Data Stack Consolidation: Why Platforms Want Your Data Pipelines

The data stack consolidation of 2025 and 2026 is usually told as a story about money and market share. Read the announcements of 5 of its deals one after the other, though, and another word keeps coming back.

• Salesforce bought Informatica to build "a critical foundation for autonomous AI agents"
• Microsoft bought Osmos to extend Fabric with "agentic AI for data engineering"
• IBM bought Confluent so that enterprises could "deploy generative and agentic AI better and faster"
• Fivetran and dbt Labs merged to create "the data infrastructure for trusted AI agents"
• SAP bought Dremio "to power agentic AI"

The word is agents. Read together, these announcements say more about where data engineering is heading than any of the deals taken one by one.

The second rebundling of the modern data stack

Data stack consolidation means that the independent tools of the modern data stack, from ingestion and transformation to streaming and query engines, end up inside a handful of large platforms. This has happened before, one layer up. In June 2019, Google announced it would buy Looker, and 4 days later Salesforce announced it would buy Tableau. 2 of the best-known independents in business intelligence joined much larger platforms.

At the time, the layer below the dashboards was going the other way. The modern data stack was a deliberate unbundling, with one tool to move the data, another to transform it, a warehouse to store it and an orchestrator to schedule the lot, each chosen on its merits and wired together by the team.

The deals of 2025 and 2026 rebundle that lower layer: the pipes that carry data from its sources to whoever uses it. In 4 of the 5, the buyer is a platform or application company that does not sell pipelines as its main business. The fifth brings together 2 of the most prominent independents of the modern data stack, which is what independents tend to do when the platforms start buying.

Why AI agents made data pipelines worth buying

A dashboard shows data to a person. When a number looks odd, someone usually notices and asks a question before anything happens. An agent acts on the data. It answers the customer or reorders the stock, and there may be nobody in between to notice.

When the consumer is an agent, the choices made inside a data pipeline stop being technical details. Which records were dropped, how a missing value was filled, which duplicates were merged, what a status code means: all of these become assumptions the agent works from without knowing it.

Take a field as plain as a customer's status. Somewhere in a pipeline, someone decided that an account with an invoice unpaid for 90 days counts as churned. A person reading a dashboard may know that rule, or ask about it. An agent drafting retention offers will simply apply it, to every account, every night.

Whoever controls the path the data takes to the agent therefore controls a large part of what the agent will do. This is what the buyers have understood, and on this point we think they are right. For years the pipeline was treated as plumbing, invisible and priced by the row. Once agents consume the data, the pipeline is where a company's reading of its own data gets decided.

Data stack consolidation risks: from failure to alignment

With independent vendors, the risk a data team managed was failure, the start-up that ran out of money or the tool nobody maintained any more. These buyers will not run out of money.

The new risk is alignment. More and more often, the ingestion and transformation layer belongs to a company whose first priority is its own platform and its own agents. Roadmaps and price lists follow priorities. That kind of risk shows up slowly, in a connector that stops getting new features or in a migration guide that points towards the parent platform. Osmos gave an early sign of the direction: its standalone product began winding down in January 2026, and its technology now lives inside Fabric.

Vendor neutrality, which the independents used to sell as a feature, is becoming scarce. A pipeline built inside a platform will increasingly be a pipeline built for that platform's agents.

How data teams should respond: rent the engines, own the decisions

Our position follows from the buyers' own reasoning. If the pipeline decides what the agent believes, it is too important to rent from the company that sells the agent.

The platforms have a fair argument on their side. An agent that reads data prepared by the same vendor, under the same governance, is easier to secure and to support, and a single contract is simpler than several. For the engines, we agree. For the rules that say what the data means, that closeness becomes a problem, because the nearer those rules sit to one vendor's agents, the harder they are to take elsewhere.

Owning the pipeline is therefore compatible with using the platforms. Their engines and their agents are often the best on the market, and data teams should use them. What stays in-house is narrower: the code that ingests and transforms your data, and the rules that decide when a data pipeline is production-ready. That layer belongs in your repository, where your engineers can read it and where it can run on the next platform as well as on this one.

A few years ago, this position was expensive. Owning pipelines meant writing and maintaining every connector by hand, and renting was the sensible choice for most teams. The balance has moved since then. Generated code has made owning cheaper, while each acquisition makes renting a little riskier. Waiting has a cost too, because vendor lock-in is built one connector at a time.

What consolidation means for data engineers and consultancies

For data engineers, the work is moving towards the very layer the platforms want to own. Fewer hours will go into writing connectors, and more into deciding what the data means and proving that the pipeline respects it.

For consultancies, the shift opens a door at a time when the IT services model based solely on daily rates is already under pressure. When vendors stop being neutral, the neutral party in the room becomes whoever builds the pipeline for the client. A consultancy that hands over a decision layer the client owns, and can carry from one platform to the next, offers the one thing a platform cannot sell, which is independence.

Before you hand your data pipelines to a platform

Over the next year, every platform will offer to take the pipelines off your hands, with an agent attached. The offer will often be good, and it will always be convenient.

Before accepting, ask one question: if this pipeline encodes how my business reads its own data, who should own it?

We think the answer is you. Data stack consolidation is one of the data engineering trends for 2027 already underway, and it is the one that makes this question urgent.

PrettyWhale.ai is an Engineering AI for ingestion code. It generates data pipelines as code in your repository, with their tests and documentation, so the layer the platforms are buying stays yours.