. -->

AI needs the best-of-breed data stack: Postgres and ClickHouse | ClickHouse<br>Skip to content

Open searchOpen region selectorEnglish<br>Japanese

48.6kSign inGet Started

->Scroll to top<br>BackBlog<br>Product<br>Copy pageCopied!More actionsView as Markdown Open this page in Markdown<br>Open in ChatGPT Ask questions about this page<br>Open in Claude Ask questions about this page<br>Open in v0 Ask questions about this page

AI needs the best-of-breed data stack: Postgres and ClickHouse

Sai Srirampur<br>Jul 7, 2026 · 10 minutes read

Over the past few years working with thousands of companies on Postgres and ClickHouse, I've watched AI change what applications demand from their databases. Every AI application is a data application: agents generate huge volumes of operational data, and users expect real-time answers built on that data, often in the same request. That changes what applications demand from their databases.

In this post, I want to explore what AI applications actually need from their data layer, why Postgres and ClickHouse are increasingly becoming the natural combination for those workloads and our investments in making this combination very accessible to developers.

TL;DR AI is accelerating data growth at an unprecedented scale and is collapsing the traditional divide between transactional (OLTP) and analytical (OLAP) workloads. Meeting these demands requires best-of-breed databases, Postgres and ClickHouse, each excelling at what it was designed to do, while still working together seamlessly.

What AI needs, and how it's changing the data landscape #

AI is pushing the data layer to new limits #

AI-native applications generate far more data than traditional software. Every prompt, response, tool call, evaluation, and user interaction becomes data, driving explosive growth in both data volume and query concurrency.

Here is a data-point showing the scale I am talking about. Across a sample of AI-native companies using Postgres and ClickHouse, we observed an average data increase of more than 1,000% over six months , adding over 85 TB of data .

Also, as AI moves beyond vibe-coded apps into production systems, the tradeoffs are changing. Applications are making mission-critical decisions and operating across a larger surface area and at greater scale, raising the bar for reliability and security at the data layer.

To keep up with these demands, AI needs a data layer that can ingest, process, and query massive volumes of data in real time, always available and secure. Fast provisioning and instant branching are nice, but they matter little if the underlying data stack isn’t fast, reliable or secure.

AI needs OLTP and OLAP working together in real-time #

LLM-powered applications, AI-generated insights, anomaly detection, recommendation engines, and natural language interfaces all demand a much tighter feedback loop between transactional and analytical databases. Data written to a transactional database must become available for analytical queries almost immediately, whether to evaluate a potentially fraudulent transaction or diagnosing a production incident.

In these scenarios, the traditional pattern of moving data into a warehouse through hourly or minutely ETL is no longer sufficient. Transactional and analytical databases must operate in real time, with only seconds separating when data is written from when it is available for analysis.

Open source foundations and costs matter #

A similar shift is underway in AI infrastructure. As teams grow more wary of proprietary LLM lock-in and mounting token costs, they're also rethinking the flexibility and economics of their data stack. Open-source databases solve both: they offer self-hosted or managed deployment, the freedom to switch vendors as needs evolve, and often significantly lower costs.

Why Postgres + ClickHouse is winning #

The above requirements are reshaping the data stacks AI-native companies choose. Postgres and ClickHouse are increasingly becoming the default choice, combining best-in-class real-time transactions and analytics on a strong open-source foundation.

Let's understand on why this is happening.

Postgres and ClickHouse are best-of-breed open source databases #

Postgres , on the one hand, is the most popular open-source transactional database in the world. It offers rock-solid support for transactions, including low-latency CRUD operations, ACID compliance, rich indexing, a full SQL interface that supports virtually any query, a robust extension framework, and a vast ORM and application ecosystem, making it the default backend for web and AI applications everywhere.

ClickHouse , on the other hand, is the fastest analytical database on the planet. It offers purpose-built capabilities including columnar storage, skip indexes, incremental materialized views, distributed cache, native JSON and full-text search, the recently introduced lazy materialization, and hundreds of optimizations, laser-focused on blazing-fast real-time...