Airbyte Cloud vs Fivetran vs Estuary Flow: Real Data Integration Cost per 100 Million Rows in 2026
The short answer: these three don't bill the same underlying quantity, and the gap matters most at genuine scale. Fivetran bills Monthly Active Rows (MAR) — distinct rows that actually changed, not raw row count moved — at a rate that declines with volume but starts around $500–$800 per million MAR and almost certainly requires a custom Enterprise negotiation well before reaching 100 million MAR a month; its $12,000 annual minimum commitment applies regardless. Airbyte Cloud bills differently depending on source type: $15 per million rows for API sources, or $10 per GB for database/warehouse sources — two different units within one product. Estuary Flow bills the most predictably of the three, at a flat $0.50 per GB of data moved plus $100 per connector per month, and at an illustrative 100 GB of data (100 million rows at roughly 1 KB each), Estuary comes to about $350/month, versus $1,000–$1,500 on Airbyte depending on source type — with Fivetran's real cost at this volume effectively unknowable from public pricing without a sales conversation.
The billing-unit mismatch, up front
Monthly Active Rows (Fivetran) counts only rows that were inserted, updated, or deleted during the billing period — a row read repeatedly without changing doesn't count again, and critically, initial bulk loads and unchanged rows re-synced don't count toward MAR either. This means "100 million rows changed per month" is not automatically "100 million rows moved" — a connector that re-syncs its entire table nightly but only 2% of rows actually change would bill far less on MAR than the raw sync volume suggests. Airbyte's unit splits by source type entirely — rows for API sources, raw gigabytes for database/warehouse sources — meaning the same logical data volume bills differently depending on where it comes from. Estuary's gigabytes-moved is the most volume-direct of the three, closest to a simple "how much data crossed the pipe" measure, independent of how many discrete rows that data represents.
What changed in 2026 (and just before it)
Fivetran's March 2025 shift to per-connector MAR billing continues to shape 2026 pricing: the change added a $5 minimum charge per connection, made data deletes billable for the first time, and removed account-level volume discounts that previously let a large, multi-connector account blend its rate across all connections — independent reporting describes this as raising costs 40–70% for multi-connector setups specifically. Any comparison or calculator still using pre-2025 Fivetran pricing logic will understate current cost materially for exactly the kind of multi-source AI SaaS data pipeline this article is modeling.
What each vendor bills
Fivetran (OFFICIAL for the mechanism; REPORTED and somewhat conflicting for exact rates). Free: 500,000 MAR/month, limited connector set. Paid tiers price per million MAR with volume-based declining rates — one sourced tiered example shows roughly $800/million MAR at 500,000 MAR, declining to about $750/million at 1 million MAR, and $400/million at 4 million MAR; a separate source cites Standard-tier pricing starting around $500/million MAR. Both describe the same general shape (lower effective rate at higher committed volume) without fully agreeing on exact figures at any single point, which this article does not attempt to reconcile. A $12,000 annual minimum commitment applies on paid plans. Connector-type multipliers mean some connectors (particularly high-change-volume database sources) consume credits faster per MAR than others.
Airbyte Cloud (OFFICIAL for the mechanism; REPORTED for exact rates). Core (self-hosted): free, unlimited, any connector, no row or volume cap — a genuine cost-elimination path for a team with the operational capacity to run it. Cloud: credit-based, with API sources billed at $15 per million rows (6 credits) and database/warehouse sources billed at $10 per GB (4 credits) — two structurally different meters within the same product, depending entirely on connector type. Data Replication Standard: from $10/month base, scaling with credit consumption; Plus: $2,083/month, adding 15-minute sync frequency. Independent analysis reports Airbyte Cloud typically running 30–50% cheaper than Fivetran for equivalent volumes, narrowing toward Fivetran's commercial-tier pricing only at very high volume.
Estuary Flow (OFFICIAL, Estuary's pricing page). Free: 10 GB/month. Paid: $0.50 per GB of data moved, plus $100 per connector per month for the first six connectors. Estuary explicitly prices real-time change data capture at the same rate as batch synchronization — no latency premium for sub-second freshness, a genuine structural difference from Fivetran's and Airbyte's typical pattern of charging more for faster sync frequency (as seen in Airbyte's $2,083/month Plus tier specifically for 15-minute syncs). Estuary's own comparative claim describes Airbyte as running "5–10x lower" than Fivetran at scale, and by extension positions itself similarly or more favorably — a vendor-sourced comparison, not an independently verified multiple.
Cost at 100 million rows/month, normalized by source type
All ILLUSTRATIVE for the rows-to-GB conversion: this article assumes an average 1 KB per row, giving 100 GB for the database/warehouse-sourced scenarios.
| Platform | Basis | Monthly cost at 100 million rows |
|---|---|---|
| Estuary Flow (100 GB, 3 connectors) | 100 GB × $0.50/GB + 3 × $100/connector | $350 |
| Airbyte Cloud, database/warehouse source | 100 GB × $10/GB | $1,000 |
| Airbyte Cloud, API source | 100,000,000 rows ÷ 1M × $15 | $1,500 |
| Fivetran | 100 million MAR vastly exceeds every published tier's volume anchor; the $12,000/year minimum applies regardless, but the actual per-MAR rate at this scale is not published | QUOTE-ONLY / UNKNOWN |
Fivetran's absence of a clean number at this volume reflects how the product is actually sold, not an unanswered question — its published tier examples top out at single-digit millions of MAR, and 100 million MAR is squarely Enterprise-negotiation territory, consistent with its $12,000 annual minimum and custom top-tier pricing.
Why "100 million rows" doesn't mean the same thing on Fivetran as on the other two
Formula: Fivetran MAR = rows inserted + rows updated + rows deleted, excluding unchanged rows from re-syncs and bulk loads. A source table with 100 million total rows, where only a fraction change in a given month, bills Fivetran far below "100 million MAR" — but a highly volatile OLTP table where most rows change daily could approach or exceed that figure even with a much smaller total table size. This is the central reason Fivetran's pricing is described as "genuinely hard to forecast" compared to Airbyte's or Estuary's more volume-direct meters: the bill depends on your data's rate of change, not its size.
Sync frequency as a hidden cost lever
Both Fivetran's and Airbyte's pricing structures implicitly or explicitly charge more for faster synchronization — Airbyte's Plus tier specifically gates 15-minute sync frequency behind a $2,083/month plan, a large step up from its entry tier. Estuary Flow's flat per-GB rate, regardless of latency, removes this lever entirely: a team needing genuinely real-time CDC pays the same per-GB rate as one running nightly batch syncs, which can make Estuary structurally cheaper specifically for low-latency requirements, independent of raw data volume.
Sensitivity
- Rate of data change, for Fivetran specifically. The single largest and most unpredictable lever on Fivetran's bill; two datasets of identical size can produce wildly different MAR totals depending on how often rows actually change.
- Source type mix, for Airbyte. API-sourced and database-sourced data bill on entirely different meters ($/row versus $/GB), so the real Airbyte cost depends on which connectors dominate the pipeline.
- Sync frequency requirement. Directly gates a higher-priced tier on Airbyte (15-minute syncs require Plus) and has no cost impact at all on Estuary's flat per-GB rate.
- Connector count, for Estuary. The $100/connector/month fee (for the first six) adds a real, linear cost as pipeline complexity grows, separate from the data-volume charge.
Budgeting traps
- Estimating Fivetran cost from total table size rather than actual row-change rate. MAR counts only changed rows, which can be dramatically lower (or, for volatile tables, comparably high) relative to raw table size.
- Assuming Airbyte's per-row API rate applies to database-sourced connectors, or vice versa. The two source types bill on genuinely different units within the same product.
- Needing fast sync frequency without checking its cost impact. On Airbyte specifically, 15-minute syncs require a materially pricier tier; Estuary's flat rate avoids this entirely.
- Budgeting Fivetran off a lower-volume tier's per-MAR rate and extrapolating linearly to 100 million MAR. The published tiers don't extend that far, and the real rate at that scale requires a sales conversation.
What to ask before you buy
Ask Fivetran directly for pricing at your actual expected MAR volume once it approaches the high tens of millions, since no published tier extends to that scale and the $12,000 annual minimum is only a floor, not a ceiling. Measure your pipeline's actual source-type mix (API versus database/warehouse) before estimating Airbyte Cloud cost, since the two bill on entirely different units and a mixed pipeline needs both calculations summed separately.