dbt Cloud vs Fivetran vs Airbyte — which data pipeline in 2026?

Pipeline costs now include transformed volume (Fivetran per-row MAR), managed orchestration (dbt Cloud per successful model), and self-hosted open-source (Airbyte infra costs). Break-even shifts as data volume grows and team scales, making the right choice highly context-dependent.

dbt Cloud for transformation-first teams. Fivetran for managed ingestion at scale. Airbyte OSS if you want to own infrastructure and avoid per-row pricing.

Blockers

Who this is for

Candidates

dbt Cloud

Managed SQL transformation orchestration billed by successful model materializations. As of 2026-03-16, the Starter plan is $100/seat/month and includes 15,000 successful models/month with $0.01 per overage model; the Developer plan is free with 3,000 models/month and jobs pause at the limit.

When to choose

Choose this for low-ops + small-team environments where transformation logic is the primary complexity and ingestion is already handled elsewhere. Best when model counts stay comfortably below plan thresholds and you want managed job scheduling, CI runs, the dbt Semantic Layer, and dbt Catalog without self-hosting. Starter is most cost-effective up to 5 developer seats and roughly 15,000 model builds/month; beyond that, negotiate Enterprise. Pairs naturally with any ingestion layer including Fivetran or Airbyte.

Tradeoffs

The billing metric — successful model materializations — counts tables, incremental models, and dynamic tables built via scheduled jobs, CI, and API runs, but explicitly excludes tests, seeds, ephemeral models, snapshots, and development IDE runs, which limits surprise bills during local iteration. Starter is capped at one project; Enterprise supports 30 projects; Enterprise Plus offers unlimited projects with custom pricing. Usage alerts fire at 75%, 90%, and 100% of monthly thresholds. The main risk is that pipelines with high model counts or heavy CI usage can exhaust included models quickly.

Cautions

dbt Cloud bills per materialization, not per data volume, so transformation cost grows with pipeline complexity rather than row count. Projects with hundreds of daily-running models on Starter can breach 15,000 models faster than expected once CI jobs are factored in. Enterprise plan pricing is annual, negotiated with sales, and not publicly disclosed. Verify current plan names and allowances at getdbt.com/pricing before committing — plan naming changed from Team to Starter since the initial consumption-based pricing launch in 2023.

Sources

Fivetran

Fully managed ELT ingestion service billed by Monthly Active Rows (MAR) per connector. As of 2026-03-16, a $5 base charge applies to any connection generating 1 to 1 million MAR per month; since January 2026, inserts, updates, and deletes all count toward paid MAR. Per-MAR tier rates use volume discounts and are published in Fivetran's Service Consumption Table.

When to choose

Choose this for low-ops + enterprise environments where managed reliability, 700+ pre-built connectors, and minimal pipeline maintenance matter more than per-row cost. Best when data sources are numerous and diverse but row volumes per connector are moderate and predictable. Free plan covers up to 500,000 MAR/month. Pairs naturally with dbt Cloud for a fully managed ingestion-plus-transformation stack when operational simplicity is the priority.

Tradeoffs

Fivetran handles connector maintenance, auth rotation, schema drift, and sync reliability with no infrastructure to manage. The cost tradeoff is per-MAR billing that compounds as row volumes and connector count grow. Since March 2025, MAR is calculated per-connector rather than account-wide, so multiple connectors no longer benefit from aggregate volume discounts. The January 2026 change making deletes count toward paid MAR is a meaningful cost increase for high-churn sources. History Mode now also incurs charges for repeated updates as of January 2026.

Cautions

As of January 2026, deletes count as paid MAR — a breaking cost change for CDC-heavy or high-churn source tables. History Mode billing now charges for each tracked repeated update; teams can switch to soft delete mode or live mode to reduce History Mode MAR. Per-connector MAR calculation (since March 2025) removes cross-connector volume pooling, raising costs for multi-connector accounts. Consult the Service Consumption Table at fivetran.com/legal/service-consumption-table for current tier dollar rates, as specific per-MAR amounts are not published on the main pricing or docs pages.

Sources

Airbyte Cloud

Managed ELT ingestion with a volume-based Standard tier and capacity-based Plus/Pro tiers using Data Workers. As of 2026-03-16, Standard starts at $10/month with volume-based credit pricing; Pro and Plus use Data Workers capacity billing for predictable spend regardless of row count; a 14-day free trial is available.

When to choose

Choose this for cost-sensitive + small-team environments that need a managed ingestion service but want to avoid Fivetran's per-MAR model at moderate volumes. Standard works when data volumes are lower and billing predictability is less critical; upgrade to capacity-based Data Workers (Plus/Pro) once volume-driven billing becomes unpredictable. Best when you want open-source Airbyte connector breadth with managed hosting and no Kubernetes overhead.

Tradeoffs

Airbyte Cloud avoids self-hosting complexity while preserving access to the open-source connector library. Standard tier credit-based billing still ties cost to data volume, but capacity-based Pro tiers decouple cost from row count — Airbyte's blog documents syncing approximately 10 GB/month at roughly $100/month on Cloud as a reference point. Governance features including RBAC, row filtering, column hashing, encryption, and multiple data regions require Pro tier. Annual billing unlocks volume discounts on Plus.

Cautions

Standard tier is still volume-based: unexpected data spikes will increase credit consumption without a hard cap. Data Workers capacity billing on Plus/Pro charges for processing power whether fully utilized or not — right-sizing matters to avoid overpaying. Compliance and enterprise features (SSO, RBAC, data masking, multiple regions) are not available on Standard. Verify current credit-to-dollar rates and Data Worker pricing at airbyte.com/pricing, as the capacity model is relatively new.

Sources

Airbyte OSS (Self-Hosted Core)

Free MIT-licensed open-source ELT platform with zero per-row software cost. As of 2026-03-16, Airbyte Core is documented as always free with no licensing fees; all infrastructure, maintenance, and operational overhead are owned by the team.

When to choose

Choose this for cost-sensitive + high-scale environments where per-row ingestion fees at Fivetran or Airbyte Cloud would exceed self-managed infrastructure costs at your data volume. Best for teams with existing DevOps capacity who can absorb Kubernetes cluster management and connector upgrade cycles in exchange for eliminating per-row charges entirely. Break-even typically favors self-hosting once managed service fees would exceed the cost of dedicated infrastructure plus engineering maintenance time.

Tradeoffs

Zero per-row or per-GB software licensing cost means total cost scales with infrastructure, not data volume, making it favorable at high and growing volumes. Access to Airbyte's full open-source connector library with sub-5-minute sync frequencies supported. The tradeoff is significant ops burden: production deployments require container orchestration, persistent storage, connector version upgrades, schema drift handling, and monitoring — all owned entirely by the team.

Cautions

Infrastructure costs are real and non-trivial — production Kubernetes deployments require cluster, storage, and network budgeting, plus recurring DevOps time for maintenance. Community support only; no SLA without upgrading to the paid Enterprise edition. Core supports a single workspace only — multitenancy, SSO, RBAC, PII masking, and multiple workspaces require the paid Enterprise edition. Model the engineering maintenance cost explicitly against managed service fees before committing to self-hosting.

Sources

Facts updated: 2026-03-16
Published: 2026-03-29

Related questions

Try with your AI agent

$ npm install -g pocketlantern
$ pocketlantern init
# Restart Claude Code, Cursor, or your MCP client, then ask:
# "dbt Cloud vs Fivetran vs Airbyte — which data pipeline in 2026?"
Missing something? Request coverage