SQL Server to Microsoft Fabric Migration
Which workloads belong in Fabric, which belong in Azure SQL, which should retire, and how to move the ones that matter. A fixed-price assessment answers the decision before anyone writes a pipeline. MVP and MCT-led.
Book a discovery callYou can migrate SQL Server workloads to Microsoft Fabric using mirroring, Data Factory pipelines and the migration tooling in Fabric Data Warehouse. But not everything should move: Fabric is an analytical platform, so reporting and warehouse workloads belong there while transactional systems stay on SQL Server or move to Azure SQL.
The question nobody on this search answers
Search for this topic and you find tool mechanics: Microsoft documentation on the migration experience, community walkthroughs of copy jobs, checklists for moving SSIS packages. All of it assumes the decision has already been made. In practice the decision is the hard part, and getting it wrong is expensive in both directions. Move a transactional workload into an analytical platform and you have rebuilt the wrong thing. Leave your reporting estate on a 2016 instance and you are paying extended support for a warehouse Microsoft stopped improving years ago.
The pressure is real and dated. SQL Server 2014 left extended support in July 2024, and SQL Server 2016 followed it on 14 July 2026, this month. If any part of your estate runs 2016 or earlier, it is receiving no security patches as of now unless you are paying for Extended Security Updates. 2017 follows in October 2027. Every year an estate stays put, the ESU bill rises and the migration gets no smaller.
| Version | Extended support ends | Position today |
|---|---|---|
| SQL Server 2014 | 9 July 2024 | Out of support two years; ESU costs escalating annually |
| SQL Server 2016 | 14 July 2026 | Out of support as of this month |
| SQL Server 2017 | 12 October 2027 | Inside the planning horizon now |
| SQL Server 2019 | 8 January 2030 | Comfortable, but the analytical estate need not wait |
Workload triage: what goes where
| Workload type | Likely destination | Why |
|---|---|---|
| Reporting databases, data marts, warehouse | Fabric (Warehouse or Lakehouse) | Analytical engine, OneLake storage, Power BI Direct Lake, no server to patch |
| ETL and overnight batch (SSIS) | Fabric Data Factory pipelines or notebooks | Modern orchestration; SSIS packages are rebuilt, not lifted |
| Transactional line-of-business databases | Stay on SQL Server or move to Azure SQL | Fabric is not an OLTP platform; see the Azure migration page |
| Databases nobody has queried in a year | Retire | Every estate carries workloads whose cheapest migration is deletion |
Mirroring deserves a specific mention because it is the honest answer for many estates: it replicates SQL Server data into OneLake continuously without moving the source system, which gives you Fabric analytics over live operational data while the transactional workload stays exactly where it is. Coexistence is a destination, not a failure to decide.
The fixed-price migration assessment
Three days across your estate for £3,750. Every database triaged to Fabric, Azure SQL, stay or retire, with a costed, phased plan, capacity sizing and the end-of-support exposure quantified. The plan is portable: you keep it whoever delivers it.
Book the assessment scoping callWhat the assessment actually delivers
Three days produces a working document, not a slide deck. The estate inventory lists every instance, database and scheduled job with its owner, its consumers and its last meaningful use, because the inventory conversation always surfaces databases nobody will claim. The workload triage assigns each item a destination with the reasoning stated, so when someone challenges a decision in six months the argument is on paper. The dependency map covers the connections that break migrations: linked servers, cross-database queries, the Access front end in finance, the SSRS subscriptions that constitute someone's entire Monday morning.
The plan itself is phased into waves with effort estimates, a capacity sizing model for the Fabric side with the workings shown, and the end-of-support exposure per instance with the ESU cost of doing nothing. The final section is the one most providers omit: what it costs to deliver each wave at the published day rate, so the plan can go into a budget paper the week you receive it. The document is deliberately portable. If you take it to another partner or deliver it in-house, it works exactly the same.
Mirroring: the option most estates should consider first
Mirroring replicates a SQL Server database into OneLake continuously, in near real time, without touching the source workload. The transactional system keeps running exactly where it is; Fabric gets a live, query-ready copy for analytics, reporting and Direct Lake models. For estates where the pain is analytical (slow reports hammering the production box, overnight ETL windows stretching into the morning) mirroring solves the presenting problem in days rather than months, and takes the reporting load off an ageing instance while the longer migration is planned properly.
It is not a full migration and should not be sold as one: the source instance still needs patching, backing up and eventually a destination of its own. But as a first wave it has two virtues the big-bang approach lacks. It delivers value before any system is decommissioned, and it lets your team learn Fabric against live data with no production risk. A meaningful share of assessments conclude that mirroring plus a phased retirement is the honest plan, and that conclusion costs £3,750 rather than a six-month programme discovering it.
How a migration actually runs
Assess
Inventory the estate, capture real usage, and triage every workload against the table above. The output is the decision document: what moves, what stays, what retires, in what order, and what it costs. Where the analytical and transactional estates split across destinations, one assessment covers both rather than selling you two.
Prove
One representative workload moves first, end to end: schema, data, pipelines, reports, security. The proof migration surfaces the estate-specific problems (data type mismatches, collation issues, the SSIS package nobody documented) while they are cheap to fix, and it gives your team the pattern for everything that follows.
Migrate
Waves ordered by dependency and business risk, delivered at £1,250 per project day against the fixed plan. Reporting cutover happens per wave, not in one weekend. Your team delivers alongside, because the objective is that wave three needs less of us than wave one.
Operate
Governance baseline, capacity monitoring, and a handover your administrators can actually run. Governance training and fractional DBA support exist where they help, as options rather than dependencies.
Two skill sets, one person
Fabric migration sits between two specialisms that rarely meet. Fabric specialists understand the destination but have never administered the estate they are emptying: backups, agent jobs, linked servers, the batch window that cannot slip. SQL Server veterans know the estate intimately but are learning the destination on your time. This practice is one Microsoft Data Platform MVP and Microsoft Certified Trainer with twenty years of SQL Server administration and current, delivered Fabric work: capacity optimisation for an insurer on F512, a Fabric deployment for a large local authority, and a data engineering build for a fintech. Both ends of the migration, one accountable person.
Frequently asked questions
Can you migrate SQL Server directly to Microsoft Fabric?
Yes, for analytical workloads. Fabric provides migration tooling into its Warehouse, pipelines for data movement, and mirroring for continuous replication from SQL Server into OneLake. Transactional workloads are the exception: Fabric is not an OLTP platform and those databases belong on SQL Server or Azure SQL.
Is Microsoft Fabric a replacement for SQL Server?
For the analytical half of the estate, increasingly yes. For transactional systems, no. Most organisations end with a two-destination estate: operational databases on SQL Server or Azure SQL, analytics in Fabric, connected by mirroring or pipelines. Planning both together is cheaper than planning them twice.
What does the migration assessment cost and include?
£3,750 fixed, exclusive of VAT, three days. It includes a full estate inventory, workload triage across Fabric, Azure SQL, stay and retire, capacity sizing with the workings shown, end-of-support and ESU exposure, and a phased migration plan with costs. The plan is yours to deliver with any partner.
What happens to our SSIS packages?
They are rebuilt, not lifted. Fabric Data Factory and notebooks replace SSIS orchestration, and the assessment maps every package to its replacement pattern before anything is committed. Simple copy-and-load packages convert quickly; the handful with embedded business logic are identified early because they set the timeline.
How long does a SQL Server to Fabric migration take?
The assessment takes three days. Delivery depends on the estate: a single warehouse and its reporting can move inside a month, while a multi-source estate with heavy SSIS typically runs in waves over three to six months. The proof migration in the first wave makes the remaining estimate reliable.
Do we need to finish the migration before end of support hits our version?
No, but you need a decision before then. The assessment quantifies which workloads carry end-of-support risk and sequences them first, and mirroring can take reporting pressure off an old instance while the longer migration runs. What you should not do is buy another year of Extended Security Updates by default.
Can our own team deliver the migration from your plan?
Yes, and the plan is written for that. Some clients deliver everything in-house after the assessment, some take delivery days for the difficult waves, some hand us the lot. The assessment price and the day rate are published so you can cost each route before choosing.
Start with the decision, not the tooling
A 30-minute call establishes what your estate looks like, whether the assessment fits, and what the honest answer to the Fabric question probably is.
Book a discovery call