What it does

Continuous replication of Dataverse tables into storage you own — Azure Data Lake Storage Gen2, Azure Synapse Analytics, or both. It's a push: creates, updates and deletes are written to the destination as they happen, with no refresh schedule to configure. This is the feature formerly called Export to data lake, renamed in May 2021.

Key facts

• You bring the storage account and, if you want it, the Synapse workspace. Synapse Link handles the provisioning and configuration around them.
• Standard and custom tables are both supported, and CUD transactions replicate — delete an account row in Dataverse and the deletion shows up in the destination.
• It relies on the change tracking attribute on the table to work out what's changed since the last sync.
• Data lands in Common Data Model format by default. To link an existing Synapse Link profile to Fabric you have to enable the Delta parquet conversion for Fabric link / Enable Parquet/Delta lake option.
• Profiles that write CSV can't be linked to Microsoft Fabric. Neither can profiles secured with managed identities.
• Where Link to Fabric selects all change-tracked tables by default, Synapse Link has system administrators pick the tables they want.
• Storage and compute costs sit with you, in your subscription, rather than showing up as Dataverse storage consumption.
• Low-latency sync — the redesigned engine that writes straight to delta parquet — applies to Fabric Link configurations, not to Synapse Link. Synapse Link continues to function on the existing pipeline.
• The Delta parquet / Delta Lake export option is itself now deprecated (announced September 2026): closed to new customers from 15 October 2026, removed December 2027. The stated replacement for delta parquet output is Link to Microsoft Fabric, not another Synapse Link setting.
• In append only mode a deleted row isn't removed from the lake straight away — it's appended with isDeleted = True. Synapse Link then hard-deletes that record from the lake after 30 days. Delta Lake output behaves the same way: soft delete on the next sync, hard delete at 30 days. To keep deleted rows longer, enable long-term retention on the table.
• Export to Data Lake (the old COMO service) has been deprecated and is being retired, with no further extension or exception process planned as of the June 2026 announcement.

When to use / skip

Synapse Link is now the specialist option, not the default. Pick it when the client genuinely needs the data in their own storage account — because a non-Fabric tool has to read it, because a data governance policy says the bytes live in their subscription, or because they have existing pipelines they aren't rewriting this year. For anything greenfield where Fabric is on the table, Link to Fabric is the shorter path and gets the engineering investment. If a client is still on Export to Data Lake, that's not a roadmap item, it's a migration with a deadline.

Configuration decisions

• Storage destination: ADLS Gen2 alone, or ADLS Gen2 plus a Synapse workspace, decided by whether anyone will actually use Synapse compute.
• Which tables replicate. Unlike Link to Fabric this is an explicit choice, so it's also an explicit governance conversation.
• Output format — and it's effectively a one-way door, because CSV profiles can't later be linked to Fabric.
• Authentication for the profile, bearing in mind managed identity profiles are currently excluded from Fabric linking.
• Whether the profile is a stepping stone to Fabric or the permanent destination, because that decides the format choice above.

Gotchas

• Deprecated 2026-09-07: the Delta Lake (Parquet) export option is going. From 15 October 2026 it isn't available to new customers, and existing customers must move to CSV output in Synapse Link or to Link to Microsoft Fabric by December 2027. See Export Dataverse data in Delta Lake format.
• Choosing CSV closes the Fabric door. It reads like a harmless format preference during setup and it isn't.
• Managed identity is the security choice most enterprises want, and it's currently the choice that blocks Fabric linking. Worth calling out to the security team before they mandate it.
• Because it's a push rather than a scheduled pull, there's no refresh interval to tune when things are slow, and no obvious dial for people used to dataflows.
• Costs are distributed. Dataverse storage doesn't move, but the client picks up storage, compute and integration charges in Azure, and those land on a different budget line than they expect.
• Deletes replicate. Downstream consumers built on the assumption that the lake is append-only will get a surprise.
• Append only isn't the archive people assume it is. The isDeleted tombstone survives 30 days and then the row goes for good, so a lake being used as the audit trail of record quietly loses deleted history a month later. Microsoft only documented the 30-day window in September 2026 — plenty of profiles were built before anyone knew. Long-term retention on the table is the fix, and it's a Dataverse setting, not a Synapse Link one.

Consultant notes

• Ask early whether Fabric is in the client's future. If the answer is yes, either go straight to Link to Fabric or at minimum set the profile up as delta parquet so the option stays open.
• Anyone still on Export to Data Lake needs a dated migration plan. Microsoft has been explicit that there's no extension process, and "we'll deal with it later" is not a plan.
• The comparison table in Microsoft's own docs is genuinely useful in a steering meeting — no copy no ETL versus your own storage, all tables versus chosen tables, Dataverse storage versus your storage. Put it on a slide.
• Existing Synapse Links can be pointed at Fabric to pick up DirectLake and Spark without rebuilding, provided the format and authentication constraints are met. That's often the cheapest route for a client who has already invested.

---

Revisit as the December 2027 Delta Lake export removal approaches, if managed identity profiles become linkable to Fabric, or if the 30-day deleted-record window changes.