A few years ago, the "modern data stack" meant assembling a dozen specialist tools. The picture in 2026 is calmer and more pragmatic: consolidation, open formats and AI features built into the platforms teams already use. Here are the trends worth understanding, and what they mean in practice.
1. Open table formats go mainstream
Apache Iceberg has become the leading open table format: a way of storing tables as files in your own cloud storage while keeping warehouse features such as transactions, schema changes and time travel. Snowflake, Databricks, Google, AWS and others all support it in some form.
Why it matters: data can stay in one open format while different engines query it. That reduces lock-in and duplication, and means your choice of query engine is less permanent than it used to be.
2. Local-first analytics with DuckDB
DuckDB has become a favourite tool for analysts and engineers: an in-process analytical database that queries CSV, Parquet and Iceberg data at impressive speed on a laptop. Paired with services like MotherDuck, it also scales into the cloud.
Why it matters: many workloads that once needed a warehouse can run locally and cheaply, which is great for development, testing and small organisations. See choosing the right database.
3. Faster tooling for analytics engineering
dbt Labs introduced the dbt Fusion engine in 2025, a rewrite of dbt's engine in Rust that understands SQL rather than treating it as text. The promise is much faster parsing, and editor features such as live error checking and column-level lineage as you type. In October 2025, dbt Labs and Fivetran also announced plans to merge, a sign of consolidation across the ingestion and transformation layers.
Why it matters: analytics engineering is converging with mainstream software development: fast feedback, strong tooling, and fewer separate vendors to manage.
4. Postgres moves next to the warehouse
In 2025, Snowflake agreed to acquire Crunchy Data and Databricks agreed to acquire Neon, both PostgreSQL specialists. The major data platforms now want to host the transactional databases behind applications, not just analytics.
Why it matters: the boundary between operational and analytical data is blurring. Expect simpler paths from application data to reporting, and more "data apps" built directly on the platform.
5. AI inside the data platform
Warehouses now ship AI features natively: Snowflake's Cortex AI functions, for example, let you call large language models from SQL to classify, summarise or extract information from text, and natural-language assistants let business users ask questions of governed data.
-- Illustrative: classify customer feedback with an AI SQL function in Snowflake
SELECT
feedback_id,
feedback_text,
SNOWFLAKE.CORTEX.SENTIMENT(feedback_text) AS sentiment
FROM raw.customer_feedback;Why it matters: unstructured text becomes something you can model and report on with SQL. It also raises the stakes for governance: AI answers are only as good as the definitions and data quality behind them.
6. The semantic layer matters more than ever
When people, and now AI assistants, ask "what was revenue last quarter?", someone has to define "revenue" exactly once. Semantic layers and metrics definitions, such as those in dbt, Power BI models and warehouse-native semantic views, are becoming the foundation that makes self-service and AI trustworthy.
What this means for teams
- Invest in fundamentals: clean models, clear definitions and tests matter more with AI, not less. Start with the star schema and data quality tests.
- Prefer open formats where practical, to keep future options open.
- Consolidate tools: fewer, better-integrated tools are easier to govern and cheaper to run.
- Try AI features on governed data, with clear definitions, before rolling them out widely.
This field moves fast. Product names, features and deals change; check vendors' official announcements for the current status before making decisions.
Written by Alessandro Ecclesie Agazzi, freelance analytics engineer in London. Updated 30 September 2026.