Fivetran is the tool people mean when they say “we don’t want to think about pipelines”. You connect a source, pick a destination, and it keeps the data flowing while someone else worries about API changes. For a decade that pitch built the biggest name in managed ingestion.
Then October 2025 happened: Fivetran and dbt Labs merged, in an all-stock deal Reuters put at close to $600 million in combined annual revenue. The two most-recommended tools in the modern data stack are now one company. That changes how I’d evaluate both, so let’s take it from the top.
The short version
Buy Fivetran when connector reliability matters more than the bill. Its 700+ connectors are maintained by the vendor, schema changes are handled for you, and the compliance list (SOC 1 and 2, ISO 27001, HIPAA, PCI DSS) gets it through procurement at companies where Airbyte’s community connectors would raise eyebrows. The trade: you pay for rows you don’t control, and the price is a moving target by design.
Monthly active rows, the meter that runs the show
Fivetran doesn’t sell seats or connectors. It counts monthly active rows: every row inserted, updated or deleted across your connections in a month, billed per million on a sliding scale. There’s also a $5 base charge on every standard connection using between 1 and 1 million rows a month, which quietly adds up if you have a long tail of small sources.
The free plan is real and worth taking seriously: 500,000 monthly active rows for connections, 3,500 for activations and 5,000 model runs, with Standard-plan features. A small team with a few steady sources can sit inside it for a year.
The problem starts when you model the paid bill, because you don’t fully control the meter: it counts what changed in your sources, and your sources answer to nobody’s budget. The bill can grow faster than the business, and predicting it means understanding the counting rules, so let’s do that properly. Run their estimator with your real row counts before you commit, and price the same workload on Airbyte while you’re at it. Annual contracts knock off up to 22%, which is genuine money, but it locks in a volume assumption you may not be able to hold.
What you get for the premium
Fairness requires saying this part plainly: the product is excellent at its job. Setup is minutes per connector. Sync frequency is 15 minutes on Standard and down to 1 minute on Enterprise. Schema migration is automatic, which sounds small until a source renames a column at 2am and nothing breaks. The Standard plan includes unlimited users, a REST API and the dbt Core integration.
Higher tiers exist for a reason too. Enterprise adds the database connectors big companies actually need (Oracle and the high-volume agents) and Business Critical buys data residency options for the strictest regimes. Both unlock the feature that surprised me most in this research pass: Hybrid Deployment. An agent runs inside your own network on Docker or Kubernetes, pipelines process your data there, and only metadata, sync metrics and MAR counts travel to Fivetran’s control plane. The old objection that Fivetran means your data transits someone else’s cloud is now a plan-tier question rather than an architecture fact, and for banks and healthcare that changes the shortlist.
How the counting actually works
Credit where due: Fivetran’s docs are unusually clear about the mechanics, and the design is fairer than folklore suggests. A monthly active row is a distinct primary key synced in a calendar month, counted once no matter how many times that row updates. A row that changes thirty times in March is one MAR in March. Where the source has no primary key, Fivetran hashes one so the counting stays honest.
The subtleties that decide real invoices sit around that rule. Keys are counted separately per table, per connection and per destination, so the same customer record flowing through two connections is two rows. Initial syncs generate historical MAR (the whole table, once), which is why the first month always spikes. And the docs themselves flag the pattern that ruins budgets: sources that modify years-old records on every sync effectively re-import history month after month. Free MAR exists too: the free plan’s allowance and each new connection’s 14-day trial don’t count towards paid usage.
What this means practically: audit which of your tables churn old rows before you estimate anything. A big table where only recent rows change is cheap under MAR; a modest table where a batch job touches every historical row monthly is expensive forever.
A worked example of the meter
Say you sync a Postgres database, Stripe, HubSpot and Google Ads. The database rewrites a 2-million-row orders table nightly because the source system updates a status column: that alone is 2 million active rows a month even though your business didn’t grow. Stripe and HubSpot are steady at 300,000 rows between them. Google Ads is small, maybe 40,000 rows, but it still triggers the $5 base charge like every standard connection over 1 million total does.
You’d cross the 500,000-row free limit in the first week of the month, and the part driving the bill would be a status column nobody looks at. This is why the estimator matters more than the rate card: the fix is often changing what you sync (exclude the churn-heavy tables, sync incrementally where the source allows it) rather than changing vendors. Fivetran lets you block schemas and tables from syncing, and using that feature well is the single biggest lever on the invoice.
Who should skip it
Three profiles, honestly. Teams with one big database and nothing else: you’re paying a managed-SaaS premium for a problem a single well-monitored sync could handle. Teams with unusual or internal sources: Fivetran builds connectors for the mass market, and your ticket will wait. And anyone who can’t get a credible answer on renewal pricing: consumption contracts are cheapest in year one by design, and if sales won’t discuss year two, that tells you the plan.
The bigger picture: an acquisition streak
The dbt Labs merger reads differently once you line up the history. HVR in 2021 brought enterprise change data capture. Teleport Data became its high-speed database replication. In May 2025 Fivetran acquired Census, one of the two names in reverse ETL, putting warehouse-to-CRM activation in the same house. Five months later came the dbt merger. That’s ingestion, replication, activation and transformation under one roof inside a year, and it’s why I keep banging on about renewal clauses: you’re no longer buying a connector tool, you’re buying into a consolidating platform, and platforms price like platforms once the consolidation settles.
What the merger means for buyers
The official line is a merger of equals: a board drawn from both companies, near cash-flow break-even, roughly $600 million in combined revenue. Fivetran was last valued at $5.6 billion in 2021, dbt Labs at $4.2 billion in 2022.
For buyers I think it cuts two ways. The good: ingestion and transformation from one vendor, with the two products already integrated, will probably get smoother and better packaged. The caution: bundling is how pricing power works. If you’re signing a multi-year contract for either product, ask directly how the combined company plans to package the two, and get protection on renewal pricing in writing. The days of playing the two vendors’ roadmaps against each other are over.
What practitioners say
The review sites lean positive: 4.3 from 839 reviews on G2 (61% five-star, 33% four, and a real if thin tail: 1% each at two and one), 4.5 from 330 on Gartner Peer Insights. For a product this widely deployed, a visible one-star tail is normal; its absence elsewhere in this category says more about smaller samples than better products. The praise is consistent (it just works, nobody maintains pipelines any more) and so is the complaint, which is always cost. One Gartner review title from June 2026 sums up the whole genre: “High reliability and support offset by prohibitive overall cost”. That’s the deal in one sentence.
Fivetran or Airbyte?
This is the comparison everyone actually runs, so here’s my honest framing. Fivetran charges for rows that changed; Airbyte Cloud charges for volume moved; self-hosted Airbyte charges for neither, you pay in servers and attention instead. If your team spends real hours each month babysitting connectors, Fivetran buys those hours back and probably pays for itself. If your sources are standard, your volumes are lumpy, or you have an engineer who’d rather own the pipeline, the Airbyte maths usually wins.
There’s a third door people forget: stay on Fivetran’s free plan for the reliable long-tail SaaS sources and run the one heavy database sync somewhere cheaper. Nothing says you have to pick a single tool.
Verdict
Still the reliability benchmark for managed ingestion, now attached to the transformation standard, and priced like a product that knows both things. Go in with your row counts modelled, an exit path sketched, and a renewal clause you’ve actually read. Then it’s a good buy.
