ELT Pricing in 2026: The Unit Decides Which Data Is Expensive
Fivetran counts rows that changed and does not charge for the initial load. Estuary counts gigabytes and charges per connector instance in both directions. Airbyte stopped publishing its rate entirely. Five tools, five units, and the one that fits depends on how your biggest table churns.
Every tool here will pull Salesforce, Postgres and Stripe into your warehouse. That is not the decision. The decision is that none of them agree on what they are charging you for, and the unit each one picked determines which shape of data is expensive for you.
A 500 GB table where 2% of rows change each month is cheap on a row-change meter and expensive on a gigabyte meter. A small table that updates constantly is the reverse. Neither vendor is cheaper; they are cheaper for different data. Prices below came off the vendor pricing pages and the GitHub API on 23 August 2026.
Fivetran: the meter skips the part you would expect it to charge for
Monthly Active Rows counts a primary key that was inserted or updated at least once in the calendar month, once, no matter how many times it changed. Two exclusions on Fivetran’s own pricing page matter more than the rate: unchanged rows retrieved during a scheduled re-sync do not count, and the initial bulk load when you set up a connector does not count at all.
That is the opposite of every volume-priced tool in this post, where the first sync of a large table is the single biggest line you will ever see. If your problem is “we need to get a 500 GB table into the warehouse,” Fivetran charges you nothing for the getting.
What it does charge is harder to pin down than the comparison tables suggest. Fivetran’s own worked example, Standard plan, a company of 1 to 200 employees:
| Connector | Median MAR | Median monthly list price |
|---|---|---|
| Facebook Ads | 34,479 | $17.23 |
| Google Analytics 4 | 21,993 | $10.99 |
| Marketo | 847,574 | $423.78 |
| Google Ads | 88,240 | $44.12 |
Every one of those works out to almost exactly $500 per million MAR. Fivetran says the rate falls as volume rises but does not publish the curve, so any comparison quoting a few dollars per million is describing a tier you cannot see from outside. Worth noting the page’s own arithmetic: it gives the total for those four lines as $549.36, which is about $53 more than the four prices add up to. Read your invoice rather than the example.
Two more things sit on top. There is a $5 base charge on every standard connection with usage between 1 and 1M MAR, so a fleet of small connectors has a floor that the per-million rate never shows. And MAR is now one of three meters: Transformations bill per model run ($0.01 above the free 5,000, falling to $0.002 past 100,000) and Activations, the reverse-ETL product, bill their own MAR at a much steeper rate. Their example puts 12,007 activation MAR at $202.01, which is roughly $16,800 per million.
The Free plan is real, though: 500,000 MAR, 5,000 model runs, 3,500 activation MAR, and no $5 base charge.
Airbyte changed the unit, which is why the old numbers do not check out
Comparisons of Airbyte still quote $10 per GB for database sources and $15 per million rows for APIs. Those rates are not on the pricing page any more. What is there: Core, self-managed and free; Standard, “starting at $10 / month” with volume-based pricing and no published rate; Pro, priced on Data Workers, capacity units that each run roughly three concurrent syncs, explicitly sold as not scaling with data volume.
The headline of that page is now agent pricing, billed in Agent Operations at $29 a month for 5,000 and $299 for 10,000, with overage at $0.004 to $0.005 each. Data replication has become the second product on its own pricing page.
The other thing worth correcting: Airbyte is not open source in the usual sense. The repository is Elastic License 2.0, which permits self-hosting but not providing the software to others as a hosted service. That distinction rarely appears in comparisons that group it with Meltano.
Estuary: count the connector instances, not the connectors
Estuary’s rate is refreshingly plain: $0.50 per GB moved, plus $100 a month for each of the first six connector instances and $50 for each after that. The free tier is 10 GB a month across up to two connector instances, with no time limit.
The trap is in the definition. A connector instance is a connection to a source or a destination, a capture or a materialization. Four sources landing in one warehouse is not four instances, it is five, so $500 a month before a single gigabyte moves. A team pricing this as “four connectors, $400” will be wrong in the direction that matters.
Where it earns that: this is the streaming, CDC-first option in the group. If your requirement is that the warehouse not be four hours stale, the batch tools do not solve it at any price. If daily is fine, you are buying latency you will not use.
Stitch: the one you can price before you sign
Standard starts at $100 a month with a row selector running from 5 million to 300 million. Advanced is $1,500 a month for 100 million rows and three destinations, Premium $3,000 for a billion rows and five destinations, both billed annually. Rows replicated, fixed tiers, no unit conversion required.
The reason it is quiet: Stitch went to Talend in 2018 and Talend to Qlik in 2023, and it is built on the Singer spec, whose own getting-started repository has not been touched since August 2025. That is a fine trade if your sources are mainstream and you value knowing the bill in advance. It is the wrong pick if you need a new SaaS connector or sub-hour latency.
If you are in healthcare, the HIPAA BAA is on Advanced and Premium. The $100 plan is not the compliance plan.
Meltano: free, MIT, and still someone’s job
Meltano Core is MIT licensed, not Apache 2.0 as most comparisons say, and it is alive: version 4.2.2 in July 2026 with commits this week. Pipelines are config files, they live in git, they run in CI, and there is no license cost or metered unit anywhere.
The bill arrives as attention. You run the infrastructure, own the taps and targets, debug Singer connector edge cases and carry the pager. That is genuinely cheaper than any of the metered options at high volume, and genuinely more expensive than all of them if you chose it to save money and then spent a senior engineer’s month on uptime.
The license line, which outlasts every rate above
| License | Can you self-host it? | |
|---|---|---|
| Meltano | MIT | Yes, without conditions |
| Airbyte | Elastic License 2.0 | Yes, but not as a service for others |
| Estuary Flow | BSL, with Apache 2.0 for some directories | Partly, under the BSL terms |
| Fivetran, Stitch | Closed | No |
Three tools that get called open source in the same sentence, three different answers.
The question that actually decides it
Take your largest table. Find what fraction of its rows change in a month, and how many gigabytes it holds.
High churn on a modest table is the Fivetran problem: the MAR meter tracks change, and history mode or CDC on a chatty dimension table is where bills go vertical. Large and quiet is the volume-priced problem: you pay $0.50 a GB on Estuary to move 200 GB once, and Fivetran charges nothing for that same first load. Wide but small is the connector-fee problem, where Estuary’s per-instance charge and Fivetran’s $5 floor both bite before any data does.
None of that requires a trial to estimate. It requires one query against your own database, and it is the only number in this post that is about you.