Fivetran vs Airbyte vs Matillion: Real Data Pipeline Cost per Million Rows
Short answer
Fivetran is by far the best-documented of the three, with a confirmed $500/month Standard tier covering 1 million Monthly Active Rows (MAR) — but its pricing has genuinely changed for the worse for many customers over 2025–2026: a March 2025 shift to per-connection cost curves eliminated blended bulk discounts across multiple connectors, and a January 2026 update added a $5 minimum charge per connection and made deleted rows count toward billable usage for the first time. Airbyte's open-source self-hosted version is free, with Airbyte Cloud priced on usage; Matillion uses its own credit-based system starting around $2/credit. Both Airbyte's and Matillion's broader pricing at scale is far less precisely documented in public sources than Fivetran's.
Pricing at a glance
| Fivetran | Airbyte | Matillion | |
|---|---|---|---|
| Public rate card | Yes, detailed and confirmed, including a published consumption curve | Partial — self-hosted open-source tier confirmed free; Airbyte Cloud pricing is usage-based but not as precisely documented in available sources | Partial — entry credit rate reported ($2.00/credit) from third-party comparison, not independently verified against Matillion's own current pricing page; full tier structure not as precisely documented |
| Free tier | Yes — 500,000 MAR/month for connections, 3,500 MAR for Activations (reverse ETL), 5,000 model runs/month for transformations | Yes — the self-hosted open-source version is free indefinitely, with infrastructure being the customer's own cost | Not confirmed |
| Entry paid tier | Standard: $500/month, includes 1M MAR, 15-minute sync intervals, 700+ connectors, dbt Core support | Airbyte Cloud: usage-based, priced per row synced — specific entry-tier dollar figures vary widely across secondary sources and were not confirmed to a single reliable figure | Credit-based, starting at $2.00/credit (reported, via third-party comparison) |
| Billing unit | Monthly Active Rows (MAR) — every row inserted, updated, or deleted counts, calculated per connection rather than blended across an account | Rows synced (Cloud); no metering on self-hosted | Credits — a different consumption unit than Fivetran's row-based model, reflecting Matillion's combined ELT-plus-transformation positioning |
| Minimum charge | $5/month minimum per standard connection (1–1M MAR band), added January 2026 | Not applicable to self-hosted; not confirmed for Cloud | Not confirmed |
| Reported cost at scale | Larger organizations syncing 20–25 million MAR are reported to pay $20,000–30,000/month, particularly on enterprise plans | Not confirmed at a comparable scale in available sources | Not confirmed at a comparable scale in available sources |
What headline pricing excludes
Fivetran's March 2025 shift to per-connection cost curves is one of the most consequential pricing changes in this category, and it's easy to miss if comparing against pre-2025 pricing research. Before this change, a company syncing 500,000 rows from Salesforce and 300,000 rows from Google Ads paid for a blended 800,000 total MAR, benefiting from volume discounts across the combined total. Now, each connector is billed on its own separate cost curve — the same company pays for 500,000 MAR (Salesforce) and 300,000 MAR (Google Ads) as two independent billing events, losing the blended-volume discount entirely. This change is reported to typically increase costs by 40–70% for organizations running multiple connectors, which describes most real-world Fivetran deployments.
Fivetran's January 2026 update made deleted rows count toward billable MAR for the first time — previously, deletes did not contribute to paid usage. A workload with heavy deletion activity (a CRM with regular record purging, for instance) will see its bill rise under this change even if actual net data volume hasn't grown.
Fivetran is effectively three separate billing meters bundled under one brand: connection MAR (standard data sync), Activation MAR (reverse ETL, its own consumption curve per destination), and transformation MMR (model runs, billed separately). A company using all three capabilities should expect three distinct usage-based charges, not one.
Certain data characteristics inflate MAR usage in ways that aren't obvious from row counts alone — nested data structures (JSON, for instance), frequent schema changes, and real-time sync configurations can all multiply effective row counts and drive MAR usage well above what a simple total-record-count estimate would suggest.
Hidden costs
- Fivetran renewal price increases of 2–4x are reported as "common" by at least one detailed source — a company should not assume its first-year Fivetran quote represents a stable, ongoing cost without specifically negotiating renewal-rate protection into the initial contract.
- A 5% uplift applies to quarterly and semi-annual Fivetran payment schedules relative to annual billing, per reported customer transaction data — a real, avoidable cost for a company willing to commit to annual payment terms instead.
- Actively evaluating Airbyte as a competitive alternative is reported to be a genuinely effective Fivetran negotiation lever — one specific reported case describes a customer's discount initially being reduced by 6%, but after two months of seriously evaluating a move to Airbyte, Fivetran restored the customer's full 39% discount. Demonstrating real evaluation effort, not just mentioning a competitor's name, is described as the key factor.
- Demonstrating 25%+ anticipated MAR growth is reported as a specific, usable lever for negotiating better connector-specific discounts or retaining existing discount rates — one reported case describes a customer securing $36,000 in "rollover MAR" by presenting expected growth during a renewal negotiation.
Worked scenarios
1 million rows/month
| Estimated monthly cost | |
|---|---|
| Fivetran Standard | $500/month, confirmed — this tier is specifically built around and priced for exactly this volume |
| Airbyte (self-hosted) | $0 in licensing cost, plus the customer's own infrastructure hosting expense, which isn't itemized in available sources |
| Matillion | Not confirmable at this specific row volume, since Matillion bills in credits rather than rows directly — a company would need to convert its expected transformation workload into Matillion's credit consumption model, which depends on job complexity rather than row count alone |
At this volume, Fivetran's confirmed, purpose-built pricing tier makes it the easiest of the three to budget with confidence, provided the workload uses a single or few connectors and doesn't trigger the multi-connector cost-curve penalty described above.
100 million rows/month
Fivetran does not publish a single confirmed rate at this exact volume, but reported figures for a comparable scale (20–25 million MAR) put enterprise-tier cost at $20,000–30,000/month — at 100 million MAR, cost would be meaningfully higher, though Fivetran's own consumption curve is designed so per-MAR cost decreases as volume increases, meaning a simple 4–5x linear extrapolation from the 20–25 million MAR figure would likely overstate the real cost. A company at this scale should request Fivetran's specific service consumption table for its actual connector mix rather than relying on any extrapolation. Neither Airbyte's Cloud pricing nor Matillion's credit consumption at this row volume could be confirmed from available sources with enough precision to estimate confidently.
1 billion rows/month
At this scale, all three vendors move into individually negotiated enterprise territory, and none can be estimated with confidence from public information. A company at this volume is very likely to be a candidate for Fivetran's negotiated Enterprise or Business Critical tiers, where the per-MAR cost curve continues to decline, but the exact resulting figure depends on connector mix, growth commitments, and negotiated terms that vendors do not disclose publicly.
Normalizing across billing units
Fivetran's MAR-based pricing and Matillion's credit-based pricing measure fundamentally different things — MAR counts actual data rows changed, while Matillion's credits are consumed by transformation job execution, reflecting its combined ELT-plus-transformation product positioning rather than pure extraction-and-load volume. A company evaluating "cost per million rows" across both vendors needs to separately model its actual transformation workload complexity for Matillion, not just its raw row-sync volume, since the two aren't measuring the same underlying activity.
Break-even and crossover
The clearest, most actionable finding in this article isn't a cross-vendor dollar crossover — it's Fivetran's own negotiation dynamic. A credible, demonstrated evaluation of Airbyte as an alternative is reported to fully restore a discount that had been reduced, in one specific case moving a customer from a reduced 6% discount back to a full 39% discount after a two-month evaluation period. This suggests the effective "market rate" for Fivetran is considerably below its list-tier pricing for any customer willing to invest real time in a competitive evaluation, rather than accepting an initial renewal quote.
Who pays more, and when
- A company with a single connector and stable, moderate row volume is well-served by Fivetran's confirmed Standard tier pricing, and is less exposed to the multi-connector cost-curve penalty that affects more complex deployments.
- A company running many connectors simultaneously should specifically model the post-March-2025 per-connection billing change, which is reported to increase costs 40–70% relative to the older blended model — this is likely the single biggest driver of an unexpectedly high Fivetran bill for a multi-source deployment.
- An engineering team with infrastructure expertise and a preference for maximum control is well-positioned to use Airbyte's self-hosted open-source option at zero licensing cost, accepting the trade-off of managing infrastructure and connector maintenance internally.
- A team needing combined ELT and transformation in one platform, rather than separate extraction/load and transformation tools, is the buyer profile Matillion's credit-based pricing is built around.
- Any company renewing a Fivetran contract should proactively negotiate using a genuine competitive evaluation (Airbyte, Stitch, or similar) and should specifically ask about renewal-rate protection, given reported 2–4x renewal increases are common absent such negotiation.
Limitations and uncertainty
Fivetran's pricing is the most thoroughly documented of the three, confirmed directly from its own pricing page and service consumption table, including specific, dated 2025–2026 pricing-model changes. Airbyte's Cloud pricing (as distinct from its free self-hosted tier) and Matillion's full credit-based tier structure are both less precisely documented in available sources — reported ranges for both vendors show significant spread across different secondary sources, some of which appear to offer very wide, low-confidence estimates rather than a tightly corroborated figure. Fivetran's specific cost at 100 million and 1 billion MAR was not confirmed and is explicitly not estimated with precision in this article, given the non-linear nature of its published consumption curve.
Official sources
- fivetran.com/docs/core-concepts/usage-based-pricing (confirmed, including 2025–2026 pricing-update documentation)
- Airbyte's self-hosted pricing (free) is confirmed on its own site; Airbyte Cloud's specific usage-based rates were not confirmed to a single reliable figure
- Matillion's entry credit rate ($2.00/credit) is reported via third-party comparison, but Matillion's current pricing page does not independently corroborate the figure; full tier pricing is not published in detail