Azure Synapse Link Delta Lake Is Being Retired: Your Dataverse and D365 F&O Export Options

Azure Synapse Link for Dataverse supports two output formats. One of them, Delta Lake (Parquet), is deprecated and has a defined removal date. This post covers the scope of the deprecation and the two supported migration targets.

#Deprecation timeline

Microsoft has deprecated the Delta Lake (Parquet) export option in Azure Synapse Link for Dataverse.

Date	What happens
15 October 2026	The Delta Lake (Parquet) export option is no longer available to new customers. Existing profiles keep running.
December 2027	The Delta Lake (Parquet) export option is removed. Existing customers must already be on Synapse Link CSV output or Link to Microsoft Fabric.

Existing profiles continue to function until the December 2027 removal date. The lead time matters because changing output format is not a configuration change on an existing profile — in most cases it requires creating a new profile and a full initial sync.

#What is actually being deprecated

Azure Synapse Link for Dataverse is not being retired. The service continues. What's being retired is one output format option within it: the one where you attach your own Synapse Spark pool and Synapse Link converts staged CSV into Delta Parquet on a schedule.

Here's what survives and what doesn't:

Component	Status
Azure Synapse Link for Dataverse	Continues. Not deprecated.
Synapse Link → CSV output to your own ADLS Gen2	Continues. A valid migration target.
Synapse Link → Delta Lake / Parquet output	Deprecated. Closed to new customers Oct 2026, removed Dec 2027.
Link to Microsoft Fabric	Actively invested in. Microsoft's recommended path for Delta output.
F&O Export to Azure Data Lake add-in	Permanently stops 30 Nov 2026. Already retired since Nov 2024.

Microsoft's own guidance points Delta/Parquet users at Link to Microsoft Fabric, citing significantly lower synchronisation latency as the reason.

flowchart LR
    DV[Dataverse / D365 F&O]

    DV --> ETL[Export to Data Lake add-in]
    DV --> SLD[Synapse Link - Delta Parquet]
    DV --> SLC[Synapse Link - CSV]
    DV --> LTF[Link to Microsoft Fabric]

    ETL --> X1[Stops 30 Nov 2026]
    SLD --> X2[Removed Dec 2027]
    SLC --> OK1[Supported - your own ADLS Gen2]
    LTF --> OK2[Supported - OneLake shortcuts]

    style X1 fill:#fde2e2,stroke:#b91c1c
    style X2 fill:#fde2e2,stroke:#b91c1c
    style OK1 fill:#dcfce7,stroke:#15803d
    style OK2 fill:#dcfce7,stroke:#15803d

#Why the Delta option is being removed

The architecture explains which replacement is the closer equivalent.

The Delta option was never a direct write. Synapse Link exported your tables to CSV first, then ran a Spark job on a pool you provisioned in your own subscription to convert that CSV into Delta, then cleaned up the CSV. A nightly maintenance job compacted and vacuumed the Delta files between 11pm and 6am local time.

That architecture has three structural characteristics:

1. The customer operates a Spark pool purely for format conversion. You size it, patch its runtime, and pay for it. An undersized pool causes conversion jobs to fail.
2. The runtime requires periodic manual upgrades. Spark 3.3 reached end of support, 3.4 was subsequently retired, and the current requirement is Spark 3.5 with Delta Lake 3.0. Each transition is an in-place upgrade with a 48-hour cache propagation window.
3. Latency is bounded by the conversion schedule rather than by the rate of source change.

Link to Fabric addresses all three by writing Delta Parquet directly and holding the replica in Dataverse storage.

#The two migration targets compared

	Synapse Link → CSV	Link to Microsoft Fabric
Where data lives	Your ADLS Gen2 storage account	Dataverse storage, surfaced to OneLake via shortcuts
Who owns access control	You	Dataverse, then Fabric/OneLake security
Storage cost basis	Azure blob storage rates	Dataverse database capacity rates (substantially higher per GB)
Compute you provision	None for export; downstream ETL is yours to build	None — Fabric capacity covers downstream work
Output format	CSV (conversion is yours to implement)	Delta Parquet, native
Latency	Batch, per configured interval	Up to 60 min on the legacy engine; lower with low-latency sync
Table selection	You choose	All change-tracked tables by default; prune afterwards
Prerequisite	ADLS Gen2 + Synapse workspace	Fabric capacity in the same region as the Dataverse environment

In summary: Link to Fabric reduces operational surface area and latency, with the replica charged against Dataverse database capacity. Synapse Link CSV keeps per-GB storage cost low and the data in storage you control, with the conversion and orchestration remaining your responsibility.

#How to choose

Choose Link to Microsoft Fabric when:

• Fabric is already your analytics platform, and you have capacity in the right region.
• You want Delta Parquet without operating a Spark pool to produce it.
• Latency matters and you can take advantage of low-latency sync.
• Your table set is selective enough to keep Dataverse storage consumption within your entitlement.

Choose Synapse Link with CSV output when:

• Your consumers aren't Fabric — Databricks, Snowflake, a third-party warehouse, or your own Spark.
• You need the data in storage you own, for governance, residency, or exit reasons.

The storage cost implications are covered separately in What Link to Microsoft Fabric Actually Costs, which includes measured replica ratios.

#Summary & takeaways

• The Delta Lake (Parquet) option in Synapse Link is deprecated: closed to new customers 15 Oct 2026, removed December 2027.
• Synapse Link itself is not deprecated. CSV output remains a fully supported destination.
• Export to Azure Data Lake permanently stops 30 November 2026 with no extensions. This is a separate deprecation with an earlier date.
• The two valid targets are Synapse Link CSV (your storage, lower per-GB cost, more implementation effort) and Link to Fabric (Dataverse database capacity, higher per-GB cost, less implementation effort).

#Sources

• Export Microsoft Dataverse data in Delta Lake format — the deprecation notice (Oct 2026 / Dec 2027), the CSV-or-Fabric transition requirement, the CSV-to-Spark-to-Delta conversion architecture, nightly compaction window, and the Spark 3.5 / Delta Lake 3.0 version requirements.
• Configure your environment and link to Microsoft Fabric — prerequisites and region requirement, low-latency sync and its F&O minimum builds, the INT64 timestamp change, unlink/relink behaviour, Change Data Feed, the 2,000-table limit, and the lakehouse naming warning.

#Related reading

• What Link to Microsoft Fabric Actually Costs — the measured storage ratios and the Dataverse-versus-OneLake price gap that should drive your destination choice.
• Export to Datalake vs Synapse Link: Enhancing Data Access in D365 Finance and Operations — the original comparison, now updated with the confirmed retirement dates.
• Synapse Link for Dataverse: Upgrading Your Synapse Spark Pool Version — the in-place Spark pool upgrade walkthrough, retargeted to Spark 3.5.
• How Microsoft Fabric Mirroring Works for Azure SQL Database — the same no-ETL replication idea applied to a different source, useful for comparing ingestion patterns.