This note covers Airbyte’s February 2025 pricing model and the hidden infrastructure costs of self-hosting — particularly NAT Gateway charges and Kubernetes overhead that are not visible on the pricing page.
The February 2025 Pricing Model
Airbyte introduced capacity-based pricing in February 2025, replacing their earlier credit-based model. The structure:
- Open-source Core: Free, forever. Self-host on your own infrastructure.
- Cloud Standard: $10/month base plus volume fees: $15 per million rows for API sources, $10 per GB for database and file sources.
- Cloud Plus and Enterprise: Higher tiers with additional features (SSO, RBAC, priority support, HIPAA compliance).
The volume-based model compares favorably to Fivetran’s MAR pricing for many workloads. Rough comparison for a mid-sized data team loading 50 million API rows plus 50GB of database data:
| Tool | Approximate Monthly Cost |
|---|---|
| Airbyte Cloud | ~$1,250 ($10 base + $750 API rows + $500 database) |
| Fivetran (MAR-based) | ~$4,445 (depending on tier and connector count) |
The Airbyte number doesn’t balloon the same way Fivetran’s does when you add connectors, because API row counts are shared across the account rather than per-connector. A row is a row regardless of which source it came from.
This matters most for teams with many connectors, where the per-connector MAR tiering eliminated the bulk discounts Fivetran customers previously received.
The Self-Hosting Option
Airbyte’s open-source Core is free, but self-hosting trades subscription fees for engineering time and infrastructure costs.
Infrastructure Requirements
Running Airbyte OSS in production requires:
- A Kubernetes cluster (EKS, GKE, or AKS) with minimum 2 cores and 8GB RAM per node
- 30GB disk space
- External PostgreSQL for metadata storage
- S3 or GCS for log storage
Each connector runs in its own Docker container; the cluster needs enough capacity to run concurrent syncs without resource contention. Kubernetes expertise and Helm chart management are required.
The NAT Gateway Problem
The cost that surprises almost everyone: NAT Gateway fees.
When Kubernetes pods connect to external APIs, traffic routes through a NAT Gateway (or equivalent) for outbound internet access. AWS NAT Gateway costs $0.045/hour (roughly $32/month just to exist) plus $0.045/GB of data processed. On GCP and Azure, similar costs apply.
Practitioners have reported NAT Gateway costs representing 80% of total infrastructure costs at moderate scale. One user noted that “NAT gateway and EC2 costs will far outweigh the rest.” A team expecting $0 for open-source deployment often ends up with $200-500/month in infrastructure before accounting for compute.
Kubernetes Operational Overhead
Beyond infrastructure costs, self-hosted Airbyte has ongoing maintenance requirements:
- Connector version management: Airbyte releases connector images regularly. Self-hosted deployments need to update them manually, or you fall behind on API changes.
- Helm chart updates: Airbyte’s platform itself releases updates. Upgrading a Kubernetes deployment is more involved than clicking “update” in a SaaS dashboard.
- Failure response: When a connector fails at 2am because the source API returned a malformed response, someone on your team gets paged.
- Kubernetes expertise: Most data teams don’t have a Kubernetes specialist. Running a production Kubernetes cluster is DevOps work. The engineering time this consumes is real money.
When Self-Hosting Saves Money
- Cloud Standard is generally better for teams without Kubernetes expertise or with moderate data volumes (under 100M rows/month).
- Self-hosting pays off for high-volume workloads (hundreds of millions of rows) where data processing fees at cloud pricing exceed infrastructure costs.
The relevant calculation includes monthly infrastructure cost plus amortized engineering time for infrastructure management.
Connector Quality
Of 600+ connectors, roughly 350 are official (Airbyte-maintained); the rest are community-contributed. Official connectors respond to API changes within reasonable timeframes. Community connectors can lag by weeks or months after a source API change.
Major marketing sources (Google Ads, Meta Ads, GA4, HubSpot, Shopify) have official connectors. For niche SaaS tools, confirm whether the connector is official or community-maintained before using it in production.
Decision Criteria
Airbyte is well-suited for teams that:
- Have engineering capacity to manage Kubernetes infrastructure (or are willing to pay for Cloud)
- Need connector breadth beyond dlt’s 60+ verified sources but cannot justify Fivetran pricing
- Have workloads where capacity-based pricing is lower than MAR-based pricing
- Can tolerate variable connector quality
Not well-suited for: teams wanting zero maintenance overhead (Fivetran), or Python-proficient teams that can build custom pipelines cheaply (dlt).
Airbyte’s headline cost is accurate; total cost is higher. Infrastructure (especially NAT Gateway), connector maintenance, and engineering time for production deployment management should be budgeted.