The Fivetran-dbt merger marks more than just two companies combining forces. It’s the clearest signal yet that the “modern data stack” approach—the idea that you should assemble your data infrastructure from best-of-breed point solutions—has fundamentally failed.
For the past decade, the promise was compelling: choose the best connector platform, the best transformation tool, the best orchestrator, and the best BI layer. Mix, match, and customize to your heart’s content. Freedom! Flexibility! Innovation!
The reality? Tool sprawl became unmanageable. Integration nightmares multiplied. Costs became impossible to predict. And data teams spent more time maintaining their Rube Goldberg-esque infrastructure than actually delivering insights.
Now the market is correcting—hard. Microsoft has Fabric. Databricks has Lakeflow. Snowflake keeps expanding its capabilities. And Fivetran just merged with dbt in what one commentator called “scrambling to catch up” with the platform giants.
The modern data stack is dead. But what comes next? And how should data leaders navigate this transition?
To understand why it failed, we need to remember why it succeeded in the first place.
Before the modern data stack, enterprise data integration was dominated by monolithic, expensive platforms like Informatica, Talend, and IBM DataStage. These tools required:
For a typical data integration project, you’d wait weeks for hardware provisioning, months for implementation, and pay six-figure license fees. Innovation moved at the speed of enterprise procurement.
The emergence of cloud data warehouses (Redshift, Snowflake, BigQuery) changed everything. Suddenly you could spin up petabyte-scale analytics infrastructure in minutes, not months.
But you still needed to get data into these warehouses. The modern data stack emerged to solve this problem:
Fivetran and Stitch offered push-button connectors. Point, click, and your Salesforce data starts flowing to Snowflake. No custom code required.
dbt revolutionized transformation. Instead of writing stored procedures, you could treat transformations like software development—version control, testing, documentation, and CI/CD.
Airflow and Dagster provided open-source orchestration that you could customize infinitely.
Looker and Mode modernized business intelligence with code-based approaches.
The pitch was irresistible:
For a few years, it worked beautifully. Data teams felt empowered. Startup valuations soared. Conference talks celebrated the “composability” of the stack.
Fivetran raised money at a $5.6 billion valuation. dbt Labs hit $4.2 billion. The ecosystem was thriving.
But cracks started appearing. And they turned into chasms.
What started as “best-of-breed” devolved into “death by a thousand tools.”
A typical modern data stack in 2024 might include:
That’s 10+ vendors. Ten sets of contracts. Ten renewal cycles. Ten support organizations. Ten sets of monitoring and alerting to configure. Ten systems that need IAM integration.
The cognitive load on data teams became crushing.
The modern data stack promised that these tools would “just work together.” In practice, integration required constant manual effort:
Data engineers spent more time being “glue code developers” than actually building data products.
Usage-based pricing sounds great in theory. But when you have eight different vendors all with different pricing models, forecasting budgets becomes impossible:
Finance teams started rejecting data platform investments because the costs were impossible to predict or control. One viral tweet showed a $60K monthly Snowflake bill for a mid-sized company—most of it from inefficient queries nobody knew were running.
Every tool required ongoing care and feeding:
What was supposed to be “low maintenance SaaS” turned into a full-time job for multiple people.
One data leader put it bluntly: “We have eight data engineers. Three of them spend 80% of their time just keeping the stack running. That’s not what I hired them for.”
While modern data stack companies fought over their narrow domains, Databricks and Snowflake systematically added capabilities:
Databricks added:
Snowflake added:
Suddenly, the “best-of-breed” tools were competing against “free” (bundled) alternatives from the platforms customers already used. And “good enough + free + integrated” started beating “best + expensive + requires integration.”
The market is now consolidating rapidly. The Fivetran-dbt merger is just one example:
All under one unified experience with shared governance, security, and costs.
Recently announced, Lakeflow provides orchestration, data quality, and lineage natively within Databricks—eliminating the need for external orchestration tools.
Snowflake continues systematically building out capabilities to reduce dependencies on external tools:
Acquisitions are accelerating:
As one industry observer noted: “The entire data industry is consolidating. Everyone’s building bigger, more integrated platforms. The modern data stack’s original promise of mix-and-match tools is dead.”
The modern data stack is dead. But you still need to make decisions about your data infrastructure. You have three paths:
The Approach: Consolidate everything onto Databricks or Snowflake and use their native capabilities for ingestion, transformation, orchestration, BI, and ML.
Best For: Organizations that value simplicity over flexibility and have confidence in their chosen platform’s long-term direction.
The Approach: Use larger vendors that have consolidated multiple capabilities through M&A. Think Fivetran-dbt, or potentially future combinations.
Best For: Organizations with significant investment in these specific tools who want gradual evolution rather than rip-and-replace.
The Approach: Choose platforms designed from day one to provide end-to-end integration capabilities in a unified architecture—but without the platform lock-in of Databricks or Snowflake.
Best For: Organizations building or modernizing data infrastructure who want best-of-breed capabilities without best-of-breed complexity.
The explosion of AI use cases makes the infrastructure question even more critical.
Generative AI and agentic systems demand:
Real-time, Fresh Data: AI agents need current information to make decisions. Batch pipelines that refresh nightly don’t cut it.
Structured and Unstructured: LLMs need access to documents, images, and other unstructured data—not just structured database tables.
Governed and Traceable: When AI makes decisions affecting customers or compliance, you need bulletproof lineage and governance.
Context-Rich Metadata: RAG systems need rich metadata to retrieve relevant context. Fragmented metadata across tools makes this nearly impossible.
Low Latency: AI applications can’t wait minutes for data pipelines. Sub-second response times are expected.
The tool sprawl of the modern data stack creates critical problems for AI:
Metadata Fragmentation: Lineage lives in dbt, catalog in Alation, quality in Monte Carlo, and usage in your BI tool. Stitching this together for AI context is a nightmare.
Unstructured Data Gaps: Most modern data stack tools focus on structured data from databases and SaaS apps. Processing documents, images, and video requires entirely separate pipelines.
No RAG Native Support: Retrieval-Augmented Generation requires vector databases, embedding generation, and semantic search. Adding these capabilities to your existing stack means… more tools.
Governance Gaps: When AI needs to explain its decisions, you need end-to-end lineage from raw source through all transformations to the vector database. With data spread across multiple tools, this is nearly impossible.
Platforms designed for the AI era provide:
This isn’t something you can bolt onto a legacy modern data stack. It requires infrastructure designed with AI as a first-class citizen.
Here’s how to evaluate which path is right for your organization:
Tool Count Audit: List every tool in your data stack. If you have more than 5 vendors, you’re feeling the pain of the modern data stack.
Time Allocation Analysis: Where do your data engineers spend time? If more than 30% goes to “keeping the lights on,” you have a maintenance problem.
Cost Predictability: Can you forecast next quarter’s data platform costs within 10%? If not, you have a pricing model problem.
Change Velocity: How long does it take to onboard a new data source? If it’s weeks, you have an agility problem.
Must-Haves:
Deal-Breakers:
For Platform Giants (Databricks/Snowflake):
For Merged Point Solutions (Fivetran-dbt):
For Converged Platforms (Nexla, others):
Don’t just compare license fees. Include:
Direct Costs:
Indirect Costs:
Opportunity Costs:
One data leader calculated that their “free” Airflow deployment actually cost $400K/year when including the two engineers maintaining it.
A large insurer was onboarding new data partners (carriers, third-party administrators) using a traditional modern data stack: Fivetran for ingestion, dbt for transformation, custom Python for special cases.
The Problem: Each partner had unique data formats, delivery methods, and schemas. Onboarding took 6 months on average—mostly spent building custom mappings and transformations.
The Solution: Migrated to a converged platform with data product architecture. Partners’ data is automatically transformed into standardized Nexsets (data products) regardless of source format.
The Results:
An ecommerce company had the full modern data stack: Fivetran, dbt, Airflow, Snowflake, plus various data quality and cataloging tools.
The Problem: Data engineers spent 70% of their time on maintenance. New analytics use cases took weeks to implement. Business users couldn’t self-serve.
The Solution: Consolidated onto a purpose-built integration platform with self-service capabilities for business users.
The Results:
A fintech company built entirely on Databricks but felt vulnerable to vendor lock-in and pricing pressure.
The Problem: Their entire data infrastructure, ML pipelines, and analytics depended on one vendor. Negotiating renewals became increasingly difficult as they had no credible alternatives.
The Solution: Implemented a converged integration platform with vendor-independent data products. Can now run workloads on Databricks, Snowflake, or Trino without rewriting pipelines.
The Results:
The modern data stack failed because it promised flexibility but delivered complexity. The market is consolidating, and you need a strategy.
The Fivetran-dbt merger is a symptom, not a solution. Two companies that built separate products are trying to compete with purpose-built platforms by stapling their offerings together. It might help them tell a better story to investors, but it doesn’t solve the fundamental problems that killed the modern data stack.
You have choices:
The right choice depends on your specific context. But making no choice—just letting your current tools dictate your future—is the worst option of all.
The modern data stack is dead. What you build next will define your data capabilities for the next decade. Choose wisely.
In the News: Watch Nexla CEO Saket Saurabh analyze the Fivetran-dbt Labs merger with John Furrier on theCUBE + NYSE Wired, exploring its strategic impact on the evolving data ecosystem.
Fivetran and Nexla are leading data integration platforms, but they take different approaches. Learn how they compare on features, deployment, and governance to find the right fit for your data strategy.
Even Fivetran, the largest of the ELT vendors, isn’t even close to being a leader….
Webinar 
