Your data catalog license costs zero dollars. Your data catalog bill might still run past one hundred thousand dollars a year — it just shows up on payroll instead of an invoice. If you are comparing DataHub, the open-source metadata platform that grew out of LinkedIn, against Collibra, the enterprise governance suite, the license-price gap is the least interesting number in the comparison. The interesting number is the engineering-hours line neither vendor puts on their pricing page.
This guide walks through what each option really costs, where the hidden spending lives, and how a small or mid-size business should budget a data catalog decision like the capital project it is.
What Collibra Actually Costs
Collibra does not publish list prices — you talk to sales — but third-party reporting gives a consistent picture. AWS Marketplace listings put the Collibra Cloud Platform base subscription at roughly $170,000 per year for platform access alone. Industry comparisons place the typical enterprise contract between $170,000 and $510,000 or more per year depending on data-asset volume, user count, and modules.
And the base subscription is only the first line item. Budget reviewers consistently flag five more:
- Module add-ons. Data lineage and data quality are separate licensed modules, each with its own pricing.
- Implementation services. Enterprise rollouts typically take six to twelve months, much of it billable professional services.
- Hosting and infrastructure. Even cloud editions can carry consumption or hosting components depending on the contract.
- People costs. Someone on your side still has to steward the glossary, approve workflows, and govern the governors.
- Ongoing maintenance. Upgrades, troubleshooting, and keeping integrations healthy as schemas and BI workspaces change.
None of this means Collibra is overpriced. For a regulated bank or insurer that needs workflow-driven governance, audit trails, and FedRAMP-authorized hosting, that spend buys compliance posture that would cost far more to build. But it sets the baseline honestly: the "expensive" option costs roughly $170,000 a year before you staff it.
What DataHub Actually Is
DataHub is an open-source metadata platform, originally built at LinkedIn and now maintained under the Apache 2.0 license with a managed edition, DataHub Cloud, sold by Acryl Data. Its pitch is the metadata graph: an event-driven system that ingests technical metadata from more than a hundred sources — warehouses, BI tools, transformation pipelines, access systems — and serves discovery, column-level lineage, a business glossary, and data contracts from one model.
There are three ways to run it:
- Self-hosted Docker deployment for development and small teams.
- Self-hosted Kubernetes deployment via Helm, the recommended path for production.
- DataHub Cloud, the managed SaaS edition with zero infrastructure to operate, SLA backing, observability monitors, and an AI assistant — priced by quote, like most enterprise SaaS.
The open-source core costs nothing to license. That sentence is true and, standing alone, dangerously incomplete.
The Honest Comparison
Put the two side by side on total cost of ownership rather than license fees and the picture changes:
| Cost line | Collibra | Self-hosted DataHub | DataHub Cloud |
|---|---|---|---|
| License | ~$170k+/yr base | $0 | Quote-based SaaS |
| Infrastructure | Bundled or consumption | Your cloud bill: compute, search, storage | Bundled |
| Implementation | 6–12 months, often with services | 2–4 weeks to first value, longer to production-harden | Days to weeks |
| Ongoing engineering | Low (vendor operates it) | 0.25–0.5 FTE or more | Low |
| Governance depth | Deepest: workflows, compliance frameworks | Build-or-buy: glossary and contracts included, workflows are yours to build | Middle ground plus monitors |
The market has noticed the gap in the middle. Modern SaaS catalogs such as Atlan and Secoda price in the roughly $50,000–$200,000 per year band with one-to-eight-week deployments — precisely the buyers who flinch at Collibra quotes but cannot staff a platform team.
The Engineering-Hours P&L Line
Here is where "free" gets its real price tag. A production self-hosted DataHub is not one container. It is a distributed system: the GMS metadata service, Kafka for the event stream, OpenSearch or Elasticsearch for search, and MySQL or Postgres for system storage — plus the ingestion pipelines that keep every connector healthy as upstream schemas drift.
Each of those components needs provisioning, upgrades, backups, monitoring, and on-call coverage. Industry analyses of self-hosted open-source stacks put the pattern at roughly a quarter to a half of a full-time engineer for ongoing maintenance of a moderately complex system, and hidden ownership costs at 60–70 percent of total cost by some estimates. One widely cited comparison found a self-hosted Sentry setup for a fifteen-engineer team running about $9,500 per year once servers, setup, upgrades, and a single major incident were counted — above the hosted alternative.
Do the arithmetic with your own loaded cost. A platform engineer with a fully loaded cost of $180,000 per year, spending a quarter of their time operating the catalog, is a $45,000 annual line item before cloud infrastructure. At half an FTE it is $90,000 — more than half the starting price of the enterprise suite you were avoiding. Add initial setup (two to four weeks), major-version migrations (roughly twice over two years for a fast-moving project), and the incident you have not had yet, and "zero license cost" can easily land between $60,000 and $120,000 a year in real money.
That is still less than Collibra's starting price. But it is not zero, it is not free, and — critically — it is engineering payroll, the scarcest budget in most small businesses. Every hour your platform engineer spends tuning OpenSearch shards is an hour not spent on your product.
Questions that surface the real number
Before you commit to self-hosting, answer these in writing:
- Who owns upgrades? Name the person, not the team. If they leave, what breaks?
- What is the on-call story? Metadata systems fail at 2 a.m. like everything else. Is there a rotation, or is it one hero?
- How many connectors will you actually run? Each integration is a maintenance surface. Ten healthy connectors cost meaningfully more than two.
- What is your cloud baseline? Price the search cluster, Kafka, and database on your cloud provider before you start, not after.
- What does month eighteen look like? Budget two major upgrades and at least one painful incident in the first two years.
If you cannot answer all five, you do not have a cost advantage — you have an unpriced risk.
When Each Option Wins
Self-hosted DataHub wins when you already have a platform team with Kubernetes and streaming experience, your governance needs are technical (discovery, lineage, contracts) rather than regulatory, and you value extensibility — the open metadata model and API let you build exactly what you need. Data teams that live in dbt, Snowflake, and Airflow tend to get value in weeks.
Collibra wins when governance is the product: regulated industries, formal stewardship workflows, audit evidence, and compliance frameworks that an auditor needs to recognize. The six-to-twelve-month implementation is a feature in this world, not a bug — it is the time it takes to stand up governed process, and the vendor's professional services have done it hundreds of times.
The managed middle wins when you want DataHub's model without operating it (DataHub Cloud removes the infrastructure and upgrade burden while keeping the same metadata standard) or you want SaaS time-to-value with collaboration baked in (the modern SaaS tier). For most businesses without a dedicated data-platform team, this band deserves a serious look before defaulting to either extreme.
Budget It Like a Capital Project
Whatever you choose, put all four lines in the budget proposal — license, infrastructure, engineering time, and implementation — over a 24-month horizon. The most common failure mode in this decision is not picking the wrong tool; it is approving the license line while the other three lines spend invisibly out of payroll and the cloud bill.
Track the engineering hours against the project the way you would track any capitalizable effort: tag the time, review it monthly, and compare actuals to the business case at month six. If self-hosting was supposed to save $100,000 a year but the time log shows half an FTE plus a $1,500 monthly cloud bill, you want to know that while you can still switch — not at renewal.
Accurate bookkeeping from day one is what makes this comparison possible. When every infrastructure invoice is categorized and engineering time is tracked against projects, the build-vs-buy review writes itself. When those costs dissolve into general overhead, every tool looks free and every renewal is a surprise.
Keep Your Tooling Spend Visible from Day One
As you weigh open-source against enterprise software, maintaining clear records of infrastructure, labor, and subscription costs is what turns a guess into a decision. Beancount.io provides plain-text accounting that gives you complete transparency and control over your financial data — no black boxes, no vendor lock-in. Get started for free and see why developers and finance professionals are switching to plain-text accounting.
