The MES deployment model you choose affects data ownership, OT network security, offline resilience, latency, and total cost of ownership for years to come. This guide gives a clear framework for making the right decision based on your factory's specific requirements.
Manufacturing operational data — OEE history, downtime events, fault patterns, production volumes, quality rates — is valuable proprietary data. Where this data physically lives has legal and competitive implications.
With an on-premise MES, all production data is stored in your PostgreSQL or SQL Server database, on hardware you control. You can audit who accessed it, move it, export it, or delete it at any time. No third party has access.
With a cloud MES, production data is stored in the vendor's cloud. Review the vendor's data processing agreement (DPA) carefully: Who owns the data? What right does the vendor have to use it (including for training ML models)? What happens if you switch vendors — can you export all historical data? What is the vendor's data breach notification obligation?
Manufacturing OT networks are increasingly targeted by ransomware and industrial espionage. Most OT security frameworks (IEC 62443, NIST SP 800-82) recommend strict segmentation between the OT network (PLCs, machines) and IT/corporate networks — with no direct internet access from the OT network.
What happens to your MES when internet connectivity drops?
For continuous process industries (pharmaceutical, chemical, food) or high-volume automotive lines where a production stoppage costs thousands of euros per minute, on-premise resilience is often non-negotiable.
| Metric | On-Premise | Cloud |
|---|---|---|
| OEE update latency | <1 second (local DB) | 2–30 seconds (cloud round-trip) |
| Andon board alert display | <1 second | 5–30 seconds |
| OPC UA subscription interval | 500ms achievable | Limited by gateway buffering |
| Shift handover / UI response | Sub-second (local LAN) | Dependent on internet speed |
For most MES functions, cloud latency (2–5 seconds) is acceptable. The exception is real-time Andon boards and alarm response systems, where a 30-second delay renders the display useless for immediate fault response.
A common misconception is that cloud MES is always cheaper. A 5-year TCO comparison tells a different story:
Cloud has lower Year 1 cost and lower upfront risk. On-premise wins on 5-year TCO for any plant with more than 3–5 lines and existing server capability.
Several industries have compliance requirements that effectively mandate on-premise deployment:
Choose on-premise if: OT network is isolated or restricted, data sovereignty is required, regulated industry, need real-time (<1s) Andon displays, 5+ lines and multi-year horizon, existing server infrastructure in facility.
Choose cloud if: Small operation (1–3 lines), no IT/infrastructure resource internally, need very fast initial deployment, internet-connected OT network is acceptable, short-term/pilot use case.
No — cloud MES platforms require internet connectivity. For isolated OT networks (common in defence, automotive Tier 1), on-premise is the only viable option. On-premise MES runs entirely on your local network and is fully operational during internet outages.
Not over 5 years. Cloud has lower Year 1 cost; on-premise has lower ongoing cost once deployed. A 5-year TCO calculation typically shows on-premise is 20–40% lower for facilities with 5+ lines and existing server infrastructure.
Shopfloor Copilot is fully on-premise → Docker deployment, all data on your servers, no internet dependency, full MES on a single VPS or on-premise server.
Explore Platform →