Fivetran is the "Apple" of ELT—expensive but it just works. Airbyte is the "Android/Linux"—open, flexible, and ubiquitous.
I've run both Fivetran and Airbyte in production at different companies, and the comparison is less about features and more about where you want the pain to live. Fivetran hides the pain inside an invoice — high, predictable, and mostly hands-off. Airbyte hides the pain inside your platform team — low license cost, but someone has to own the Kubernetes cluster, connector upgrades, and the occasional 2am "why did the Salesforce sync stop?" alert. Both move data from API A to warehouse B. They just price and operate differently.
At one company we ran Fivetran for Salesforce, Stripe, Zendesk, and 40 Postgres CDC tables. Bill hit $30K/month, mostly on one high-write Postgres table with ~500M MAR/month. We moved that single source to Airbyte self-hosted on an existing EKS cluster — migration took a sprint, ongoing ops was maybe 2 hours/week, infrastructure cost was ~$400/month. Net savings: $29K/month, minus the opportunity cost of the 2 hrs/week. For that one workload, Airbyte was dramatically better economically.
For Salesforce though, we kept Fivetran. The Salesforce schema changes constantly (custom fields, new objects). Fivetran handles schema drift silently; Airbyte requires manual intervention for some changes. One Fivetran invoice line item for Salesforce was worth the cost avoidance of a flaky pipeline that would block the CEO's weekly revenue dashboard. Horses for courses.
Choose Fivetran if:
For most small-to-medium data teams, start with Fivetran for the 10–20 high-value connectors that matter most (Salesforce, Stripe, a couple of databases) and accept the license cost as the price of not needing a platform team. As volume grows, profile your MAR usage and consider moving the top 1–2 highest-cost sources to Airbyte self-hosted — that's usually where the biggest savings hide. For companies that already have strong Kubernetes/DevOps capability or have strict data residency requirements, Airbyte (self-hosted or Enterprise) is the right answer from day one. And honestly, running both is fine — the idea that you must pick one tool for all data is marketing, not engineering.
Airbyte Core (OSS) is free under the Elastic License — no license cost. But running it in production means infrastructure costs (Kubernetes or Docker hosts, storage, network) plus engineer time for operations (upgrades, connector version management, monitoring, on-call when a sync fails at 3am). Realistically, a production Airbyte deployment for a mid-size team costs $500–3K/month in infrastructure plus 0.25–0.5 FTE of platform engineer time. "Free" is a relative term. Airbyte Cloud is a paid managed service that removes most of the ops overhead and shifts cost back to license.
Fivetran charges based on Monthly Active Rows (MAR) — the number of unique rows added, modified, or deleted each month per source. A row that changes 30 times in a month counts as 1 MAR; a row that exists and doesn't change counts as 0. This pricing is gentle for append-heavy or low-churn sources (e.g., Stripe charges, Shopify orders) and brutal for high-update sources (e.g., a Postgres CDC table with frequent updates, or Salesforce with many active users). Model your top 3 sources carefully before committing — pricing surprises almost always come from CDC and high-write tables.
For many enterprise use cases, yes — especially with Airbyte Enterprise, which adds SSO, RBAC, audit logs, private networking, and SLAs. Where Fivetran still tends to win: (1) schema drift handling is more mature and silent — Fivetran notices a new Salesforce field and just adds the column; Airbyte may need attention. (2) Enterprise compliance certifications (SOC 2 Type II, HIPAA BAA, GDPR addendums) are turnkey with Fivetran; with Airbyte self-hosted, compliance is your problem. (3) Long-running connector reliability for the top 20 SaaS sources is still better on Fivetran. Pilot both on your real pipelines before committing.
Absolutely, and it's a very common pattern at scale. Use Fivetran for high-value, low-volume sources where reliability matters most (Salesforce, NetSuite, financial systems) and Airbyte for high-volume sources where MAR pricing would be painful (large CDC tables, high-event SaaS products) or long-tail custom connectors. The two tools don't conflict; you just split sources across them based on cost/reliability trade-offs per source.
Fivetran handles schema drift automatically for nearly all connectors — a new column appears in your warehouse the next sync, no human action. Airbyte handles common drift cases, but for complex scenarios (renamed fields, dropped columns, non-additive changes) you may need to manually refresh the schema or reset the stream. For Salesforce specifically, where new custom fields appear constantly, Fivetran's drift handling is noticeably smoother. This is one of the top reasons teams stay on Fivetran even when MAR pricing is expensive.
Fivetran has first-class CDC for Postgres, MySQL, SQL Server, Oracle, MongoDB, and a few others — log-based, low-latency, handles backfills and resyncs gracefully. Airbyte also has CDC for Postgres/MySQL/SQL Server via Debezium; quality has improved dramatically in the past year but still has more rough edges than Fivetran (occasional need for resyncs, state management quirks). For mission-critical CDC, Fivetran is the safer choice today. For cost-sensitive CDC where some ops overhead is acceptable, Airbyte can work well.
Both integrate with dbt well, since both just land raw data in your warehouse and let dbt handle the T. Fivetran has a slight edge here because they publish and maintain dbt packages for the top 50 connectors — you get pre-built staging models that normalize the raw Fivetran schemas into clean dbt-friendly tables. Airbyte has some official and community dbt packages but coverage is thinner. If you're a heavy dbt shop, factor in Fivetran's dbt package ecosystem when modeling TCO.