Airbyte VS Fivetran

Estuary was founded in 2019. But the core technology, the Gazette open source project, has been evolving for a decade within the Ad Tech space, which is where many other real-time data technologies have started.

Estuary Flow is the only real-time and ETL data pipeline vendor in this comparison. There are some other ETL and real-time vendors in the honorable mention section, but those are not as viable a replacement for Fivetran.

While Estuary Flow is also a great option for batch sources and targets, where it really shines is any combination change data capture (CDC), real-time and batch ETL or ELT, and loading multiple destinations with the same pipeline. Estuary Flow currently is the only vendor to offer a private cloud deployment, which is the combination of a dedicated data plane deployed in a private customer account that is managed as SaaS by a shared control plane. It combines the security and dedicated compute of on prem with the simplicity of SaaS.

CDC works by reading record changes written to the write-ahead log (WAL) that records each record change exactly once as part of each database transaction. It is the easiest, lowest latency, and lowest-load for extracting all changes, including deletes, which otherwise are not captured by default from sources. Unfortunately ELT vendors like Airbyte, Fivetran, Meltano, and Hevo all rely on batch mode for CDC. This puts a load on a CDC source by requiring the write-ahead log to hold onto older data. This is not the intended use of CDC and can put a source in distress, or lead to failures.

Estuary Flow has a unique architecture where it streams and stores streaming or batch data as collections of data, which are transactionally guaranteed to deliver exactly once from each source to the target. With CDC it means any (record) change is immediately captured once for multiple targets or later use. Estuary Flow uses collections for transactional guarantees and for later backfilling, restreaming, transforms, or other compute. The result is the lowest load and latency for any source, and the ability to reuse the same data for multiple real-time or batch targets across analytics, apps, and AI, or for other workloads such as stream processing, or monitoring and alerting.

Estuary Flow also has broad packaged and custom connectivity, making it one of the top ETL tools. It has 150+ native connectors that are built for low latency and/or scale. While may seem low, these are connectors built for low latency and scale. In addition, Estuary is the only vendor to support Airbyte, Meltano, and Stitch connectors as well, which easily adds 500+ more connectors. Getting official support for the connector is a quick “request-and-test” with Estuary to make sure it supports the use case in production. Most of these connectors are not as scalable as Estuary-native, Fivetran, or some ETL connectors, so it’s important to confirm they will work for you. Flow’s support for TypeScript and SQL also enables ETL.

• Modern data pipeline: Estuary Flow has the best support for schema drift, evolution, and automation, as well as modern DataOps.
• Modern transforms: Flow is also both low-code and code-friendly with support for SQL, TypeScript (and Python coming) for ETL, and dbt for ELT.
• Lowest latency: Several ETL vendors support low latency. But of these Estuary can achieve the lowest, with sub-100ms latency. ELT vendors generally are batch only. 
• High scale: Unlike most ELT vendors, leading ETL vendors do scale. Estuary is proven to scale with one production pipeline moving 7GB+/sec at sub-second latency.
• Most efficient: Estuary alone has the fastest and most efficient CDC connectors. It is also the only vendor to enable exactly-and-only-once capture, which puts the least load on a system, especially when you’re supporting multiple destinations including a data warehouse, high performance analytics database, and AI engine or vector database.
• Deployment options: Of the ETL and ELT vendors, Estuary is currently the only vendor to offer open source, private cloud, and public multi-tenant SaaS.
• Reliability: Estuary’s exactly-once transactional delivery and durable stream storage makes it very reliable.
• Ease of use: Estuary is one of the easiest to use tools. Most customers are able to get their first pipelines running in hours and generally improve productivity 4x over time. 
• Lowest cost: for data at any volume, Estuary is the clear low-cost winner in this evaluation. Rivery is second.
• Great support: Customers consistently cite great support as one of the reasons for adopting Estuary.

• On premises connectors: Estuary has 150+ native connectors and supports 500+ Airbyte, Meltano, and Stitch open source connectors. But if you need on premises app or data warehouse connectivity make sure you have all the connectivity you need.
• Graphical ETL: Estuary has been more focused on SQL and dbt than graphical transformations. While it does infer data types and convert between sources and targets, there is currently no graphical transformation UI.

Of the various ELT and ETL vendors, Estuary is the lowest total cost option. Estuary only charges $0.50 per GB of data moved from each source or to each target, and $100 per connector per month. So you can expect to pay a minimum of a few thousand per year. But it quickly becomes the lowest cost pricing. Rivery, the next lowest cost option, is the only other vendor that publishes pricing of 1 RPU per 100MB, which is $7.50 to $12.50 per GB depending on the plan you choose. Estuary becomes the lowest cost option by the time you reach the 10s of GB/month. By the time you reach 1TB a month Estuary is 10x lower cost than the rest.

Airbyte was founded in 2020 as an open-source data integration company, and launched its cloud service in 2022.

Airbyte started as a Singer-based ELT tool, but has since changed their protocol and connectors to be different. Airbyte has kept Singer compatibility so that it can support Singer taps as needed. Airbyte has also kept many of the same principles, including being batch-based. This is eventually where Airbyte’s limitations come from as well.

If you go by pricing calculators and customers, Airbyte is the second-lowest-cost vendor in the evaluation after Estuary.

• Ease of use: Airbyte is an easy-to-use (harder to operate), modern ELT product.
• Open source: Airbyte is open source, which means you can self-host. In addition, open source has fewer limits, such as being able to run more frequent batch intervals.
• Low cost: Airbyte Cloud is one of the lowest-cost options for batch ETL.
• Widely used: Even though Airbyte is only 4 years old, it is widely used. Most of the customers use the open-source version. The official 1.0 product launch, the big milestone for any open-source project, was September, 2024.

• Only 50+ managed connectors: While Airbyte lists 300+ connectors, only 50+ of these are connectors actively developed by Airbyte. The rest are open source connectors listed as Marketplace connectors for Airbyte Cloud. Make sure you evaluate the connectors based on your needs.
• High Latency: While Airbyte has CDC source connectors mostly built on Debezium (except for a new Postgres CDC connector), and also has Kafka and Kinesis source connectors, everything is loaded in intervals of 5 minutes or more with the open source version. Airbyte Cloud is much worse. It only supports 1+ hour intervals and one source connector at a time. There is no staging or storage, so if something goes wrong with either source or target the pipeline stops. 
  Also, Airbyte is pulling from source connectors in batch intervals. When using CDC, this can put a load on the source databases. In addition, because all Airbyte CDC connectors (other than the new Postgres connector) use Debezium, it is not exactly-once, but at-least-once guaranteed delivery. Latency is also made worse with batch ELT because you need to wait for loads and transforms in the target data warehouse.
• Reliability: There are some issues with reliability you will need to manage. Most CDC sources, because they’re built on Debezium, only ensure at-least-once delivery. It means you will need to deduplicate (dedup) at the target. Airbyte does have both incremental and deduped modes you can use though. You just need to remember to turn them on. Also, Debezium does put less of a load on a source because it uses Kafka. This does make it less of a load on a source than Fivetran CDC. A bigger reliability issue is failure of under-sized workers. There is no scale-out option. Once a worker gets overloaded you will have reliability issues (see scalability.) There is also no staging or storage within an Airbyte pipeline to preserve state. If you need the data again, you’ll have to re-extract from the source.
• Scalability: Airbyte is not known for scalability. It has scalability issues that may not make it suitable for your larger workloads. For example, each airbyte operation of extracting from a source or loading into a target is done by one worker. The source worker is generally the most important component, and its most important component is memory. The source worker will read up to 10,000 records into memory, which could lead to GBs of RAM. By default only 25% of each instance’s memory is allocated to the worker container, which you have little control over in Airbyte Cloud.
  Airbyte is working on scalability. The new PostgreSQL CDC connector does have improved performance. Its latest benchmark as of the time of this writing produced 9MB/sec throughput, higher than Fivetran’s (non HVR) connector. But this is still only 0.5TB a day or so depending on how loads vary throughout the day.
• ELT only: Airbyte cloud supports dbt cloud. This is different from dbt core used by Fivetran. If you have implemented on dbt core in a way that makes it portable (which you should) the move can be relatively straightforward. But if you want to implement transforms outside of the data warehouse, Airbyte does not support that.
• DataOps: Airbyte provides UI-based replication designed for ease of use. It does not give you an “as code” mode that helps with automating end-to-end pipelines, adding tests, or managing schema evolution. But there is Octavia which acts as a CLI for Airbyte.

Airbyte starts at $10 per GB of data moved from a database, and $15 per million rows of data moved via an API (or custom source.) There are volume-based discounts. You do pay for backfills as well. While this is solely volume-based, Estuary becomes less expensive for 10s of GB per month.

Fivetran was founded in 2012 by data scientists who wanted an integrated stack to capture and analyze data. The name was a play on Fortran and meant to refer to a programming language for big data. After a few years the focus shifted to providing just the data integration part because that’s what so many prospects wanted. Fivetran was designed as an ELT (Extract, Load, and Transform) architecture because in data science you don’t usually know what you’re looking for, so you want the raw data. 

In 2018, Fivetran raised their series A, and then added more transformation capabilities in 2020 when it released Data Build Tool (dbt) support. That year Fivetran also started to support CDC. Fivetran has since continued to invest more in CDC with its HVR acquisition.

Fivetran’s design worked well for many companies adopting cloud data warehouses starting a decade ago. While all ETL vendors also supported “EL” and it was occasionally used that way, Fivetran was cloud-native, which helped make it much easier to use. The “EL” is mostly configured, not coded, and the transformations are built on dbt core (SQL and Jinja), which many data engineers are comfortable using.

• Ease of Use: Fivetran is modern SaaS ELT with an easy-to-use UI, especially in comparison to more traditional eTL tools. It allows you to set up a data pipeline without coding.
• Pre-built Connectors: Fivetran offers nearly 300 native connectors and an additional 300+ “lite” connectors based on APIs.
• Scalability: Fivetran is known for scaling better than many of its competitors.
• Integration with dbt: Fivetran has done a good job of integrating dbt core into the Fivetran platform.
• Focus on replication: Fivetran is good at data extraction and loading (EL), even if it is batch only, making it a strong choice if your primary goal is to efficiently move data into your warehouse for analysis.
• Advanced schema evolution: Fivetran and Estuary are the two leading vendors with support for automating how changes in sources are passed through to destinations.

• Latency: While Fivetran uses change data capture at the source, it is batch CDC, not streaming. Enterprise-level is guaranteed to be 15 minutes of latency. Business critical is 1 minute of latency, but costs more than 2x the standard edition. Its ELT architecture can also be slowed down by the target load and transformation times.
• Costs: Another major complaint are Fivetran’s high vendor costs, which have been 5x the cost of Estuary as stated by customers. Fivetran costs are based on monthly active rows (MAR) that change at least once per month. This may seem low, but for several reasons (see below and the pricing section) it can quickly add up.
• Unpredictable costs: Another major reason for high costs is that MARs are based on Fivetran’s internal representation of rows, not rows as you see them in the source.
  For some data sources you have to extract all the data across tables, which can mean many more rows. Fivetran also converts data from non-relational sources such as SaaS apps into highly normalized relational data. Both make MARs and costs unexpectedly soar. This also does not account for the initial load where all rows count.
• Reliability: Another reason for replacing Fivetran is reliability. Customers have struggled with a combination of alerts of load failures, and subsequent support calls that result in a longer time to resolution. There have been several complaints about reliability with MySQL and Postgres CDC, which is due in part because Fivetran uses batch CDC. Fivetran also had a 2.5 day outage in 2022. Make sure you understand Fivetran’s current SLA in detail. Fivetran has had an “allowed downtime interval” of 12 hours before downtime SLAs start to go into effect on the downtime of connectors. They also do not include any downtime from their cloud provider.
• Deployment options: while Fivetran claims private cloud as an option, it’s not really an option. Its private cloud deployment requires some installation work and only supports 8 sources and 5 destinations. There is also a self-hosted option for HVR only.
• Support: Customers also complain about Fivetran support being slow to respond. Combined with reliability issues, this can lead to a substantial amount of data engineering time being lost to troubleshooting and administration.
• DataOps: Fivetran does not provide much control or transparency into what they do with data and schema. They alter field names and change data structures and do not allow you to rename columns. This can make it harder to migrate to other technologies. Fivetran also doesn’t always bring in all the data depending on the data structure, and does not explain why.
• Roadmap: Customers frequently comment Fivetran does not reveal as much of a future direction or roadmap compared to the others in this comparison, and do not adequately address many of the above points.

Fivetran's pricing is based on monthly active rows (MAR). This can be very unpredictable because MARs are based on Fivetran’s internal representation of data, not yours. Any non-relational or nested data gets turned into highly normalized rows that raise costs.

Lower latency is also very expensive. To reduce latency from 1 hour to 15 minutes can cost you 33-50% more (1.5x) per million MAR, and 100% (2x) or more to reduce latency to 1 minute, which is rarely deployed. Some connectors require all data to be extracted each time, which also becomes more expensive as you lower latency and increase the number of extracts.

Even then, you still have the latency of the data warehouse load and transformations. The additional costs of frequent ingestions and transformations in the data warehouse can also be expensive and take time. Companies often keep latency high to save money.

While a small deployment (2M MARs/month) can cost $700-$2667, 10M MARs/month get you into $10K a month. It is not unheard of for Fivetran costs to reach 6 digits annually, especially with certain high-cost connectors that end up having many more MARs.