Self-Hosted Matrix vs Slack and Teams: Lessons From the EU Commission's 40,000-User Rollout
The European Commission took Matrix from proof of concept to every staff laptop in nine months. Here is the stack it runs, what broke at 40,000 users, what self-hosting Matrix costs you in day-2 work, and when Slack or Teams is still the better call.
Key takeaways
- The European Commission rolled Matrix out to more than 40,000 staff in nine months, on Element Server Suite Pro run under its own control.
- Self-hosted Matrix gives you data ownership, end-to-end encryption and an open protocol, but you run Synapse, Postgres, auth and media yourself.
- Pick it when control of chat is a requirement; keep Slack or Teams when it is a preference.
On this page
Self-hosted Matrix is now a proven choice for organisations that need to own their chat. The European Commission runs it for more than 40,000 staff. Run it as a production service with its own on-call, or don't run it: the protocol is mature, but the operations are yours.
Verified against
Element's November 19, 2026 write-up of the Commission's Matrix Conference talk and its device verification post, the ESS Community 26.11.2 README and maintenance guide, Synapse v1.165 docs, Slack's data residency and EKM pages and Microsoft Learn's Teams guides. Checked November 2026.
What the Commission actually did
Per Element's account of the talk, a proof of concept landed in December 2025, production infrastructure was built from January to June 2026, and after laptop and mobile pilots the laptop rollout went out in waves from July to September. The stack is Element Server Suite Pro, Element's enterprise Matrix distribution, which the Commission operates under its own control on sovereign cloud providers picked through a tender, not in its own data centres.
Three decisions are worth copying. Login goes through EU Login, the sign-in staff already use. Federation with outside servers is still under assessment, because it needs governance first. And the tool complements Teams rather than replacing it, so there is no one-to-one swap to promise users.
Scale problems a pilot won't show
The most useful part of the talk is what only appeared at 40,000 users:
- Directory search across homeservers. The Commission runs several Synapse homeservers, and early on users on different servers could not always find each other. Element and the Commission fixed it so one search covers everyone.
- User-led account creation. Accounts appear only after a user signs up, so you cannot find a colleague who has not registered yet.
- Presence. Load from live online/away status started to affect performance, so the team switched it off.
- Wave-shaped load. Three months of rollout waves meant uneven demand, so the team tuned the infrastructure to scale itself.
Decide presence on day one. In Synapse it is one setting:
# homeserver.yaml (in ESS, set it under synapse.additional)
presence:
enabled: false
The stack you would actually run
The quickest serious route is Element Server Suite Community, an AGPLv3 Helm chart, aimed at non-commercial use, that deploys the whole stack:
- Synapse, the homeserver
- Matrix Authentication Service (MAS), which manages users and authentication
- Element Web, Element Admin and Element Call's Matrix RTC backend
- HAProxy routing,
.well-knowndelegation and an optional bundled Postgres
# Install or upgrade (the same command does both)
helm upgrade --install --namespace "ess" ess \
oci://ghcr.io/element-hq/ess-helm/matrix-stack \
-f ~/ess-config-values/hostnames.yaml -f ~/ess-config-values/tls.yaml --wait
# Registration is off by default: create the first user through MAS
kubectl exec -n ess -it deploy/ess-matrix-authentication-service -- \
mas-cli manage register-user
Pick the server name carefully: it becomes the suffix of every Matrix ID (@alice:example.com), and the README says you cannot change it without resetting the database. Calls need TCP 30001 and UDP 30002 open for the RTC backend, alongside TCP 80 and 443.
ESS Community enforces no user limit, but Element tunes it for roughly 100 users. Synapse scales by splitting work into worker processes that coordinate over Redis, and workers only run on PostgreSQL. ESS Pro adds automatic worker scaling, in-cluster high availability, S3 media storage and LDAP/SCIM provisioning. A 40,000-seat rollout runs on the Pro tier or on a team that has rebuilt those pieces itself.
Day-2 work you take on
A consistent backup covers more than the database:
- Scale Synapse and MAS to zero.
- Dump Postgres.
- Save your Helm values.
- Export the chart-generated Secrets, which include Synapse's signing key.
- Export the deployment-marker ConfigMaps.
- Copy the media volume.
The guide warns that changing the signing key without extra configuration has impact, so treat it like a root credential. Our guide to testing database restores applies directly: the documented restore ends with a TRUNCATE e2e_one_time_keys_json step, the kind of detail to rehearse, not improvise.
Encryption adds support load Slack admins never see. Since Element Web and Desktop 1.12.33, verifying a new session is mandatory by default, and a user who loses every device needs their recovery key to unlock the encrypted key vault. The server admin cannot see it, so onboarding has to teach it.
| Self-hosted Matrix (ESS) | Slack | Microsoft Teams | |
|---|---|---|---|
| Where data lives | Your servers or chosen cloud | US by default; Business+ and up can pin messages, files and search to one of 12 regions | Your tenant's Microsoft 365 geography |
| End-to-end encryption | Default for new DMs and invite-only rooms in Element | No; encrypted at rest and in transit, customer-held keys via EKM on Enterprise+ | Opt-in for one-to-one calls; meetings need Teams Premium |
| Cross-org chat | Open federation, governed by you | Slack Connect | External access |
| Who runs it | You | Slack | Microsoft |
| Exit cost | Open protocol, your database | Workspace export | Teams Export APIs |
Where it hurts
You inherit the pager. Synapse has shipped a minor release roughly every two weeks through autumn 2026, and security fixes do not wait for your change window. The maintenance guide recommends running Postgres outside the chart. Integrations are thinner: Hookshot connects GitHub, GitLab, Jira and more, but incident bots may need rewriting. And outages become yours: when Microsoft 365 went down on November 12, its customers waited on Microsoft, while a self-hosted homeserver going down is your incident.
Questions people ask
Is Matrix a replacement for Slack or Teams?
For chat, rooms and calls, yes. For the wider Teams suite, no: the Commission positions Element as a complement to Teams for secure internal communication.
How many users can a self-hosted Synapse handle?
There is no fixed ceiling. Larger deployments split Synapse into workers on PostgreSQL and Redis, and the Commission serves more than 40,000 users across several homeservers. Element tunes ESS Community for about 100 users and positions ESS Pro above that.
Do I need federation?
No. The Commission runs without external federation while it works out governance, and opening up later means policy work, not just a config change.
The decision, concretely
- Data ownership or sovereignty is a legal or contractual requirement? Self-hosted Matrix, and budget for ESS Pro or an equivalent in-house platform team.
- Non-commercial community under about 100 users? ESS Community on a single K3s node is the fastest route.
- You need chat to be someone else's problem? Stay on Slack or Teams, and use data residency settings where they exist.
- Heavy SaaS integrations and bots? Map each one to Hookshot or a bridge before you commit.
The call we'd make
If control of your communications is a requirement, run Matrix on ESS: disable presence before the first wave, put Postgres outside the chart, and rehearse a full restore including the signing key. If it is only a preference, the on-call cost outweighs it and Slack or Teams is cheaper. The caveat: document device verification and recovery before launch, as the Commission did, and treat federation as a governance project, not a config change.
Sources
- Element: How the European Commission rolled out Matrix to more than 40,000 staff
- Element: Device verification becomes mandatory by default
- Element: ESS Community Helm chart (ess-helm)
- Synapse: Scaling Synapse via workers
- Matrix.org: End-to-end encryption by default
- Slack: Data residency for Slack
- Slack: Enterprise Key Management
- Microsoft: End-to-end encryption for Teams meetings and calls
- Microsoft: Microsoft 365 data locations
- Microsoft: Teams external access
Cite this article
Kiril Urbonas (2026). Self-Hosted Matrix vs Slack and Teams: Lessons From the EU Commission's 40,000-User Rollout. DevOpsNess. https://www.devopsness.com/blog/self-hosted-matrix-vs-slack-teams
[Self-Hosted Matrix vs Slack and Teams: Lessons From the EU Commission's 40,000-User Rollout](https://www.devopsness.com/blog/self-hosted-matrix-vs-slack-teams) (Kiril Urbonas, DevOpsNess, 2026-11-19)
More from DevOps
Explore more articles in this category
Container Vulnerability Scanning Tools: Trivy vs Grype vs Snyk vs Docker Scout
A practitioner's comparison of Trivy, Grype, and Snyk for finding CVEs in container images, plus how to wire scanning into CI without drowning in noise.
Platform Engineering Golden Paths: Build Your First One in 6 Steps
A golden path is a template plus the defaults that make the right way the easy way. How to build the first one, service template, CI, deploy and observability, how to measure whether anyone uses it, and the two ways platform teams kill their own paths.
18 Kubernetes Interview Questions With Answers, by Level
Kubernetes interviews in 2026 are scenario-led. Eighteen questions grouped junior, mid and senior, each with the answer, the first command, and what separates a strong answer from a memorized one.
You might have missed
Evergreen posts worth revisiting.