The Real TCO of Self-Managed Grafana
By Brad Bell, Principal Observability Lead
“Grafana is free” might be the most expensive sentence in observability. It’s also true, which is what makes it dangerous. The software genuinely costs nothing to download, and the team that self-hosts to save money is not wrong about the license. They’re wrong about the bill, because the license was never where the cost lived.
The real total cost of ownership of self-managed Grafana is not the software. It’s the engineer-time to operate the LGTM stack reliably at scale, plus the infrastructure it runs on, plus the value of the Grafana Cloud capabilities you forgo by staying purely open source. Model those three honestly and the “free” version often costs more than the managed one, sometimes much more. Model them badly, which is the default, and you discover the real number two years in, when three of your best engineers are spending half their time keeping the observability platform alive instead of building product.
We build on Grafana, and self-managed is genuinely the right answer for a meaningful set of organizations. This piece is not an argument against it. It’s an argument for costing it honestly, because a self-hosting decision made on the license price alone is a decision made on the wrong number.
The cost nobody puts in the spreadsheet
When a team compares self-managed Grafana against a commercial platform or Grafana Cloud, the spreadsheet usually has infrastructure on one line and a big fat zero next to “software.” The line that’s missing is the expensive one: people.
Operating the LGTM stack well is a real job, often several. Someone has to run Mimir, Loki, and Tempo as production services, each with its own scaling behavior, storage tuning, and failure modes. Someone has to keep the collection layer healthy. Someone has to handle upgrades, capacity planning, and the 2 a.m. page when the thing that watches everything else falls over. At small scale a single capable engineer can do this part-time. At enterprise scale it’s a platform team, and platform engineers are not cheap. This is the cost that turns “free” into a number that would make Finance wince if anyone ever wrote it down, which is precisely why it rarely gets written down.
The honest way to model it is to count the fraction of engineer time your observability platform actually consumes, multiply by loaded cost, and put that number on the line where “$0 software” currently sits. That single correction changes most self-hosting business cases.
The infrastructure is not the small number either
Beneath the people cost is a real infrastructure cost that scales with your telemetry, not with your headcount.
High-cardinality metrics in Mimir, high-volume logs in Loki, and trace storage in Tempo all consume compute and storage, and running them at high availability means replication, which means multiplying that consumption for redundancy. Grafana has done serious engineering to make this more efficient, the Loki architecture was substantially reworked in 2026 to scan far less data on aggregation queries, and Mimir is built to scale horizontally, but efficient is not free, and the bill grows as your environment does. The teams that get surprised here are the ones who sized their infrastructure against last year’s telemetry volume and then shipped fifty new services.
The upgrade and maintenance tax
Self-managed means you own the upgrade path, and the upgrade path has teeth.
Grafana moves quickly, which is good for features and demanding for operators. Major releases carry breaking changes, and staying current while staying stable is ongoing work. Grafana 13, for example, tightened RBAC enforcement in ways that surface sloppily scoped roles on upgrade, exactly the kind of thing that turns a routine version bump into an afternoon of firefighting if nobody read the notes. Grafana 13’s new Advisor tool helps by running health checks and surfacing recommendations, which is a genuine improvement, but it assists the maintenance work rather than removing it. Every version you fall behind makes the eventual catch-up harder, so “we’ll upgrade later” quietly compounds into technical debt with an interest rate.
The opportunity cost is the one that actually hurts
Add up the people, the infrastructure, and the maintenance, and you have a real number. But the cost that does the most damage isn’t on any invoice. It’s what your senior engineers are not building while they keep the observability platform alive.
The people good enough to operate Mimir, Loki, and Tempo at high availability are, by definition, some of your strongest infrastructure engineers. Every hour they spend on observability toil is an hour not spent on the product, the platform, or the reliability work that actually differentiates your business. For some organizations, running observability in-house is core and that tradeoff is worth it. For many, it’s paying premium engineers to reinvent a managed service, and the opportunity cost dwarfs the infrastructure line everyone was worried about.
The feature gap, stated honestly
There’s one more line that belongs in an honest model: some Grafana capabilities live in Grafana Cloud, not in the open source stack.
The clearest example is the Adaptive Telemetry suite, Adaptive Metrics, Logs, Traces, and Profiles, which automatically reduce telemetry volume and cost by identifying and trimming low-value data. These are Grafana Cloud features. A self-hosted team can absolutely manage cardinality and volume, but they do it with their own engineering effort rather than with a built-in adaptive capability, which loops straight back to the people cost. This is not a knock on open source Grafana, it’s a real factor in the comparison: part of what you pay for with Cloud is cost-optimization machinery you’d otherwise build and run yourself.
When self-managed genuinely wins
None of this means self-hosting is the wrong call. It’s the right call in identifiable situations, and naming them is part of costing the decision honestly.
Self-managed wins when data residency or air-gapped requirements make a managed cloud service a non-starter, and regulated and sovereign environments often land here. It wins when you already have a strong platform team for whom operating this stack is squarely in their wheelhouse and only a marginal addition. It can win at very large, stable telemetry volumes, where the economics of running your own infrastructure eventually beat usage-based pricing, provided you’ve honestly counted the people cost and it still pencils out. And it wins when deep control over the stack is itself a requirement rather than a preference.
The common thread is that these are decisions made with the full cost in view, not decisions made because the download was free.
How to model it honestly
A defensible self-managed TCO model has five lines, not one. Infrastructure, the compute and storage for the stack at your real volume, with high-availability replication included. People, the loaded cost of the engineer time the platform actually consumes, which is usually the largest line. Maintenance, the ongoing upgrade and operational burden. Opportunity cost, an honest read on what those engineers would otherwise build. And the capability gap, the value of the Cloud features you’re choosing to reproduce yourself or do without. Put all five against the fully loaded cost of the managed alternative over a three-year horizon, and you have a real comparison instead of a license-price illusion.
Optimize first, then decide applies here as much as anywhere. Cost your current setup honestly, and either self-managed clearly wins for your situation, or it doesn’t, and now you have a real number to take to leadership instead of a gut feeling about “free.”
Modeling the true cost of a self-managed stack, people, infrastructure, maintenance, opportunity cost, and the capability gap, is exactly what our assessment produces. The output is a number you own and can defend, whatever you decide to do with it.
Learn more here!
About the Author
Brad Bell is the Principal Observability Lead at TekStream Solutions, a Digital Resilience partner that helps organizations operate, recover, and adapt with confidence by modernizing, securing, and optimizing their digital environments. His 29 years of experience spans large-scale telecommunications infrastructure, enterprise cloud strategy, and operational transformation across industries ranging from financial services to quick-service restaurants — work that included scaling monitoring platforms to 12,000+ nodes, architecting next-generation networks, and turning reactive ops teams into reliability engineering organizations. A Grafana Certified Solutions Architect and PreSales Solution Architect, he teaches SRE immersion courses at major financial institutions and consults on observability strategy at the enterprise level, helping clients navigate the convergence of cost pressure, tool sprawl, and AI-driven operations that is reshaping how enterprises think about reliability. His perspective isn’t theoretical — it comes from having been the person on the other end of a 2 AM page, and from watching organizations make expensive, avoidable mistakes with their observability investments. At TekStream, his focus is helping enterprises build platforms that solve today’s problems and hold up against where the industry is headed.