For most EU startups we recommend Grafana Cloud on an EU stack: it keeps data in the EU, is built on open-source backends that speak OpenTelemetry, and its usage-based bill is easier to keep small. Datadog is the better product for teams that want the most polished all-in-one experience and can afford per-host pricing. Self-hosting Prometheus, Loki, Tempo and Grafana only pays off when you already run Kubernetes well and have someone who owns it.
Everything below about regions, pricing units and features was checked on 1 November 2026 against the vendors' official pages, linked inline. Prices are list prices before tax and before any negotiated discount.
The comparison at a glance
The table covers what actually decides the choice for an EU team; it is not a full feature matrix.
| Datadog | Grafana Cloud | Self-hosted (Prometheus, Loki, Tempo, Grafana) | |
|---|---|---|---|
| EU data residency | EU1 site (app.datadoghq.eu), located in Germany. Separate UK1 site. | EU stacks in Germany, Ireland and Sweden (AWS), Belgium (GCP) and the Netherlands (Azure), plus Switzerland and the UK. | Wherever you run it: your own EU region or your own servers. |
| Main pricing units | Per host for infrastructure and APM; per GB ingested plus per million indexed events for logs; per session for RUM. | Per active metric series; per GB for logs, traces and profiles; per session for frontend; per host-hour for application observability. | Your compute, storage and engineering time. |
| Free tier | Limited free tier for infrastructure; most paid products have trials. | Free forever: 10k series, 50 GB each of logs, traces and profiles, 14-day retention. | Software is free; operations are not. |
| OpenTelemetry | Supported through the Agent, the DDOT Collector or direct OTLP ingest; Datadog's own agent and libraries are the most complete path. | OTLP-native, built on open-source backends. | Native. |
| Lock-in | High once dashboards, monitors and proprietary query syntax pile up. | Low to medium: PromQL, LogQL and TraceQL also work self-hosted. | Lowest, but you own the upgrades. |
| Best fit | Teams that value one polished tool and have budget. | Most startups and scale-ups from 2 to 50 engineers. | Platform teams already running Kubernetes with an owner for the stack. |
EU data residency: both vendors can keep data in the EU, if you pick it at sign-up
Datadog and Grafana Cloud both offer EU hosting, but in both cases the choice is made when the account or stack is created, and it is easy to get wrong.
Datadog runs independent sites. EU1 lives at app.datadoghq.eu and is located in Germany; the docs say each site is completely independent and data cannot be shared across sites. An organisation created on the default US1 site stays there, so moving to the EU means a new organisation on EU1 and reconfiguring every agent, API key and integration. Check which site you are on before you sign anything.
Grafana Cloud lets you choose a region per stack. The EU options are AWS Frankfurt, Ireland and Stockholm (sold directly), GCP Belgium and Azure Netherlands (sold through the cloud marketplaces), and the Grafana docs are explicit that it does not support changing the region of an existing stack. Same lesson: pick an EU region on day one.
Region is only half the GDPR picture. Both vendors are US-owned companies, so you still need a signed DPA, an up-to-date list of sub-processors, and a check of which features (support access, AI assistants, some integrations) process data outside the region. We cover the full checklist, including what to strip from logs before they leave your network, in our guide to GDPR-compliant logging in EU regions.
Pricing models: per host vs per unit of data
Datadog's bill grows mainly with the number of hosts and the volume of logs you index; Grafana Cloud's grows with active metric series and gigabytes of logs and traces. Which one is cheaper depends on your shape, not on the headline rate.
Datadog
Datadog prices infrastructure monitoring and APM per host per month, with lower rates for annual commitment than for on-demand. Logs are billed twice: once per GB ingested and again per million events indexed, with the indexing price rising with retention. RUM is billed per session, and custom metrics beyond a per-host allowance are billed separately. Check the current figures on Datadog's pricing page for your plan; we don't reproduce them here because they vary by commitment and contract.
The pattern we see most: the APM and infrastructure line is predictable, the log line is not. One noisy service shipping debug logs to an indexed pipeline can outgrow the rest of the bill. Datadog gives you tools for this (exclusion filters, Flex Logs, archives to your own bucket), but someone has to configure them.
Grafana Cloud
Grafana Cloud's Pro plan has a $19 monthly platform fee, then usage. As listed on 1 November 2026:
- Metrics: from $6.50 per 1,000 active series.
- Logs: split into three charges: $0.05 per GB processed, $0.40 per GB written, and $0.10 per GB for each extra 30 days of retention beyond the included 30.
- Traces and profiles: the same processing, writing and retention structure.
- Frontend Observability: from $0.75 per 1,000 sessions.
- Application Observability: from $0.025 per host-hour.
- Users: from $8 per active user.
The free tier includes 10k active series, 50 GB each of logs, traces and profiles, 14-day retention and 3 users, which is enough for a small product in production. The split logs pricing matters: data you drop with Adaptive Logs still pays the processing fee but not the much larger write fee, so filtering inside Grafana Cloud is cheap, and filtering before you send is cheaper still.
The metric to watch on Grafana Cloud is cardinality. A label like user_id or a raw URL path on a Prometheus metric can multiply your active series overnight.
Self-hosted
The software costs nothing. You pay for compute, block and object storage, and above all engineer time. For a small team the time line is almost always the biggest one.
Features: Datadog is broader, Grafana Cloud covers what most teams use
Both platforms cover the core signals: metrics, logs, traces, frontend monitoring and alerting. Datadog's advantage is depth and integration between products; Grafana's is that each piece is an open backend you can query in a standard language.
| Signal | Datadog | Grafana Cloud | Self-hosted |
|---|---|---|---|
| Metrics | Infrastructure + custom metrics | Mimir (PromQL) | Prometheus, plus Mimir or Thanos for long-term storage |
| Logs | Log Management, Flex Logs | Loki (LogQL) | Loki on object storage |
| Traces / APM | APM with service maps and profiling | Tempo (TraceQL), Application Observability | Tempo on object storage |
| Profiles | Continuous Profiler | Pyroscope | Pyroscope |
| RUM / frontend | RUM, Session Replay | Frontend Observability (Faro) | Faro SDK into your own stack, more assembly needed |
| Synthetics | Synthetic Monitoring | Synthetic Monitoring, k6 | Blackbox exporter, k6 |
Where Datadog clearly wins is correlation out of the box: a trace links to the host, the logs and the deployment with no setup. On Grafana you get the same with consistent labels and a bit of configuration. On self-hosted you build it yourself.
Lock-in and OpenTelemetry: instrument once, keep the exit open
Instrument with OpenTelemetry and run your own Collector, and switching vendors becomes a config change instead of a re-instrumentation project. That is our default advice whichever backend you choose.
Datadog supports OpenTelemetry through several paths: the DDOT Collector embedded in the Agent (Datadog's recommended option), OTLP ingest in the Agent, or a standalone Collector with the Datadog exporter. Note that OTLP log ingestion in the Agent is off by default. Grafana Cloud accepts OTLP natively, and its backends are open source under AGPLv3 (Prometheus itself is Apache 2.0), so the same queries keep working if you later self-host.
The lock-in that remains is above the data: dashboards, monitors, SLOs and on-call routing. Keep those in code (Terraform providers exist for both vendors) and the exit stays realistic.
A Collector config that scrubs personal data once and can ship to either EU backend:
receivers:
otlp:
protocols:
grpc:
http:
processors:
batch:
attributes/gdpr:
actions:
- key: client.address # client IP
action: delete
- key: enduser.id
action: hash
exporters:
datadog:
api:
site: datadoghq.eu # EU1, not the US default
key: ${env:DD_API_KEY}
otlphttp/grafana:
endpoint: ${env:GRAFANA_OTLP_ENDPOINT} # from your EU stack's OTLP page
auth:
authenticator: basicauth/grafana
extensions:
basicauth/grafana:
client_auth:
username: ${env:GRAFANA_INSTANCE_ID}
password: ${env:GRAFANA_API_TOKEN}
service:
extensions: [basicauth/grafana]
pipelines:
traces:
receivers: [otlp]
processors: [attributes/gdpr, batch]
exporters: [otlphttp/grafana] # or [datadog]
Use the Collector contrib distribution (or Grafana Alloy, or Datadog's DDOT) so the Datadog exporter and the basic auth extension are included. Scrubbing in the pipeline is the cheapest place to do it: the data never reaches a processor, and you are not paying to ingest it.
The real cost of self-hosting Prometheus, Loki, Tempo and Grafana
Self-hosting is cheap in infrastructure and expensive in attention. Expect it to need an owner, not just an install.
- Storage and retention. Prometheus stores data locally on one node; for months of retention or high availability you add Mimir or Thanos. Loki and Tempo need object storage, compaction and retention configured correctly.
- Scaling. Loki and Mimir run as several components in microservices mode once volume grows. Each has its own tuning for ingestion limits, caching and query parallelism.
- Upgrades. Four or five projects, each with its own release cadence and breaking changes, usually deployed with Helm charts that also change.
- Alerting on the alerting. If your monitoring runs in the same cluster as production, an outage can take both down. You need an external heartbeat and an Alertmanager setup that pages through a path that doesn't depend on the cluster.
- Security and backups. Auth in front of Grafana, TLS between components, encrypted buckets, and access controls on logs that contain personal data.
None of this is exotic, but it is steady work. In our experience a small self-hosted stack that nobody owns is the most common reason a team discovers during an incident that logs stopped arriving two weeks ago.
Which one fits your team size
Team size and in-house Kubernetes skill decide this more than feature lists do.
- 1 to 5 engineers, no dedicated ops: Grafana Cloud free or Pro on an EU stack. You will likely stay in or near the free tier for a while, and nothing needs patching.
- 5 to 30 engineers, growing product: Grafana Cloud Pro in most cases. Datadog on EU1 if the team strongly prefers one integrated product and the budget conversation is easy; in that case set log indexing rules and per-team usage attribution in month one.
- 30+ engineers or a platform team: any of the three. Datadog for breadth, Grafana Cloud for cost control and open standards, self-hosted if you already run Kubernetes reliably and have a hard data-sovereignty requirement.
- Regulated workloads with a "no third-party processor" rule: self-hosted in your own EU account, or Grafana Enterprise run by you.
Decision guide
Answer these in order; the first clear "yes" usually decides it.
- Must telemetry stay out of any third-party processor? Self-host in your EU region.
- Is budget tight and is the team small? Grafana Cloud on an EU stack, starting on the free tier.
- Do you want one product with the deepest out-of-the-box correlation and is per-host pricing acceptable? Datadog on the EU1 site.
- Is lock-in a board-level concern? Grafana Cloud or self-hosted, with OpenTelemetry and a Collector in front either way.
Whichever you pick: create the account in the EU from the start, scrub personal data in the Collector, keep dashboards and alerts in code, and set a monthly usage review. The GDPR side of that list is covered step by step in GDPR logging and EU regions, and if you are still choosing a cloud provider, Hetzner vs AWS for EU startups and AWS vs GCP vs Azure cover that decision. For what to do once alerts fire, see our incident response runbook template.
If you would rather not run observability yourself, our managed DevOps service sets up OpenTelemetry, EU-hosted dashboards and alerting, and keeps the stack and the bill in check month after month.
Frequently asked questions
Does Datadog have an EU region for GDPR?
Yes. Datadog's EU1 site at app.datadoghq.eu is located in Germany, and each Datadog site is fully independent. You choose the site when you create the organisation, so EU teams should sign up on EU1 directly. A region alone does not make you GDPR compliant; you still need a DPA and should scrub personal data from logs.
Is Grafana Cloud cheaper than Datadog?
For most small and mid-sized teams, yes, because Grafana Cloud has a generous free tier and bills by data volume and active series rather than per host. It can get expensive if you create high-cardinality metrics or keep large log volumes for long. Compare both with your real host count, log volume and series count before deciding.
Can I move my Grafana Cloud stack from the US to the EU?
No. Grafana documents that the region of an existing stack cannot be changed. You create a new stack in an EU region and point your agents and Collectors at it. Datadog works the same way across sites, so for both vendors pick the EU location when you first sign up.
Is self-hosting Prometheus, Loki and Tempo worth it for a startup?
Usually not for a team without dedicated platform engineers. The software is free, but storage, scaling, upgrades and alerting on the monitoring itself need an owner. Self-hosting makes sense when you already run Kubernetes reliably or when contracts forbid sending telemetry to a third-party processor.