What’s in today’s newsletter:

Cloud Database Insider — State of the Union 2025→2026

This edition of the newsletter is going to be very different from the usual important weekly news, and the deep dive I bring forth on a weekly basis.

As you may well be aware, I started working in the IT realm, back in the prime of the ”dot com” boom and subsequent bust. Back then, it was primarily a relational database world. It was the realm of Sybase, SQL Server, Oracle, DB2, and a bit of Informix. Client-server was the architecture of the day, and we had to worry about RAID, and placing our data files and log files on separate physical disks.

And to those readers of my vintage, do you remember tape backups, and off site storage?

Fast forward 25 years, and the database world is vastly different. I am no longer in the trenches worried about 50 different nightly backups, indexing, data loads via the reviled and detested SSIS packages (I’ve never met someone that liked them), and snarky Business Analysts that thought they knew more than me, and had the wisdom of creating views with 5 layers of nested views.

I work with the most talented people. I have discussions and contact periodically with a fair bit of the folks that work for the companies that I actually write about. I am also a still a practitioner, albeit I don’t get those calls 3 AM Sunday morning saying, “Gladstone, the London SQL Server is slow”. “Slow” is such a nebulous term.

My day-to-day is a bit more of a strategic sphere as opposed to an operational sphere.

I have to pour over hundreds of articles weekly to put together the newsletter. I don’t mind one iota as it makes me far more aware of the different aspects of the database world that helps tremendously with my actual 9 to 5 job, and putting the newsletter together, usually late on a Friday night.

With that preamble out of the way, here are the trends that I am seeing currently, whether I am working with them myself, or need to know about for myself and for the benefit of the readers and subscribers of the newsletter. Note that these are just my thoughts and I am not in this edition going to qualify them with links and loads of empirical evidence:

TL;DR

• Postgres Moves into the Data Platform

• An End to the Format Wars (Delta Lake vs. Iceberg)

• Data Platforms Become Application Hubs 

• Vector Search is Everywhere 

• Data Clean Rooms Go Mainstream 

• Smarter Economics for Cost and Performance

• Compliance and Sovereignty Drive Design

• The Decline of Bespoke Data Pipelines

• Graph Databases and AI Find Synergy 

• FinOps Moves Beyond Infrastructure to the Query Level 

1. Postgres Moves into the Data Platform The big shift for 2026 is bringing application databases (OLTP) directly into analytics platforms. Companies are adding serverless Postgres, allowing developers to build apps right where the governed data and AI models already reside. This simplifies the tech stack, speeds up development, and boosts security by reducing the need to move data and credentials between services.

2. An End to the Format Wars (Delta Lake vs. Iceberg) The rivalry between data formats like Delta Lake and Apache Iceberg is giving way to interoperability. You can now realistically plan for one physical copy of your data that can be used by many different engines. This approach dramatically cuts down on storage costs and operational overhead while making data governance much easier to manage.

3. Data Platforms Become Application Hubs Data warehouses and lakehouses are evolving from places you query data to places you run software. With support for containerized apps, AI agents, and marketplaces, you can now deploy tools and workflows directly on your data. This shortens the path from insight to action and simplifies security by keeping everything within a single, governed environment.

4. Vector Search is Everywhere For most AI applications, a separate, dedicated vector database is no longer necessary. Vector search is quickly becoming a built-in feature in warehouses, search tools, and even Postgres. The challenge is shifting away from managing infrastructure and toward better data practices, as search quality usually depends more on good data preparation than on a specialized engine.

5. Data Clean Rooms Go Mainstream Secure data clean rooms are becoming a practical reality beyond ad tech, expanding into finance, healthcare, and other regulated industries. These tools allow partners to collaborate on sensitive data without ever moving or exposing it. The key to success is treating it like a product: start with a clear goal, define shared metrics, and focus on delivering insights quickly.

6. Smarter Economics for Cost and Performance Serverless is now the default for data workloads, offering better performance and predictability. At the same time, new GPU advancements are changing the cost equation for AI tasks. The best strategy is to use CPU as the default and treat GPUs as accelerators for specific, high-impact workloads. The biggest savings, however, often come from reducing data copies and unnecessary data movement.

7. Compliance and Sovereignty Drive Design Data sovereignty is no longer an afterthought but a core requirement. Platforms now offer features like sovereign regions and customer-managed keys to satisfy regulators without building entirely separate systems. The goal is to bake compliance directly into your development process, turning audits into a routine, automated check.

8. The Decline of Bespoke Data Pipelines The tedious work of building custom ETL pipelines is shrinking. Modern tools favor serverless connectors and declarative pipelines with data quality and observability built-in. This frees up engineers to focus on delivering value instead of just managing the plumbing, resulting in more reliable data and faster insights.

9. Graph Databases and AI Find Synergy Graph databases, which map the complex relationships between data points, are becoming critical for fraud detection, customer-360 views, and risk analysis. The emerging trend is combining graphs with vector search ("GraphRAG"). This powerful duo provides AI models with context that is both factually accurate and semantically rich, leading to fewer AI "hallucinations" and more explainable results.

10. FinOps Moves Beyond Infrastructure to the Query Level As data platforms and AI workloads become major cost centers, FinOps (Cloud Financial Operations) is evolving. It's no longer enough to just track virtual machine and storage costs. The focus is shifting to granular, query-level cost analysis. Expect to see more tools and practices dedicated to understanding the financial impact of every query, optimizing AI model training expenses, and providing engineers with direct visibility into how their code translates into dollars spent.

It has been an absolute pleasure, creating this for the last year, as the readership and subscriber base has grown into the thousands.

I know at times the elephant in the room, AI, is all the rage right now. But keep in mind two things.

Firstly, the elephant ALWAYS needs data. Secondly, the database world is so vibrant and diverse, that I could write about it 5 days a week and still never run out of news to write about. Trust me on that one.

Look out for more improvements and engagement in the weeks and months to come. Have a productive and informed day. Thank you.

Gladstone Benjamin