Back to blog
Ronald Kraijesteijn, Founder
·March 25, 2026·12 min read

On-premise MDM vs cloud MDM: what's actually driving the shift back

Home/Blog/On-premise MDM vs cloud MDM: what's actually driving the shift back
Last updated: November 2026
MDM deployment: key differences
On-Premise
Data locationYour servers
SQL Server integrationDirect
Compliance auditYour infra, your call
Exit complexityLow
Pricing as you growFlat
GDPRDORANIS2HIPAA
Cloud MDM
Data locationVendor platform
SQL Server integrationVia API / ETL
Compliance auditVendor-dependent
Exit complexityHigh
Pricing as you growScales up
Compliance varies by vendor & region
Compliance fit depends on your jurisdiction and sector — always validate with your legal team.

On-premise MDM runs on servers you control. Cloud MDM runs on a vendor’s servers. Choose on-premise when a regulator, a contract or your own security policy says where master data may sit, or when you already run SQL Server and have a DBA. Choose cloud when you have no database team and no rule about data location.

GDPR does not force anyone on-premise. It limits transfers of personal data outside the EU. DORA, which has applied to EU financial firms since 17 January 2025, requires contracts with ICT providers to state where data is processed. Read your vendor’s data location clause before you sign.

How cloud MDM became the default answer

For most of the 2010s, the direction looked obvious. No servers to manage, automatic updates, built-in redundancy. MDM vendors that pivoted to cloud SaaS pulled ahead in analyst rankings. The narrative became self-reinforcing: on-premise MDM was legacy thinking, a sign your IT team was behind the times.

The pitch wasn’t wrong for everyone. If you had no DBA team, no existing SQL Server investment, and no data residency requirements, cloud MDM solved real operational problems. The assumption built into that pitch — that most enterprise buyers fit that description — turned out to be less accurate than the vendors hoped.

Most mid-market organizations already had SQL Server. They had DBA teams. And by 2022, a growing number of them had data sovereignty problems they hadn’t anticipated when they signed their cloud MDM contracts.

What changed the math between 2022 and 2025

Three regulatory developments hit close enough together to force a real reconsideration.

GDPR enforcement got expensive. EU supervisory authorities issued over €2.5 billion in fines between 2020 and 2024, with data transfer violations accounting for a growing share. The Schrems II decision had already made clear that standard contractual clauses aren’t a blanket solution. Legal teams that had approved cloud MDM contracts on the basis of “we have SCCs” started revisiting those decisions as enforcement escalated.

DORA landed in January 2025 for EU financial services firms. Article 28 requires organizations to audit third-party ICT providers and maintain contracts that specify data location, data format ownership, and exit provisions. A lot of cloud MDM contracts — written before DORA existed — didn’t pass that audit without expensive renegotiation.

NIS2 took effect in October 2024 for critical infrastructure operators across manufacturing, energy, and healthcare. Organizations that hadn’t thought seriously about where their master data was processed had to start.

None of this made cloud MDM impossible. It made the procurement process longer, the legal review more expensive, and the ongoing audit burden heavier. For organizations in regulated sectors, the math stopped working.

The lock-in cost nobody puts in the business case

Cloud MDM vendors don’t advertise exit complexity. It’s not in the sales deck.

Your entity models, attribute mappings, approval workflows, and validation rules live in the vendor’s proprietary schema. When you decide to migrate — to a different tool, or back on-premise — you export to flat files and start over. There is no standard MDM interchange format. The relationship hierarchies you built over three years don’t transfer. The governance rules you tuned through six months of stakeholder workshops don’t transfer. You rebuild.

Pricing escalation compounds this. Most cloud MDM platforms start with a base fee, then layer on per-domain charges (governing products and suppliers separately each costs extra), per-user charges for data steward access, and sometimes API call charges when downstream systems query frequently. What looks like a reasonable five-figure annual license at contract signature tends to look different three years in, once usage has grown and expansion pricing kicks in.

Switching costs and pricing escalation together create a hold. Most teams only fully understand it once they’re inside it.

When the cloud itself breaks

Regulation is the slow reason teams bring master data back in-house. Outages are the fast one. A Barclays survey of enterprise CIOs found 83% planning to move at least some workloads off the public cloud. Below are six public cases, each linked to its source.

Google deleted a pension fund (May 2024)

A setup mistake at Google Cloud deleted the entire cloud subscription of UniSuper, an Australian pension fund with more than 600,000 members. Members could not reach their accounts from 2 to 15 May. Google’s CEO called it a “one-of-a-kind occurrence”. The fund came back because it also kept a backup with a different provider.

Nobody outside your company can delete a server in your own data center by changing a setting.

A court lost its prosecutor’s mailbox (May 2025)

After US sanctions against the International Criminal Court, its chief prosecutor lost his Microsoft email account and moved to Proton Mail, AP reported. Microsoft says it never stopped its services to the court itself. Handelsblatt later reported that the court plans to replace the Microsoft software on its workstations with openDesk, an open-source suite.

A sanctions list in another country cannot switch off a database on your own server.

One AWS region stopped apps, games and banks (October 2025)

On 19 and 20 October 2025, two automation processes in Amazon’s DNS system raced each other and deleted the address of DynamoDB in us-east-1. Services that depended on that region failed for about 14.5 hours, by AWS’s own count.

Master data on your own SQL Server does not depend on the health of a region in Virginia. Your ERP reads it over your own network.

Rackspace switched off email for good (December 2022)

Ransomware hit Rackspace’s hosted Exchange service on 2 December 2022. Rackspace counted about 30,000 affected customers. It never rebuilt the service and moved customers to Microsoft 365. A month later, a little over half of them could download some or all of their old mail.

When master data sits in your own SQL Server database, you can read it with plain SQL, whatever happens to the vendor that wrote the software.

Atlassian deleted 775 customers’ sites (April 2022)

A script meant to switch off one app deleted 883 Atlassian cloud sites belonging to 775 customers. Jira and Confluence were gone for those teams, and some waited 14 days to get them back. Atlassian says no customer lost more than five minutes of data. That is true, and little comfort to a team without its tickets for two weeks.

The script ran on Atlassian’s cloud. Teams that ran Jira on their own servers never noticed.

OVHcloud burned, and the backups with it (March 2021)

Fire destroyed OVHcloud’s SBG2 data center in Strasbourg on 10 March 2021. Some customers had paid for backups, but OVH kept those backups in the same building. A French court later ordered OVH to pay two of them about €100,000 and €150,000. OVH appealed.

Your own server room can burn too, so the lesson here is the backup. With your data on your own SQL Server, you choose where the second copy lives, and you can check that it is really there.

What on-premise does not fix

On 19 July 2024, a faulty CrowdStrike update crashed about 8.5 million Windows devices. Many of them stood in companies’ own offices and server rooms. Running on-premise helped none of those that day.

On-premise cuts the number of outsiders who can take your data away. It also hands you the work: your DBA team runs the backups, applies the patches and tests the restore. If nobody on your team will do that, a cloud vendor is the safer choice.

The SQL Server integration argument

For organizations already running SQL Server — which covers most mid-market European and North American businesses — on-premise MDM has a practical advantage that tends to get underweighted in the cloud discussion.

When your MDM system lives on the same infrastructure as your ERP and data warehouse, the master data views your ERP reads are SQL Server views. Direct. No API calls across the internet, no added latency, no authentication layer to maintain, no per-call charges. When the DBA team needs to troubleshoot, they use the same tools they already know. Primentra works this way: it installs on your own SQL Server, and the master data stays there.

Cloud MDM integration with on-premise ERP requires an extra layer: an ETL pipeline, a middleware connector, or a custom sync job. Each additional component is another failure point, another team that needs to understand it, and another thing that breaks at 2am.

When cloud MDM is the right call

On-premise MDM isn’t always the better option. It’s the better option for specific situations.

Cloud makes sense when you have no internal DBA team and no interest in building server infrastructure. When your core systems are already SaaS-based, so integration complexity is symmetric either way. When you need the vendor to manage geo-redundancy across multiple regions. Or when your industry has no data residency requirements and your legal team has reviewed the cross-border data processing.

A US-headquartered company with 150 employees, all-cloud infrastructure, and no regulated personal data probably has no argument for adding on-premise infrastructure just for MDM. Cloud is the right answer there.

Mapping your situation to the decision

The choice usually comes down to a handful of practical questions: where does your regulated data need to stay, do you already have SQL Server and a DBA team, what does your five-year cost look like as usage grows, and how painful would switching vendors be in year four?

Your situationOn-premiseCloud MDM
GDPR / DORA / NIS2 applies to your org✓ Clear path✗ Needs careful vendor vetting
Already running SQL Server + DBA team✓ Native fit~ Adds integration overhead
No IT team, SaaS-first infrastructure✗ Adds overhead✓ Vendor handles it
5-year cost predictability matters✓ Flat pricing✗ Usually escalates
Likely to switch tools eventually✓ Low exit cost✗ Proprietary schema hold
Multi-region, global operations~ Requires own geo setup✓ Vendor manages it

The honest answer

Most organizations that contact us after running cloud MDM for two or three years aren’t unhappy with the tool. They’re unhappy with the infrastructure decision — the compliance audit that got complicated, the pricing that didn’t scale the way the initial model suggested, or the integration layer held together by a contractor who left six months ago.

Those problems are solvable. They’re just easier to avoid at the start than to fix midway through a three-year contract.

If your organization runs SQL Server, has a DBA team, and operates in a regulated sector, the case for on-premise MDM is stronger now than it was five years ago. Not because cloud MDM got worse — it didn’t. But because the regulatory environment around data sovereignty got more demanding, and the full cost of cloud lock-in became clearer once enough organizations had actually experienced it.


MDM that runs where your data already lives

Primentra installs on your SQL Server, keeps your master data on infrastructure you control, and integrates directly with your ERP via staging tables — no middleware, no per-call charges, no data leaving your network. The 60-day trial includes everything.

More from the blog

Data silos: the word everyone uses, and the fix that never removes one8 min readSingle source of truth: the phrase everyone uses and almost nobody has8 min readMaster data vs reference data: two different problems people keep solving with one tool8 min read

Ready to migrate from Microsoft MDS?

Download Primentra and run it on your own server, or try the live demo first. All features included.

Download Free TrialTry DemoCompare MDM tools