management, or checksum verification), and a single job size that can handle up to 200 billion items.
In addition, Amazon S3 Tables, designed for Apache Iceberg workloads, has added two key new features:
• Supports Intelligent-Tiering:It can automatically move data between three access levels based on the access mode, saving up to 80% of storage costs without affecting performance.
• Automatic Replication:It supports automatic form copying across regions and accounts, enabling teams located in different places to access local data to improve performance, while simplifying compliance and backup management.
FSx for NetApp ONTAP data seamlessly integrates with the Amazon S3 storage ecosystem
Finally, AWS has expanded the functionality of Amazon S3 Access Points to include support for Amazon FSx for NetApp ONTAP. This integration means that customers can access files in FSx as easily as they would access data stored on Amazon S3.
Enterprise users can now directly utilize AI and analytics services originally designed for Amazon S3, such as Amazon SageMaker, Amazon Athena, or Amazon Bedrock, to process data stored in FSx for NetApp ONTAP, significantly lowering the application threshold after migrating NetApp ONTAP data from the ground to the cloud.
Bonus Feature: EC2 Hardware Upgrade and Database Service Update
In addition to storage services, AWS also released several updates for computing, databases, and modern tools at re:Invent 2025:
Computational Examples and the Evolution of Serverless Computing:
• Amazon EC2 X8aedz instance:Equipped with a 5th generation AMD EPYC server processor, it offers processing speeds up to 5GHz and 3TB of memory capacity, designed for workloads requiring high single-threaded performance, such as electronic design automation (EDA) and memory-intensive databases.
• AWS Lambda Managed Instances:By combining the simplicity of serverless computing with the flexibility of EC2, Lambda functions can run on EC2 computing resources, facilitating access to special hardware and optimizing costs.
• AWS Lambda Durable Functions:It allows the creation of applications that can coordinate multiple steps and operate continuously (from seconds to a year) without incurring additional idle computing costs while waiting for external events.
Database and search optimization:
• Database Savings Plans:Introducing a new pricing model for AWS Repository services, offering more flexible cost optimization options.
• Amazon OpenSearch Service acceleration:By incorporating GPU acceleration and automatic optimization features, the speed of building large-scale vector databases can be increased by 10 times, and the cost can be reduced to a quarter.
• RDS for SQL Server/Oracle:Improve scalability and optimize costs, including support for Microsoft SQL Server Developer Edition and Oracle M7i/R7i instances.
Application modernization and management:
• AWS Transform:The company launched a Custom feature that uses AI to learn organizational patterns to automate code transformation, and also announced modern features for all Windows and mainframe platforms to help enterprises eliminate technology debt.
• Amazon EKS Capabilities:Simplify the development process of Kubernetes containerized applications and provide fully managed workload orchestration and cloud resource management.
• Amazon CloudWatch:It introduces unified data management and analysis, and supports OCSF and Apache Iceberg formats.
Network and Partners:
• Amazon Route 53 Global Resolver (Preview):It provides unified hybrid DNS management, resolving public and private domains globally through Anycast technology.
• AWS Partner Central:It is now directly integrated into the AWS Management Console, simplifying the solution management process for partners.
