A five-person DevOps team opened their Terraform Cloud renewal email in Q1 2026. The annual bill jumped from $180K to $270K. No new features justified the increase. The team lead forwarded it to Slack with one line: “We need to talk.”
Three months later, they migrated 600 Terraform modules to OpenTofu, replaced Fivetran with self-hosted Airbyte, and swapped Datadog for Grafana plus SigNoz. Annual infrastructure tooling spend dropped 58%.
This is not an isolated incident. In 2026, open source infrastructure is systematically eroding commercial SaaS territory across IaC, data integration, and observability.
The license trust collapse
The story starts in August 2023.
HashiCorp switched Terraform from Mozilla Public License (MPL 2.0) to Business Source License (BSL 1.1). BSL is not an OSI-approved open source license. It prohibits anyone from using Terraform to build a “competing product,” and the definition of “competing” is vague enough to make legal teams nervous.
The community responded fast. By September 2023, the Linux Foundation announced OpenTofu, a fork based on Terraform’s last MPL-licensed version. IBM, Alibaba, Gruntwork, and Spacelift backed it immediately.
Then in late 2024, IBM acquired HashiCorp for $6.4 billion. Terraform, Vault, Consul, Nomad became IBM assets. IBM needs to recoup that investment. Tighter enterprise licensing, feature gating, ecosystem lock-in are the predictable commercial path.
This was not the first time. Elastic changed Elasticsearch from Apache 2.0 to SSPL in 2021, and AWS forked OpenSearch within weeks. MongoDB did the same move in 2018. In 2024, Elastic added an AGPL option under community pressure, but trust, once broken, costs more to repair than to maintain.
As of April 2026, 38% of Terraform users are evaluating or actively migrating to alternatives. OpenTofu downloads grew over 300% year-over-year, approaching 10 million cumulative downloads. GitHub stars reached 29,300 by June 2026.
The logic is simple: when your infrastructure tooling vendor can change the rules at any time, you either accept the risk or take back control.
The cost math
Cost is the primary migration driver. It beats ideology every time.
Take data integration. Fivetran charges by Monthly Active Rows (MAR). A mid-size SaaS company processing 50 million rows per month typically pays $120K-$180K annually. Self-hosted Airbyte running on a Kubernetes cluster costs roughly $20K-$40K per year in infrastructure, plus about half an engineer’s time for maintenance. Net savings: 50-70%.
A more extreme case: an e-commerce platform had a single high-frequency table updating 8 million rows daily. Their Fivetran bill tripled in one quarter. After switching to Airbyte, the same volume produced only infrastructure costs. No per-row billing surprises.
The gap in observability is even wider. Datadog prices on a per-host, per-module SKU basis. Infrastructure monitoring runs $15-23 per host per month, APM adds $31 per host per month, log management starts at $0.10 per GB, and Synthetics charges per test run. A 50-host team running Infrastructure plus APM plus Logs easily crosses $5,000 per month. That is $60K or more annually.
Self-hosted SigNoz with ClickHouse and OpenTelemetry? Infrastructure costs for the same 50-host scale land around $800-$1,500 per month (cloud instances plus storage), covering logs, traces, and metrics. Even including engineering maintenance time, total cost is 20-30% of Datadog. SigNoz’s own published data claims customers save up to 80% after migration.
IaC follows the same pattern. Terraform Cloud’s Team and Governance tier starts at $70 per user per month. Enterprise pricing is custom but generally exceeds $100 per user per month. OpenTofu paired with open source CI/CD pipelines (GitHub Actions, GitLab CI) brings tooling-layer costs to near zero, assuming feature requirements are met.
Technical maturity has caught up
Two years ago, choosing open source alternatives required a leap of faith. Not enough connectors. Incomplete documentation. Small communities. Those were reasonable concerns.
In 2026, the situation has changed.
OpenTofu has reached version 1.12.x, and its codebase diverged significantly from Terraform in 2025. Features unique to OpenTofu include native state file encryption (supporting AWS KMS, PBKDF2, and other key providers), dynamic provider configuration, looped import operations, and provider iteration. Some of these do not exist in Terraform at all. Others require the paid Enterprise edition.
The migration path remains smooth for now. Both projects share a compatible core language and state file format. Existing Terraform modules run on OpenTofu with minimal or zero modification. But this window is narrowing. As each project introduces exclusive features, future migration complexity will increase.
Airbyte offers over 350 connectors in its open source version as of 2026, covering mainstream SaaS products, databases, and API sources. The Connector Development Kit (CDK) lets Python developers build custom connectors in a day or two. Airbyte Cloud provides a managed option, typically priced 30-50% below Fivetran.
For comparison: Fivetran completed its merger with dbt Labs in June 2026, and its connector count exceeds 500 with higher consistency (all maintained by paid engineering staff). But Airbyte’s advantage is flexibility. Self-hosting means data never leaves your network, which is a hard requirement for regulated industries.
Grafana plus SigNoz covers the three pillars of observability: metrics, logs, and traces. SigNoz is built on OpenTelemetry natively and uses ClickHouse for storage (the same columnar database that Uber and ByteDance run at scale). Query performance holds up well under heavy data volumes. Grafana’s dashboard ecosystem and alerting engine are already industry standard. Together, they match Datadog’s core scenarios.
PostgreSQL barely needs an argument. It is the de facto open source relational database standard, and its performance and features have caught or surpassed many commercial options. Choosing AWS RDS PostgreSQL or Google Cloud SQL for PostgreSQL over Aurora or Cloud Spanner preserves migration capability. Your data format and query syntax work in any PostgreSQL-compatible environment.
Composable over all-in-one
Fivetran plus dbt Labs completed their merger in June 2026, serving over 100,000 data teams. The combined pitch is end-to-end data infrastructure on a single platform: ingestion, transformation, and governance bundled together.
The open source camp rejects this model.
Composable architecture means choosing the best tool for each layer and using open standards as the glue. SQL, OpenTelemetry, and Apache Iceberg are the integration points. Data integration runs on Airbyte. Transformation runs on dbt Core (still open source under Apache 2.0). Orchestration uses Dagster or Prefect. Storage uses Iceberg format. Analytics runs on DuckDB or ClickHouse.
The benefits are concrete.
No single point of failure. A Fivetran outage takes down your entire pipeline. Self-hosted Airbyte’s availability depends on your own Kubernetes cluster, which you control.
Transparent pricing. You pay for compute and storage, not for a “platform.” No MAR multipliers creating billing surprises.
Freedom to swap. When a better tool emerges, replacing one component costs far less than replacing an entire platform. OpenTelemetry’s OTLP protocol means your instrumentation code works unchanged when you switch backends, from SigNoz to Grafana Tempo or back.
The trade-off is real: you own the integration layer. Version compatibility testing, glue code, and operational overhead fall on your team. This is not free. It is a deliberate exchange of engineering time for control.
Why open source ships features faster
OpenTofu delivered multiple features within two years of forking that Terraform had failed to ship for much longer. Client-side state encryption was an RFC submitted to Terraform in 2016. OpenTofu landed it in under a year.
The difference is governance. OpenTofu operates under the Linux Foundation with a Technical Steering Committee that meets biweekly. Community members participate directly in RFC discussions and code contributions. Terraform, under IBM, follows a product roadmap driven by commercial priorities. New features are more likely to appear in the paid Terraform Cloud or Enterprise tier than in the open source version.
Airbyte’s connector development model follows a similar pattern. The CDK lowers the contribution barrier, and community developers respond quickly to new SaaS product integrations. When Notion updates its API, community patches often land faster than a commercial vendor’s scheduled release.
Grafana’s ecosystem is the textbook example. Grafana Labs keeps core products (Grafana, Loki, Tempo, Mimir) open source while generating revenue through Grafana Cloud and enterprise support. The model proves open source and commercial viability can coexist, as long as you do not treat the open source community as unpaid beta testers.
The honest counter-argument
Open source alternatives are not a universal answer. The narrative often glosses over several realities.
Operational overhead is not zero. Self-hosted Airbyte requires Kubernetes cluster maintenance: upgrades, scaling, incident response. OpenTofu state management needs you to build and maintain a remote backend. SigNoz’s ClickHouse cluster needs tuning as data volumes grow. All of this requires dedicated people. A 20-person startup without a platform engineering team may find the hidden costs of self-hosted Airbyte exceed Fivetran’s bill.
SLAs and commercial support carry real value. Datadog commits to 99.9% availability with dedicated technical account managers. Self-hosted SigNoz availability is your problem. When ClickHouse runs out of memory at 3 AM, nobody is on call for you. For financial services, healthcare, and other regulated industries, vendor SOC 2 Type II certification, HIPAA compliance, and FedRAMP authorization are not optional.
UX gaps exist. Fivetran handles schema drift automatically, includes anomaly detection, and manages retries out of the box. Airbyte is less polished in these areas. Datadog’s Session Replay, Synthetic Monitoring, and Real User Monitoring form a digital experience suite that has no direct open source equivalent.
Not every organization can self-host. A data team with three analysts and zero infrastructure engineers cannot realistically operate an Airbyte cluster on Kubernetes. Fivetran’s zero-maintenance value proposition is real for these teams. They are paying an insurance premium for not having to worry about it.
A pragmatic migration path
The smart strategy is neither “go all-in on open source” nor “stay locked into commercial tools.” It is understanding where your organization’s boundaries sit.
Three questions to evaluate:
How much engineering bandwidth can you dedicate to infrastructure maintenance? If the answer is “almost none,” full self-hosting may not fit. Airbyte Cloud or SigNoz Cloud offer a middle path: open source core, managed operations.
How strict are your data compliance requirements? If data cannot leave a specific geographic region or network boundary, self-hosting is close to your only option regardless of cost.
What does your growth curve look like? If data volume grows more than 3x annually, usage-based commercial SaaS pricing escalates fast. Self-hosted open source has near-linear marginal costs, driven mainly by storage and compute.
Migrate in stages rather than all at once. Start in non-critical environments. Use OpenTofu for dev environment infrastructure. Sync a few non-core data sources through Airbyte. Monitor a handful of microservices with SigNoz. Build operational confidence, then expand to production.
For Terraform-to-OpenTofu specifically, the migration window is still friendly. Core language and state file compatibility keep risk low and engineering effort small. But as both projects diverge further, the cost of waiting only goes up.
What 2027 looks like
Industry consensus points to accelerating open source infrastructure adoption over the next two years. The drivers: AI workload costs creating pressure to cut spending elsewhere (GPUs are expensive enough without observability and data pipelines eating budget too), multi-cloud becoming standard (amplifying vendor lock-in risk), and platform engineering culture spreading (more companies building internal infrastructure teams with the skills to self-host).
The Fivetran-dbt Labs merger is a defensive move by the commercial camp, reducing customer switching motivation through bundling. IBM will likely gate more Terraform features behind paid tiers, pushing OpenTofu adoption further. Datadog’s 15-plus product SKU model is being structurally unbundled by the OpenTelemetry ecosystem.
Open source infrastructure will not fully replace commercial SaaS, the same way Linux did not kill Windows. But it is resetting the baseline for what counts as a reasonable price and an acceptable level of control. When open source solutions cover 80% of the functionality at 20-30% of the cost, the remaining 20% feature gap requires commercial vendors to make a very convincing case.
Ultimately, this is not a technology selection problem. It is an organizational capability question: can you operate the tool stack you choose? If yes, open source gives you freedom and economics. If no, commercial SaaS sells you the service of not having to worry about it. Both are valid choices. The key is being honest about where you stand.
Stay updated with our latest AI insights
