Cloud ERP vs On-Premise ERP

A comprehensive guide to choosing between cloud-hosted and self-managed ERP — covering cost, security, control, and future-proofing.

Best for
Any business evaluating deployment strategy
Pricing
Varies widely by vendor and user count

Quick verdict

This is a decision guide, not a product choice: read it when deciding between cloud and self-managed deployment. Compare TCO, control, and risk before picking a vendor.

Who this is for: Anyone about to sign an ERP contract, because the deployment clause determines your cost curve, your upgrade rights and your exit options for the next five years. This is a decision guide, not a product comparison — pick the model first, then shortlist products that support it.

Commercial disclosure. ERP Guide Hub is an editorial resource. We do not present ourselves as an authorised partner, reseller or distributor of any product named on this page. We may receive commission or referral fees on products we cover, Chanjet products included. That does not buy placement, a better position in a list, a softer verdict or the removal of a limitation. Nothing here is an offer to supply any product named on this page. Any figure or feature claim still needs checking against the vendor's current documentation. Read our editorial policy for the full terms.

This page specifically. No vendor paid for this page, and the deployment models described here are not tied to any product we might supply — the products we cover are cloud-hosted in mainland China, which is itself a constraint on the on-premise option.

Cloud (vendor-hosted) vs On-premise / self-managed

The same dimensions for both models, including the ones vendors tend to leave out — who is responsible at three in the morning, and who controls when your core system changes.

On a phone, swipe the table sideways to see both columns.

DimensionCloud (vendor-hosted)On-premise / self-managed
Cost structureRecurring subscription plus implementation. No hardware, but watch consumption tiers, storage overage and renewal uplifts.Capital hardware on a three-to-five year cycle, database and OS licences, backup and DR, power and cooling — plus the people to run it.
Who is responsible when it breaksThe vendor for infrastructure and availability; you for process and data quality.You, entirely. Including at three in the morning, and including the restore you have never tested.
Customisation freedomMulti-tenant SaaS generally forbids arbitrary private code inside the core — which blocks the deep localisations Chinese rebate work often needs.Anything you can build, you can run. The cost is that you also own its maintenance forever.
Upgrade cadence and who controls itThe vendor's calendar, often quarterly. Features arrive without a project, and breaking changes arrive the same way.Your calendar. You can defer an upgrade for years — and eventually pay for it in one painful jump.
Data residency and connectivityData sits in a region the vendor chooses. Overseas sites depend on the public internet; Chinese sites depend on cross-border links.Data stays where you put it. Site connectivity is your design problem, not a third party's.
Security realityProvider security investment — dedicated teams, independent audits, continuous testing, geographic redundancy — generally exceeds what one company can build.As good as your team and your patch discipline. For most mid-sized businesses that is the weaker option, whatever the intuition says.
Exit and portabilityContract-dependent. Extraction format, frequency and cost should be negotiated before signature, not after.You already hold the data; the exit risk is skills and hardware refresh rather than access.
  • Cost structure: Over five years the two totals frequently land within about twenty per cent of each other. Cost rarely decides this.
  • Customisation freedom: This is the single most consequential row for a Chinese exporter. Confirm the extension model before choosing.

When cloud (vendor-hosted) is the right answer

  • +You have no one who can genuinely own database administration, patching and backup testing
  • +You want the vendor contractually responsible for availability
  • +Your sites have reliable connectivity, including overseas ones
  • +You need the system live quickly and want to avoid a hardware cycle

When on-premise / self-managed is the right answer

  • +A regulation or data-residency rule genuinely forbids offshore hosting
  • +Production sites have unreliable links — workshops, mines, vessels, remote warehouses
  • +Deep integration with on-site hardware is core: MES equipment, weighbridges, PLCs, legacy scanners
  • +Latency-sensitive shop-floor transactions cannot tolerate a round trip
  • +Your rebate or customs localisation needs private code that multi-tenant SaaS forbids

What actually goes in the five-year model

No figures here on purpose. Licence prices change per deal and per country, so a number invented in a comparison table is worse than no number at all. The line items, though, are the same everywhere.

Line itemCloud (vendor-hosted)On-premise / self-managed
SoftwareSubscription per user or per consumption unitPerpetual licence plus annual maintenance
InfrastructureIncluded in the subscriptionServers, storage, backup, DR site, power and cooling
PeopleProcess owner and super-usersThe above, plus a database administrator who must be retained
UpgradesVendor-run, with your regression testing each cycleProject-based, often deferred, then expensive
Implementation and migrationQuoted separately — usually the largest single line in year oneQuoted separately, plus hardware lead time
Localisation maintenanceConstrained by the extension model; verify who ships when rules changeBuildable, but you own it permanently

Decide in this order

  1. 1Write down the real constraint first: is it cost, control, connectivity, or a compliance rule? Most buyers assume cost and discover it was control.
  2. 2Ask who restores Tuesday's data, and when that restore was last tested. If nobody knows, the answer decides the model.
  3. 3Take your hardest localisation requirement — usually export rebate — and ask each shortlisted vendor to describe the extension mechanism in writing.
  4. 4Model five years including hardware refresh, database administration and DR, not just the subscription.
  5. 5Check latency from an actual overseas site and an actual factory floor during peak hours.
  6. 6Negotiate the exit clause: data format, extraction frequency and cost, at signature.
  7. 7If you choose hybrid, design the interfaces before go-live — retrofitted synchronisation is where hybrid projects fail.

Sources and when we checked

These pages are cited for the deployment models each vendor documents, not as price evidence. The twenty per cent observation above is a pattern we describe, not a measured benchmark.

What we could not verify

  • Any specific five-year TCO figure — the twenty per cent observation is a pattern we describe, not a measured benchmark
  • Which vendors currently allow private code inside a multi-tenant deployment, and under what extension model
  • Data-residency options available to a mainland-China legal entity from each major provider
  • Restore-time commitments in standard contracts — these are negotiated, not published

Not sure which model your case allows?

The decision is not cloud versus on-premise in the abstract. It is: who can restore your data fastest, and does your hardest localisation fit inside the extension model you are buying?

Going deeper

The five-year cost nobody models properly

Most comparisons collapse into a licence-versus-subscription argument, which is the smallest part of either answer. A credible model has to account for what each option removes from your plate.

On-premise carries hardware capital expense on a three-to-five year replacement cycle, operating system and database licences, storage and backup software, virtualisation, monitoring, power and cooling, plus a disaster-recovery site you will pray you never need. Above all it carries people: someone with database administration skills is on staff or on retainer, and replacing them is expensive and slow.

Cloud removes most of those lines and replaces them with a recurring subscription. What it does not remove is implementation, data migration and integration middleware, all of which are usually quoted separately. Watch too for consumption tiers — API call limits, storage overage, sandbox environments — and for renewal uplifts written into multi-year contracts.

The honest conclusion usually surprises both camps: over five years the totals are frequently within twenty per cent of each other. The decision should turn on capability, not cost arithmetic — specifically on whether your organisation can operate infrastructure well, which most cannot prioritise.

When owning the server is genuinely the right answer

There are defensible reasons to keep ERP local, and they are narrower than most people assume.

The strong cases: regulatory or data residency rules that genuinely forbid offshore hosting; production sites with unreliable connectivity such as workshops, mines, vessels or remote warehouses; deep integration with on-site hardware including MES equipment, weighbridges, PLCs and legacy scanner networks; and latency-sensitive shop-floor transactions that cannot tolerate a round trip.

The weak case is 'cloud is not secure'. For the overwhelming majority of companies, a major cloud provider's security investment — dedicated teams, independent audits, continuous penetration testing, physical controls, geographic redundancy — exceeds anything a single business can build for itself. Ask a sharper question instead: who can restore Tuesday's data faster, and when did you last test that restore?

In practice hybrid is now the norm. Core transactions may sit on-premise while analytics, collaboration and customer-facing tools run in cloud, or the reverse. Hybrid only works when interfaces are designed deliberately; retrofitted synchronisation is where these projects fail.

What the deployment choice means for exporters

Three checks apply regardless of which way you lean, and each one has broken real projects.

Connectivity to local platforms. Chinese customs portals, tax interfaces, bank systems and e-Port access have specific network and certificate requirements. An on-premise system behind a locked-down firewall needs deliberate, documented egress rules; a cloud deployment needs confirmed reachability and often a decision about which region hosts the data. Decide this during design, not during go-live week.

Customisation limits — this one surprises people. Multi-tenant SaaS generally does not permit arbitrary private code inside the core application. If your export rebate module must be deeply customised, as Chinese rebate work usually is, pure multi-tenant deployment can block it. The workable paths are an extension layer outside the core, a single-tenant or private cloud edition, or a partner-supplied localisation the vendor supports. Confirm which applies before committing. The depth of what rebate actually requires is set out in our export tax rebate guide.

Exit terms. Require the contract to state data export formats, frequency and any associated cost. Cloud contracts that are silent on extraction tend to become expensive precisely when you most need to leave.

Frequently asked questions

Is cloud ERP always cheaper than on-premise?

No. Over a five-year horizon the totals frequently land within twenty per cent once you count hardware replacement, database administration, power, disaster recovery and staffing. Decide on capability and risk, then compare costs.

Is on-premise more secure than cloud?

Rarely, in practice. Major providers invest far more in security controls, auditing and redundancy than most single organisations can. The exceptions are genuine regulatory prohibitions and environments where data must never leave a controlled site.

Can we move from cloud back to on-premise later?

Technically yes if the vendor supports export or offers a deployable edition, but expect substantial cost and downtime. Negotiate exit terms and data formats at signing rather than at departure.

What happens to operations during an internet outage?

With cloud ERP, transactional work stops — which is why the business case must include connectivity resilience. Sites with unreliable links usually need a local component or a hybrid design for floor-level transactions.

Does the deployment model affect China export rebate handling?

Yes. Rebate and customs work usually needs localised code, and multi-tenant SaaS typically restricts private code in the core. Choose single-tenant, private cloud or a supported partner extension before committing to a deployment model.

Is hybrid just a compromise?

Often it is the correct architecture, provided integration is designed from the start. Retrofitting synchronisation between systems after go-live is where hybrid projects fail, not the hybrid concept itself.

Last updated: