Skip to content
Engineering deep-divesAll articles

Airbyte vs Fivetran in 2026: MAR, dbt Merger, CDC & TCO

Engineering and 36-month TCO benchmark of Airbyte vs Fivetran in 2026: Fivetran + dbt Labs merger lock-in, unpooled MAR vs Data Workers, PostgreSQL 17 pgoutput WAL slot guards, S3 Iceberg ELT, and 180M MAR/mo math.

Artem Zaitsev
Artem ZaitsevCTO · Idealogic
Author: Artem ZaitsevPublished: Updated: Reading time: 20 minHow we write →
On this page
Idealogic 2026 engineering comparison of Airbyte vs Fivetran covering unpooled MAR vs Data Workers, the Fivetran and dbt Labs merger, PostgreSQL 17 pgoutput WAL guards, S3 Iceberg ELT, and 36-month TCO

Airbyte vs Fivetran comes down to how you want to pay for moving data. Fivetran is a managed service that bills per Monthly Active Row (MAR) on each connection. Airbyte is open core: you self-host the ELv2-licensed platform for free, or you buy capacity-priced Data Workers in its cloud. In our worked model of 35 production connectors and 180M MAR a month, it's a three-year pipeline bill of $234,900 to $432,900, and the 2026 Fivetran and dbt Labs merger adds a renewal question on top.

This guide covers one layer of the data stack: managed and owned ELT and Change Data Capture (CDC) pipelines that move operational Postgres and SaaS records into S3, ClickHouse, Snowflake or BigQuery. For the other layers, ClickHouse vs TimescaleDB benchmarks the analytical storage engine, Metabase vs Superset covers BI, Redpanda vs Kafka compares event-streaming brokers, and Dagster vs Prefect picks up orchestration once the data has landed.

At Idealogic, our custom software development, SaaS development, software development consulting, and dedicated development team squads build data backbones across fintech, manufacturing, and logistics and supply chain. Here we compare Airbyte vs Fivetran against a third option, an owned PostgreSQL pgoutput CDC pipeline into S3 Iceberg, over 36 months at 180M MAR a month. The examples run on PostgreSQL 17, but the replication features they use work on PG15 and later.

The short version

  • The Fivetran and dbt Labs merger closed on June 1, 2026. Announced October 13, 2025 and completed on June 1, 2026 (Fivetran's announcement), the all-stock deal puts ingestion and dbt transformation under one vendor with close to $600M in combined ARR. One contract for both is convenient. It's also a reason to read per-connection MAR pricing closely before you renew.
  • Airbyte has two very different cost shapes. Self-hosted Airbyte Core costs $0 in license (ELv2 platform, MIT connectors) but ships without SSO or RBAC, and we budget 36 hours a month of Kubernetes work to run it. Airbyte Pro and Enterprise Flex sell flat-rate Data Workers, so a spike in row mutations doesn't change the invoice.
  • A stalled pgoutput replication slot can fill an RDS disk. It doesn't matter whether Fivetran, Airbyte or Debezium reads the slot. Without heartbeat writes and a max_slot_wal_keep_size limit, Postgres retains WAL until primary storage is full.
  • At 180M MAR a month across 35 connectors, the owned pipeline has the lowest modeled 36-month TCO: $107,400 at CEE rates, against $165,600 for Airbyte Core on EKS ($306,000 at US rates), $234,900 for Airbyte Enterprise Flex and $432,900 for Fivetran Enterprise. That's $325,500 (75.2%) less than Fivetran. Vendor subscriptions are modeled assumptions, not list prices.

Airbyte vs Fivetran in 2026: MAR, Data Workers and the dbt merger

Both tools ship hundreds of prebuilt connectors (Airbyte advertises 700+), and both extract Stripe, Salesforce, HubSpot and NetSuite cleanly. So a Fivetran vs Airbyte decision rarely turns on connector count. The 36-month cost diverges on four things: how MAR is metered, what the merger does to your contract, how much memory Kubernetes connector pods need, and whether your PostgreSQL WAL is protected.

Four commercial, operational, and data-residency boundaries separating Fivetran, Airbyte Cloud/OSS, and owned S3 Iceberg CDC pipelines in 2026
Airbyte vs Fivetran at a glanceFivetranAirbyte
Pricing unitMonthly Active Rows, metered per connectionCredits on Standard and Plus, capacity-based Data Workers on Pro and Enterprise Flex
Entry priceFree plan up to 500,000 connection MAR a month; Standard adds a $5 monthly base per connectionStandard from $20 a month (5 credits), Plus from $189 a month (40 credits)
DeletesCount toward paid MAR since January 1, 2026Not billed per row on capacity plans
LicenseProprietary SaaSELv2 platform, MIT connectors
DeploymentVendor-managed SaaSSelf-hosted, Airbyte Cloud, or hybrid

Fivetran MAR pricing and the dbt Labs merger

Monthly Active Rows (MAR) is Fivetran's billing unit: the number of distinct rows inserted, updated or deleted on a connection in a month. Two things shape a Fivetran contract in late 2026.

  1. The merger. Fivetran and dbt Labs announced it on October 13, 2025 and closed it on June 1, 2026, an all-stock deal with close to $600M in combined ARR. Joint metadata lineage gets cleaner. The trade-off is that your EL connectors and your T orchestrator land on the same paper, which gives the vendor more room at renewal. Our advice is to keep dbt-core portable and keep high-volume CDC tables off per-row meters.
  2. Per-connection metering. Under Fivetran's usage-based pricing, the Free plan covers up to 500,000 connection MAR a month. The paid Standard, Enterprise and Business Critical plans meter MAR on each connection separately instead of pooling volume across the account. On Standard, every connection carries a $5 monthly base (1 to 1M MAR) and climbs its own rate curve, so a small connector never shares your largest table's volume discount. Deletes have counted toward paid MAR since January 1, 2026. A row counts once per month however many times it changes, but a schema migration that touches updated_at on 60M orders still adds 60M MAR to that month.

Fivetran vs Airbyte licensing: Core, credits and Data Workers

Airbyte is open core. The platform is under the Elastic License 2.0, the connectors are MIT, and you either run it yourself or pay Airbyte to run it.

  1. Airbyte Core on Kubernetes. The license is $0. Deploying the Airbyte Helm charts on Amazon EKS starts separate orchestrator, source and destination containers for every sync. At 35 connectors syncing every 15 minutes, that's 3,360 sync jobs a day. Pod churn, OOMKilled retries on wide rows, and the auth proxy you need because Core has no SSO or RBAC add up to about 36 hours a month of DevOps time in our planning.
  2. Airbyte Cloud. Standard costs $20 a month with 5 credits included, and Plus starts at $189 a month for 40 credits. Both are credit-based, so cost still follows volume. Pro and Enterprise Flex work differently. A Data Worker is a fixed slot of sync capacity sold at a flat monthly rate, and a 10x backfill queues on the worker instead of raising the bill. How many syncs one worker carries depends on the connector, the data volume and the destination, so size it with a load test. Enterprise Flex also supports hybrid deployment: Airbyte manages the control plane while regulated EU or HIPAA data stays in your own AWS account.

The PostgreSQL pgoutput replication slot trap on AWS RDS

Logical replication through pgoutput is how CDC tools read changes out of PostgreSQL. In our model most of the row volume comes from a handful of transactional PostgreSQL databases, the pattern we see in manufacturing telemetry (smart factory architectures) and warehouse management system development ledgers. Fivetran RDS PostgreSQL CDC, Airbyte Postgres CDC and Debezium all read PostgreSQL logical replication (pgoutput) on AWS RDS PostgreSQL, and they share one failure mode:

  • Stalled slots. Postgres keeps WAL files on the primary disk until every slot (pg_replication_slots) advances its restart_lsn. If a connector stalls on a schema mismatch, or tracks an idle database on a busy multi-database RDS instance with no heartbeat writes, retained WAL can grow by tens of GB an hour on write-heavy instances. Eventually the disk fills and writes stop, customer checkout included.
  • The defense has two parts. Set max_slot_wal_keep_size (available since PG13) so Postgres invalidates a broken slot before it blocks OLTP writes. The cost: an invalidated slot is marked lost and the connector needs a full resync. Then write a heartbeat every 60 seconds into a published table, or call pg_logical_emit_message, in the same database as the slot. The connector gets a change to acknowledge, so the slot keeps advancing.
Architectural dimensionFivetran (Enterprise)Airbyte (Pro / Ent. Flex)Airbyte Core (EKS)Owned PG17 CDC + S3 Iceberg
Primary billing unitPer-connection MAR, not pooledCapacity Data Workers (flat rate)$0 license plus K8s EC2 and SRE timeAWS ECS and S3 ($650/mo infra)
Backfill / UPDATE spikeEach touched row adds MAR (deletes included)$0 overage (queues on the worker)$0 vendor fee (risks pod OOM kills)$0 license ($0.023/GB S3 list price)
2026 commercial lock-inHigh (Fivetran + dbt Labs bundle)Moderate (open CDK connectors)Low (ELv2 platform, MIT connectors)No vendor lock-in (Apache 2.0)
SSO / SAML & RBACHigher tiersSSO from Plus, RBAC from ProNot in CoreNative AWS IAM + Okta BFF
EU / HIPAA isolationHigher tiers (check current plan features)Enterprise Flex hybrid deploymentSelf-hosted inside your VPCRuns entirely inside your VPC

Where Stitch, Hevo, Meltano and Estuary fit

Teams comparing Airbyte vs Fivetran usually shortlist a few others. Stitch and Hevo are managed ELT services in the Fivetran mould, Meltano is an open-source, code-first tool you run yourself, and Estuary focuses on streaming CDC. Ask each one the same three questions: what the billing unit is, who runs the workers, and what happens to the bill during a backfill. We haven't modeled their pricing here.

Owned CDC blueprint: PostgreSQL slot guard and S3 Iceberg staging

Instead of paying a modeled $378,000 in MAR fees to copy your own PostgreSQL tables and five core SaaS APIs (Stripe, Salesforce, HubSpot, NetSuite, Zendesk), you can own the pipeline. It has two parts. The first guards the pgoutput replication slot so it can't fill the primary disk. The second is a small worker that compacts CDC micro-batches and stages them on Amazon S3 for Apache Parquet and Apache Iceberg tables. From there, ClickHouse S3Queue ingests new files as they land, or Snowflake MERGE loads them.

The PostgreSQL side is small. Set max_slot_wal_keep_size = '40GB' in postgresql.conf or the RDS parameter group, then add a heartbeat table, a column-scoped publication (listing only the columns the warehouse needs, with publish_via_partition_root enabled) and a view that reports how much WAL each slot is holding. Column lists and publish_via_partition_root need PG15 or later, and the design targets PG17.

The rest of the pipeline is a TypeScript worker. These are the design decisions that matter for cost and safety across data pipelines, multi-tenant SaaS and event-driven architecture setups:

Design decisionWhere it's enforcedWhat it buys you
WAL ceiling on every logical slotDatabase (parameter group)Postgres drops a stalled slot before the disk fills and OLTP writes stop. The price is a full resync for that connector.
Heartbeat every 60 seconds, then a slot health checkWorker writes the heartbeat row; the view reports retentionIdle slots keep advancing, and the worker raises an alert at 16 GB retained or a lost slot, well under the 40 GB ceiling.
Column lists on the publicationDatabasePII columns such as customer_email and card_last4 never enter the replication stream.
Compact each micro-batch by primary keyWorkerOnly the latest change per row (by LSN) is kept, so a row that moves through four statuses in 60 seconds lands as one upsert.
Idempotent staging and loadS3 key per tenant, connector, date and LSN; warehouse MERGE checks commit timeReplaying a batch after a crash can't overwrite newer rows or duplicate them.
Checkpoint only after the load succeedsPer-tenant watermark table under row-level securityThe high LSN and S3 manifest are written in one transaction after the warehouse load, so a failure replays the batch instead of skipping it.

To keep the first version simple, the worker stages gzipped NDJSON. Parquet and Iceberg compaction comes later. No change in this pipeline is billed per row, and at this volume that is the reason to own it.

Fivetran MAR renewals spiking after the dbt merger, or Airbyte EKS pods OOM-crashing?
Send your active connector list, monthly CDC row churn, sync SLA, and warehouse target. Our data platform architects will return a 36-month TCO comparison.
Audit your ELT & CDC TCO

Airbyte vs Fivetran 36-month TCO at 180M MAR/mo across 35 connectors

The model is a mid-market data platform over 36 months. It moves 180M Monthly Active Rows a month across 35 production connectors: five high-churn AWS RDS PostgreSQL CDC databases generating 155M MAR a month including deletes, and 30 Stripe, Salesforce, HubSpot, NetSuite and webhook SaaS sources generating 25M. Requirements are a 15-minute SLA, SAML SSO with RBAC, EU and US data residency, and SOC 2 Type II plus PCI DSS v4.0.1 compliance. Neither vendor publishes list prices at this volume, so the $10,500 and $5,000 monthly subscriptions below are modeled assumptions. Swap in your own quotes.

Labor compares Idealogic's $75/hr senior CEE rate with a $165/hr US agency rate (2.2x). Both are agency rates with overhead and margin included, so they sit well above wage data. For context, EU whole-economy labour cost averaged EUR 34.9 an hour in 2025, and the BLS publishes US software developer wages. AWS figures are us-east-1 EKS and S3 list prices, so check them against current pricing:

  1. Option A: Fivetran Enterprise (35 connections, 180M MAR/mo, 15-minute sync, SSO)
    • Setup and ops (732h = $54,900 CEE; $120,780 US): 192h of Terraform connector IaC, column hashing and dbt contracts ($14,400 CEE / $31,680 US), plus 15h/mo x 36 = 540h of MAR and schema governance ($40,500 CEE / $89,100 US).
    • 36-month subscription ($10,500/mo x 36 = $378,000, modeled): five CDC and 30 SaaS connections, each on its own MAR curve, deletes and status updates included. Total = $432,900 CEE ($498,780 US).
  2. Option B: Airbyte Pro / Enterprise Flex (6 Data Workers, hybrid deployment, SAML and RBAC)
    • Setup and ops (732h = $54,900 CEE; $120,780 US): 192h for the hybrid deployment and 35 connector setups ($14,400 CEE), plus 15h/mo x 36 = 540h of schema and CDK ops ($40,500 CEE).
    • 36-month Enterprise Flex ($5,000/mo x 36 = $180,000, modeled): six Data Workers is our sizing assumption for 35 connectors at a 15-minute cadence, and you should confirm it with a load test. Row spikes add no overage. Total = $234,900 CEE ($300,780 US).
  3. Option C: self-hosted Airbyte Core on AWS EKS ($0 license, heavy K8s maintenance)
    • Setup and K8s ops (1,560h = $117,000 CEE; $257,400 US): 264h of EKS Helm setup and an OIDC auth wrapper ($19,800 CEE / $43,560 US), plus 36h/mo x 36 = 1,296h of pod OOMKilled tuning, upgrades and RBAC maintenance ($97,200 CEE / $213,840 US).
    • 36-month AWS EKS ($1,350/mo x 36 = $48,600): the EKS worker cluster, with no license fee. Total = $165,600 CEE ($306,000 US). At $165/hr US rates, self-hosting costs $5,220 more than the modeled Enterprise Flex option.
  4. Option D: owned PG17 pgoutput CDC + S3 Apache Iceberg / Parquet ELT (S3Queue / MERGE)
    • Turnkey build (580h at $75/hr CEE = $43,500; at $165/hr US = $95,700): WAL-guarded pgoutput CDC (cdc_heartbeat plus max_slot_wal_keep_size), the S3 Iceberg/Parquet compactor, typed SaaS REST pollers with dlt workers for the long tail, and idempotent S3Queue / MERGE loaders.
    • 36-month AWS compute and S3 ($650/mo x 36 = $23,400): stateless ECS/EKS CDC tasks ($460/mo) plus S3 Iceberg staging and a Gateway Endpoint ($190/mo).
    • 36-month SRE and schema ops (15h/mo x 36 = 540h): $40,500 CEE ($89,100 US). Total = $107,400 CEE ($208,200 US).
Worked 36-month ELT & CDC TCO ($k) at 180M MAR/mo across 35 connectors (15-min SLA, CEE engineering): Owned PG17 + S3 Iceberg saves $325.5k (75.2%) vs Fivetran Ent. and $127.5k vs Airbyte Ent. Flex. Vendor subscriptions are modeled assumptions, not list prices
36-month line item (180M MAR, 35 conn.)Option A: Fivetran Ent.Option B: Airbyte Pro/Ent.Option C: Airbyte Core EKSOption D: Owned PG17 + S3Economic driver
1. Setup / build ($75/hr CEE)$14,400 (192h; $31.7k US)$14,400 (192h; $31.7k US)$19,800 (264h; $43.6k US)$43,500 (580h; $95.7k US)pgoutput + Iceberg + pollers
2. 36-mo SaaS (modeled) / AWS infra$378,000 ($10,500/mo)$180,000 ($5,000/mo)$48,600 ($1,350/mo EKS)$23,400 ($650/mo ECS + S3)No per-row fees, no connector pod fleet
3. 36-mo data / SRE ops ($75/hr)$40,500 (15h/mo; 540h)$40,500 (15h/mo; 540h)$97,200 (36h/mo; 1,296h)$40,500 (15h/mo; 540h)Core K8s upkeep takes 36h/mo
Total 36-month TCO (CEE)$432,900$234,900$165,600$107,400 (CapEx incl.)Saves $127.5k to $325.5k vs SaaS ($58.2k vs Core)
Total 36-month TCO (US $165/hr)$498,780$300,780$306,000 (exceeds SaaS)$208,200 (CapEx incl.)Wins by $92.6k to $290.6k US
Year-2+ annual run-rate (CEE)$139,500/yr$73,500/yr$48,600/yr$21,300/yrSaves $52.2k to $118.2k/yr vs SaaS ($27.3k vs Core)
Table format and lock-inProprietary sync + dbt CloudCapacity workers + CDKSelf-hosted K8sOpen S3 Iceberg + dbt-corePortable lakehouse files

Sensitivity check: the 80/20 split, sub-10M MAR startups and CapEx payback

  • Below 10M MAR a month across fewer than 10 low-churn SaaS apps: stay on Fivetran Free or Standard, or Airbyte Cloud Standard (under $400/mo). A custom pipeline is overkill at seed volume.
  • The 80/20 split is the pragmatic middle. Move the five high-churn PostgreSQL CDC databases (86% of MAR in this model) to the owned pipeline and leave the 30 SaaS connectors on Airbyte Cloud or Fivetran.
  • Payback at 180M MAR a month ($75/hr CEE): Option D needs $29,100 more initial CapEx than a managed setup ($43,500 minus $14,400). In return, monthly OpEx drops from $11,625 (Fivetran) and $6,125 (Airbyte Ent. Flex) to $1,775 ($650 AWS plus $1,125 ops). The extra CapEx pays back in 2.95 months vs Fivetran Ent., 6.69 months vs Airbyte Ent. Flex, and 10.42 months vs Airbyte Core on EKS.

Fivetran or Airbyte is still the right answer in plenty of cases. The owned build only pays off when a few databases produce most of your rows and you have engineers to own it. If your volume is spread across dozens of long-tail SaaS APIs, maintaining those connectors yourself costs more than a subscription, and that's the work both vendors do well.

2026 ELT and CDC pipeline build costs (senior CEE vs US)

Our custom software development, SaaS development, and software development consulting teams deliver fixed-milestone data pipelines at $75/hr senior CEE rates, against $165/hr at US agencies (2.2x):

Engagement tierTimelineHoursSenior CEE ($75/hr)US agency ($165/hr)Deliverables
PG17 CDC and WAL guard sprint2 to 3 wks120 to 180h$9,000 to $13,500$19,800 to $29,700pgoutput publication, cdc_heartbeat and WAL alerts
Managed Airbyte / Fivetran IaC3 to 5 wks180 to 260h$13,500 to $19,500$29,700 to $42,900Hybrid deployment workers, Terraform IaC and dbt-core silver layer
Owned CDC + S3 Iceberg (580h)8 to 11 wks480 to 680h$36,000 to $51,000 ($43.5k)$79,200 to $112,200 ($95.7k)PG17 CDC, S3 Iceberg compactor, SaaS pollers and MERGE
Multi-region streaming lakehouse12 to 16 wks800 to 1,120h$60,000 to $84,000$132,000 to $184,800Redpanda/Kafka CDC mesh, Iceberg REST catalog and PII vault
Dedicated data platform podOngoing240h/mo (2)From $18,000/moFrom $39,600/moSenior CDC, ClickHouse/Snowflake and dbt engineers

Inside the worked 580-hour ($43,500 CEE vs $95,700 US) owned CDC and Iceberg stack

Worked model: 580-hour ($43,500 at $75/hr CEE vs $95,700 at $165/hr US) owned PostgreSQL 17 pgoutput CDC + S3 Apache Iceberg / Parquet ELT pipeline
  • pgoutput guard (25% / 145h / $10,875 CEE vs $23,925 US): column-scoped publications, the cdc_heartbeat cron, max_slot_wal_keep_size circuit breakers and the FORCE RLS watermark ledger.
  • S3 Iceberg compaction (25% / 145h / $10,875 CEE vs $23,925 US): in-window PK mutation deduplication, a ZSTD Parquet/Iceberg partition writer, a schema-evolution registry and S3 Gateway Endpoint IaC.
  • SaaS REST pollers (20% / 116h / $8,700 CEE vs $19,140 US): rate-limit-aware Stripe, Salesforce, HubSpot and NetSuite incremental cursor and webhook workers, plus dlt adapters for the long tail.
  • MERGE and S3Queue (20% / 116h / $8,700 CEE vs $19,140 US): ClickHouse S3Queue into ReplacingMergeTree, idempotent Snowflake/BigQuery MERGE, and open dbt-core models.
  • Audit and cutover (10% / 58h / $4,350 CEE vs $9,570 US): a zero-downtime historical snapshot with LSN handoff, row-hash reconciliation against Fivetran or Airbyte, and PagerDuty runbooks.

Post-merger, paying unpooled Fivetran MAR fees to copy your own PostgreSQL tables into cloud storage makes zero sense, while thirty-five JVM pods on Airbyte OSS trade SaaS bills for Kubernetes OOM firefighting. A WAL-guarded pgoutput to S3 Iceberg pipeline cuts three-year TCO seventy-five percent versus Fivetran, in open format.

Artem Zaitsev · CTO at Idealogic

Airbyte vs Fivetran decision checklist for CTOs

Our view on Airbyte vs Fivetran comes from pipelines we've shipped: transactional CDC, financial reconciliation and supply-chain ELT.

Five checks settle Airbyte vs Fivetran vs an owned PostgreSQL and S3 Iceberg pipeline in 2026:

  1. Under 10M MAR a month across fewer than 10 low-churn SaaS tools? Use Fivetran Standard or Airbyte Cloud (under $400/mo).
  2. Dozens of long-tail SaaS connectors and EU residency rules? Get a quote for Airbyte Enterprise Flex and its Data Workers.
  3. Assuming Airbyte Core is free? Price 36 hours a month of K8s SRE work first. At US rates the three-year total ($306k) passes our modeled Airbyte Flex figure ($300.8k).
  4. Running pgoutput CDC on AWS RDS? Set max_slot_wal_keep_size and write a cdc_heartbeat row every 60 seconds before a stalled slot fills the disk.
  5. Around 180M MAR a month, mostly PostgreSQL or MySQL CDC plus core SaaS APIs? Build the owned pgoutput and S3 Iceberg pipeline (ClickHouse S3Queue or Snowflake MERGE, with dbt-core). In our model that saves $127.5k to $325.5k over 36 months. Contact our team for a 48-hour ELT TCO audit.
Weighing Fivetran MAR renewals, Airbyte Data Workers, or owned S3 Iceberg CDC?
Share your monthly active rows, connector mix, Postgres WAL volume, and warehouse stack. Our senior CEE data architects will return a 36-month TCO blueprint within 48 hours.
Request a 48-hour ELT & CDC audit

Occasional field notes on building software, no spam

Protected by Cloudflare Turnstile · Privacy · Terms

Artem Zaitsev
Artem ZaitsevCTO · Idealogic Development OÜ

CTO and Co-founder at Idealogic. Directing engineering, systems architecture, security, and full-stack delivery across web, mobile, and blockchain.

Frequently asked questions

  • Pick Fivetran when a lean analytics team wants managed SaaS connectors across dozens of business apps and monthly row churn is low. Pick Airbyte Pro or Enterprise Flex when high-churn PostgreSQL or MySQL CDC tables would run up per-row billing, or when European data residency calls for a hybrid deployment.

  • The merger was announced on October 13, 2025 and closed on June 1, 2026, an all-stock deal with close to 600 million dollars in combined ARR. ELT ingestion and SQL transformation now sit with one commercial vendor. Fivetran MAR ingestion and dbt Cloud seats can now renew on one contract, which raises lock-in, so we advise keeping open dbt Core and portable Iceberg pipelines.

  • Fivetran meters Monthly Active Rows (MAR) on each connection separately instead of pooling row volume across the account. On the Standard plan every connection carries a five dollar monthly base from one up to one million MAR and climbs its own rate curve. Deletes have counted toward paid MAR since January 1, 2026. A row counts once per month however often it changes, but an ORM backfill that touches updated_at on sixty million rows still adds sixty million MAR to that month.

  • Airbyte Core is free to self-host on Kubernetes (ELv2 platform, MIT connectors). It ships without SSO or RBAC and starts separate source, destination, and orchestrator pods for every sync. Airbyte Cloud Standard and Plus are credit-based, so cost follows volume. Pro and Enterprise Flex sell flat-rate Data Workers, fixed slots of sync capacity whose price doesn't move when row mutations spike. How many syncs one worker carries depends on connector, volume, and destination, so size it with a load test.

  • The license is zero dollars, but the labor isn't. In our model, Airbyte Core on Kubernetes needs about thirty-six SRE hours a month: connector pods that get OOMKilled on wide JSONB tables, upgrades, connector patches, and a reverse-proxy wrapper to stand in for the SSO and RBAC that Core lacks. At $165 per hour US engineering rates, that labor pushes the three-year total past our modeled Airbyte Enterprise Flex quote.

  • PostgreSQL logical replication keeps every Write-Ahead Log (WAL) segment on primary storage until the downstream CDC consumer acknowledges it. If a Fivetran, Airbyte, or Debezium connector pauses, or tracks a quiet database on a busy RDS instance with no heartbeat writes, the slot keeps retaining WAL until the disk is full.

  • Set max_slot_wal_keep_size in postgresql.conf or the RDS parameter group and Postgres invalidates a lagging replication slot before it fills the disk and blocks OLTP writes. The trade-off: an invalidated slot is lost and the connector needs a full resync. Pair it with a sixty second job that updates a small heartbeat table in your publication or calls pg_logical_emit_message. That gives the connector a change to acknowledge, so the slot keeps advancing while tenant schemas sit idle.

  • With Idealogic's senior Central and Eastern European data engineers at $75 per hour (versus $165 per hour at US agencies), a PostgreSQL CDC WAL guard sprint runs $9,000 to $13,500 and a managed Airbyte Flex or Fivetran setup costs $13,500 to $19,500. The full 580 hour owned pgoutput CDC plus S3 Apache Iceberg and Parquet pipeline loading ClickHouse or Snowflake costs $43,500 ($95,700 at US rates). In our 36-month model it saves $127,500 to $325,500 against the managed options.

Still unanswered
Ask us directly

A senior engineer replies under 4 hours.