Amazon Web Services (AWS) has made a significant advancement in data management by embedding the open-source analytics engine DuckDB into its relational database, Amazon Aurora PostgreSQL. This integration, announced on September 30, 2026, allows users to query Apache Iceberg tables and Parquet files directly from data lakes without the need for traditional ETL (extract, transform, load) processes. As a result, both operational and historical data can be accessed seamlessly within a single SQL statement, all while avoiding additional licensing fees.
The implementation of DuckDB within Aurora PostgreSQL marks a transformative shift in how users interact with their data. Previously, accessing historical data stored in Amazon S3 for operational systems necessitated an ETL process to replicate that data into the operational database. By removing this requirement, users can now combine live operational data with historical data residing in the data lake in a single query.
To leverage this new feature, users simply enable the aurora_analytics extension on their cluster, assign an IAM role that grants access to S3 and AWS Glue, and create external tables that point to data lake files using standard PostgreSQL syntax. Notably, for Parquet files, the system automatically infers the schema from file metadata, even if certain columns are omitted from the table definition.
Expanding Data Lake Integration
This new capability supports four types of data sources: Iceberg tables managed by AWS Glue Data Catalog, Iceberg REST Catalog-compatible external catalogs connected via Glue federation, Parquet files on S3, and S3 Tables. On the performance front, predicate pushdown and column pruning are executed on the data lake side, minimizing the amount of data read, while frequently accessed data is cached on the Aurora instance. Users can monitor their queries through dedicated functions that provide insights into rows read, bytes read, and cache hit status.
AWS has identified several primary use cases for this feature, including the creation of real-time dashboards that integrate current and historical data, enriching transactional data with historical context, and enabling AI agents to perform cross-temporal reasoning over both current and past data. Esra Kayabali, a Principal Solutions Architect at AWS, emphasized that as AI agents become more integrated into applications, the challenge of predicting and pre-replicating the data they require becomes increasingly complex.
This feature is available for Aurora PostgreSQL versions 17 (17.11 and later) and 18 (18.6 and later), and it is accessible across all commercial regions as well as AWS GovCloud (US). Importantly, there are no additional charges for this capability; users are only responsible for the Aurora compute and S3 request fees incurred during usage.
The Major Contracts Surrounding AWS
In parallel with these advancements, AWS’s substantial contract with AI company Anthropic has garnered considerable attention. According to Anthropic’s draft IPO prospectus, the company anticipates booking 8 billion (approximately ¥81.9 trillion) in cloud, computing, and infrastructure-related contractual obligations over the next several years. Of this total, approximately 0 billion (around ¥17.4 trillion) is projected to be paid to Amazon between May 2026 and April 2036, stemming from a 10-year agreement announced in April 2026, which commits over 0 billion (approximately ¥15.8 trillion) to AWS as its primary training and cloud provider.
Anthropic’s revenue trajectory has been impressive, with first-quarter 2026 revenue reaching .73 billion (approximately ¥750 billion), surpassing the entire 2025 figure of .6 billion (approximately ¥730 billion). The second-quarter revenue soared to .5 billion (approximately ¥1.8 trillion), marking a year-over-year increase of more than 14 times. If this growth continues, the anticipated annual payment of roughly billion (approximately ¥1.6 trillion) to AWS seems achievable.
However, there are inherent risks. The draft prospectus indicates that about one-quarter of Anthropic’s 2025 revenue was derived from just two customers, and the company cautions that many of its larger clients are not bound by long-term contracts, leaving the potential for reduced or halted spending. This structure creates a scenario where AWS’s contractual obligations extend over a decade, while much of the revenue necessary to support them remains uncommitted.
Amazon serves as both Anthropic’s primary compute provider and an investor, having invested billion (approximately ¥2.8 trillion) to date. As of the end of June 2026, this stake is valued at approximately 0 billion (around ¥30 trillion), contributing positively to Amazon’s second-quarter 2026 net income. However, some analysts have raised concerns that this valuation may not accurately reflect the core business’s earnings potential.
