IT services / Guide

On-premises or SaaS: where should you host your IT services?

Installing a service on your own premises, hosting it on a rented server or subscribing to online software: the concrete differences in responsibilities, security, skills, reversibility and sovereignty, what changes in 2026, a service-by-service decision method and the questions to ask before signing.

IT services Guide All levels Published October 20, 2026 Translated November 20, 2026 18 min read Version française
A server rack in a data room
Photo: Kevin Ache on Unsplash

In plain words

To get around, you can buy a car, lease one long-term, or take a taxi. With your own car, you choose everything but you handle the servicing, insurance and repairs. With a lease, the leasing company takes care of the major servicing. With a taxi, you handle nothing, but you choose neither the exact route nor the vehicle, and you depend on the driver. IT works the same way: on-premises (at your place), hosted (at a provider's) or SaaS (a ready-to-use service). No formula is better in absolute terms: it all depends on the service and the organisation.

No prices are given here. Information checked on 20 October 2026.

1. The four ways to host a service

Model Where the software runs Who administers it Example
On-premises (on site) On your servers, in your premises You Active Directory, file server, GLPI on a Proxmox VM at the office
Self-hosted at a hosting provider (IaaS, VPS, dedicated server) On a machine rented in a data centre You (the system and the application) Nextcloud or BookStack on a VPS in Europe
Hosted and managed (private cloud, managed service) At a provider's The provider, according to the contract GLPI hosted by an integrator, managed database
SaaS (Software as a Service) At the vendor's, shared between its customers The vendor Microsoft 365, Google Workspace, Confluence Cloud, Intune

Between the two extremes, most organisations are hybrid: local directory synchronised with the cloud, email as SaaS, line-of-business application on site, backups at a hosting provider. For the main families of cloud (IaaS, PaaS, SaaS) and the providers, see the cloud comparison.

2. Who is responsible for what?

This is the most frequently misunderstood point: moving to SaaS doesn't transfer all the responsibility.

Item On-premises IaaS / VPS SaaS
Building, electricity, cooling You Hosting provider Vendor
Hardware (servers, disks, network) You Hosting provider Vendor
Operating system and patches You You Vendor
Application, updates You You Vendor
Service availability You Shared Vendor (according to the contract)
Accounts, permissions, two-factor authentication You You You
Security configuration of the service You You You
Data: content, organisation, sharing You You You
Backing up your data You You Often you (see the box)
GDPR compliance You You, with the hosting provider as processor You, with the vendor as processor

A SaaS is not a backup

The big online services protect their platform against failures, but not your data against your own mistakes: a file deleted six months ago, the mailbox of a departed employee erased, a hacked account that encrypts OneDrive. Recycle bins and retention have limited durations. Plan an independent backup of Microsoft 365, Google Workspace or your other critical SaaS, following the 3-2-1 rule.

3. Point-by-point comparison

Criterion On-premises / self-hosted SaaS
Getting started Slow: hardware, installation, configuration Fast: account created, service available
Maintenance Your job: patches, updates, hardware, monitoring Handled by the vendor
Skills needed System, network, security, backup, on-call Functional administration, account security
Control and customisation Total: version, configuration, extensions, integrations Limited to what the vendor offers
Pace of change You decide when to update The vendor imposes new features and the removal of others
Availability Depends on your infrastructure (power, redundancy) Generally high, but a vendor outage hits you with no immediate recourse
Dependence on the internet The internal service works even if the connection drops No access without the internet
Remote access Has to be organised (VPN, reverse proxy, controlled exposure) Built in, from anywhere
Data location You know exactly where it is According to the contract and the vendor's options
Applicable law Law of the country where your servers are The vendor's country's law on top (for example the CLOUD Act for an American vendor)
Reversibility Your data is at your place; the format remains an issue Depends on the exports offered and the contract
Cost structure Upfront investment, then maintenance and staff time Subscription per user or per use, no upfront investment
Main hidden cost The team's time and on-call The pile-up of subscriptions and the dependence

4. What changes in 2026

  • Vendors are pushing towards SaaS or subscriptions. Atlassian is ending its self-installed Data Center offers (end of life on 28 March 2029). Microsoft has replaced Exchange Server 2016 and 2019, unsupported since 14 October 2025, with Exchange Server Subscription Edition, which can still be installed on site but with licences maintained by subscription or Software Assurance. An "on-site" solution is therefore no longer necessarily a one-off purchase.
  • Sovereignty is becoming a common criterion. Gartner talks of "geopatriation": organisations are bringing data and applications back from foreign hyperscalers to sovereign or local clouds, or to their own servers. The firm expects most European organisations to have done so by 2030, more towards European clouds than by a complete return on site.
  • The European Data Act makes it easier to switch cloud provider. Applicable since 12 September 2025, it requires providers to make it easier to migrate to a competitor or to your own servers, and switching charges must disappear completely from 12 January 2027.
  • Qualified offers for sensitive data are growing, such as SecNumCloud in France. See the cloud comparison.

5. Service by service

There is no global answer: you decide service by service. Here are the trends for a small or medium business, to adapt.

Service Reasonable trend Why
Email and calendar SaaS Exchange on site takes a lot of expertise and remains a favourite target of attacks; availability and filtering are better handled by a specialised service
Collaborative office suite, file storage SaaS, or self-hosted Nextcloud if sovereignty comes first Co-editing and mobile access; self-hosting is realistic with a competent team
Directory and identities Hybrid Local Active Directory for workstations and internal applications, synchronised with Entra ID for the cloud. See AD, Entra and Microsoft 365 hybrid identity.
Device management (MDM) SaaS Devices are everywhere, and automatic enrolment goes through the manufacturers' services anyway. See MDM.
Inventory and helpdesk Either GLPI installs easily on site (see installing GLPI) and also exists as a hosted version
Wiki and documentation Either Simple self-hosting (BookStack), but keep an offline copy of the emergency procedures. See knowledge bases.
Password manager End-to-end encrypted SaaS, or self-hosted with great rigour An outage or loss is critical; self-hosting demands backups and high availability. See password managers.
Line-of-business application, ERP Depends on the vendor and the integrations Links with machines, the workshop or other local software often favour on site
Technical file server, large volumes On site Local throughput, large volumes, low latency
Telephony SaaS (hosted telephony) On-site phone systems are becoming rare
Backup Hybrid A local copy to restore quickly, an off-site copy protected against deletion
Monitoring On site, with an external probe Monitoring must work when everything else falls over, and someone must see from outside if the whole site is cut off

6. A decision method

For each service, give each criterion a score from 1 to 5 for both options, multiply by the weight, and compare the totals. Adjust the weights to your context.

Criterion Weight On-premises (score) SaaS (score)
Data sensitivity and regulatory constraints 3
Skills and time available in the team 3
Need for remote access and mobility 2
Expected availability and fault tolerance 2
Integration with other systems 2
Customisation needed 1
Reversibility and risk of lock-in 2
Total cost over 5 years (hardware, licences, staff time) 2
Weighted total

Count the staff time

An on-site server isn't "free" once bought: monthly patches, backups to check, hardware renewal every 5 to 7 years, monitoring, on-call, documentation. If nobody has that time, the on-site service quietly degrades until the day of the incident.

7. Before signing up for a SaaS

The questions to ask, and to find in the contract:

  1. Where is the data, including backups and support? Which subcontractors have access to it?
  2. Which law applies to the vendor? Is it subject to non-European legislation that can force it to hand over data?
  3. GDPR data processing agreement (Article 28): is it provided, and does it cover security, breach notification and deletion at the end of the contract?
  4. Reversibility: which exports, in which formats, free of charge? For how long after termination can the data still be recovered?
  5. Service level agreement (SLA): guaranteed availability, response times, penalties, public status page?
  6. Security: single sign-on (SAML, OIDC), enforceable two-factor authentication, exportable activity logs, certifications (ISO 27001, or sector-specific qualifications as relevant)?
  7. Backup: what does the vendor back up, for how long, and can a single item be restored?
  8. Change: how are changes to features, terms and prices announced?

8. Before keeping or installing a service on site

  1. Is someone responsible for the service, with a deputy?
  2. Are patches applied every month, and major updates planned?
  3. Do backups follow the 3-2-1 rule, with an offline or immutable copy and restore tests?
  4. Does monitoring alert on an outage, a full disk or an expired certificate?
  5. Is the hardware under warranty, and its renewal planned?
  6. Are power and network protected (UPS, redundancy if needed)?
  7. Is the recovery plan written down: what do you do if the server room is unusable?
  8. Is exposure to the internet limited to the bare minimum, behind a VPN or a reverse proxy?

9. Checklist

Item Done?
Each service has a chosen and justified hosting model ☐
The responsibility table is known to the team for each SaaS ☐
Critical SaaS have an independent backup ☐
SaaS contracts include a GDPR data processing agreement and a reversibility clause ☐
A full export has been tested at least once for each critical SaaS ☐
On-site services have an owner, patches, monitoring and tested backups ☐
End-of-support dates for on-site software are tracked (vendors' lifecycle calendars) ☐
The decision is reviewed every year or at each contract end ☐

Read next