Every data tool that died

Every tool on this chart was, at some point, someone’s recommended architecture. Most of them were good. Almost none of them exist as independent products anymore. Here is the last twelve years of data-tool exits, one dot per company, coloured by what actually happened to the product afterwards.

Read the colours first

Orange is the one to watch. “Absorbed” means the acquirer took the team and the technology and the product you were paying for stopped being a product. Sometimes it survives as a feature, like Dataform inside BigQuery or Transform becoming the dbt Semantic Layer. Often it just fades.

Blue is the gentle outcome: the brand keeps running under a new owner. Tableau, Looker, Segment, Mode. Prices change, roadmaps slow, but the login still works.

Red is rare because most companies do not shut a product down. They absorb it and let the docs rot. Chartio is the honest exception: Atlassian bought it, announced the shutdown, and gave customers a year.

Three waves

2018 to 2019: BI gets bought. Salesforce paid $15.7B for Tableau and Google paid $2.6B for Looker within the same month. The BI category, which had spent a decade as the top of the stack, became a feature of two clouds and one CRM. Periscope, Zoomdata, Numeracy, and Platfora all went in the same window. If you were selling dashboards in 2019, you were either acquired or about to be repositioned.

2021 to 2023: the modern data stack starts folding. This is the era of the ten-logo architecture diagram, and it is where the diagram started losing logos. Ingest tools (Alooma, Datacoral, HVR, Grouparoo, Arcion), transformation (Trifacta, Transform), catalogs (Okera, Stemma, Samooha), and the first wave of ML tooling all went to bigger platforms. Talend was taken private and then sold again within two years, which tells you what the private equity math looked like.

2024 to 2026: the engines eat everything. Switch to the “who bought” view. Databricks and Snowflake are running an acquisition every couple of months, and the things they buy are exactly the categories that used to be separate products: Postgres (Neon, Crunchy Data), feature stores (Tecton), catalogs, quality, observability (Observe). Then the two biggest independent pieces of the stack, Fivetran and dbt Labs, merged with each other, and IBM bought Confluent for $11B. There is no longer a “middle of the stack” to sell into.

What it means for the tool you are choosing now

The acquirer decides your roadmap, not the founder. If you adopt a standalone tool today, price in a change of control. Ask who the natural buyers are. If the answer is “the engine we already run on,” the risk is lower: the product will probably be absorbed into something you already pay for. If the answer is “a private equity firm,” expect the price to rise and the roadmap to freeze.

Open source changes the fate, not the exit. SQLMesh went to the Linux Foundation when Tobiko was bought. Redash and Superset are still forks away from any owner. If a tool is critical, prefer the one you can fork.

Categories collapse before companies do. Every category on this chart shrank from many products to a feature of two or three platforms. If you are adopting the fourth-best tool in a category with six players, you are picking a dot on next year’s chart.

Independence is a phase. Fivetran and dbt were the two most independent tools of the last decade, and they are one company now. That is not a failure. It is what winning looks like when the middle of the stack stops being a market.

The other reading

You could also read this chart as: the industry works. Ideas get proven by small companies, then rolled into platforms where more people use them. Dataform is used by more teams inside BigQuery than it ever was on its own. The semantic layer idea reaches more people as part of dbt than Transform reached alone.

That is true, and it is also cold comfort if you were the customer who picked the standalone version. Both things hold. Choose accordingly.