Fivetran vs Airbyte

Airbyte offers 3 shapes of bill. Standard and Plus on Airbyte Cloud are volume-based: you pay in credits for the data you sync, with Standard from $20 a month at hourly syncs and Plus from $189 a month at 15 minute syncs. Pro and Enterprise Flex are capacity-based, priced by Data Workers, each of which typically runs about 3 syncs at once, with 5 minute syncs and custom quotes. Airbyte Core is free software that you host.

A volume bill tracks the data a sync moves, so the 2 models answer different questions. A large table fully refreshed every night moves its whole size each run, which a volume meter sees, while MAR counts only the keys that changed. Airbyte's pricing page quotes credits against gigabytes differently for the trial and for Standard, and meters differ by source type, so price your own sources in the calculator before comparing totals.

Capacity pricing flips the question from how much data to how many concurrent syncs. If your jobs are predictable and heavy, a fixed number of workers is easier to budget than a meter. Self-hosting removes the licence fee and adds the cost of the cluster, upgrades, on-call and security review, which is real engineering time.

Fivetran runs the connections for you. By default its runtime is on Google Cloud in the region of your destination. The Enterprise and Business Critical plans add a choice of cloud provider and the Hybrid Deployment model. Business Critical adds customer-managed encryption keys and PCI DSS Level 1.

Airbyte gives a wider range of places to run. Core runs wherever you deploy it, Cloud is hosted by Airbyte, and Enterprise Flex adds hybrid, on-premises and multi-region deployments where data stays inside your boundary. If a policy says source data may not pass through a vendor's network, both vendors offer a hybrid option on their top plans, and a self-hosted Core is the one that never leaves your network at all.

On licensing, read the page, not the label. Airbyte's licence page places its connectors and public repositories under the Elastic License 2.0 and the Airbyte Protocol under MIT. It says Cloud and Enterprise need a commercial licence, as do Agents. ELv2 allows internal use and embedding, and restricts offering Airbyte as a managed service or selling a product that exposes its UI or API. Older documentation described the connectors as MIT, so check the current text if a legal review depends on it.

Airbyte names its sync modes by how the source is read and how the destination is written. Incremental reads only new records, by a cursor column or by change data capture where the source supports it, and Full Refresh reads everything. On the write side the modes are Overwrite and Append, plus the deduplicating pair Append Deduped and Overwrite Deduped. The UI offers 5 combinations, such as Incremental Append + Deduped, which appends a change history and builds a final table deduplicated on the primary key. A mode appears only when both the source and the destination support it for that stream.

That surface is explicit, which suits teams that want to choose between keeping history and keeping the latest row per stream. Fivetran makes more of those choices for you: it manages the destination schema and counts deletes captured by supporting connectors, as well as changes to history-mode tables, in MAR. For a CDC pipeline the question to ask each vendor is the same: does the connector for my database read the log, how does it surface deletes and how are re-syncs billed. CDC and log-based capture has the mechanics, and the answers are per connector.

Every ELT tool has gaps, and the long tail of an in-house API is where they show. Fivetran's Connector SDK lets you write a connector in Python that Fivetran runs in an isolated environment on your schedule. It is available on all plans and billed by MAR. It does not support webhooks or History Mode, and Fivetran's support covers the platform while problems in your own code fall to Professional Services.

Airbyte's connectors are in public repositories, so a team can read how an existing one works and fix it, then run its own build in a self-hosted deployment. That openness costs you the maintenance of whatever you fork. In either tool, count the sources with no maintained connector before you decide, because 1 custom API owned by your team is a pipeline you maintain regardless of the vendor.

Fivetran fits when the team is small, the sources are mainstream SaaS and databases, and nobody wants to run connector infrastructure. It also fits when the biggest tables change slowly, so a MAR bill stays low. The cost of that is a bill you cannot cap by running fewer servers, and little control over how a connector behaves.

Airbyte fits when volume or residency rules make a managed service awkward, when you want to inspect or modify connectors, or when the budget prefers engineering time to a licence fee. It also fits when your changing data is high in volume but low in distinct keys per month, because a credit or capacity bill is then predictable. The cost is operational: on Core you own upgrades and scaling as well as failures, and on Cloud you choose a plan whose meter you must understand.

Many teams run both: a managed tool for the dozen standard SaaS sources and a self-hosted one for a database that can't leave the network. Before committing, load 1 real month of your busiest connection into each trial and compare the actual bill, not the list price.

Interviewers rarely ask you to name a vendor. They ask how data gets from a source system into the warehouse, and the strong answers show the trade-offs on this page: full refresh vs incremental, how deletes arrive, what a re-sync costs and how you backfill. Practice that reasoning with data pipeline interview questions and data pipeline practice problems, and the ingestion angle in CDC pipeline interview questions.

For the wider picture of how loops are structured, see the data engineer interview prep guide. If the open-source side of this choice is what you are weighing, open source ETL tools compares Airbyte with its alternatives, and ETL vs ELT explains why both tools load first and transform in the warehouse.