#85 - Is It Time for OpenTofu? Our HashiConf Takeaways
Terraform Actions, Stacks and HCP pricing put the cost of staying with Terraform under scrutiny.
16 min read
We break down 10 years of HashiConf and this year’s Terraform-heavy news. What do Terraform Actions with Ansible, Stacks GA, and HCP-only features mean for day two work? Is open source getting left behind, and is OpenTofu worth a look?
Staying on Terraform is now a decision, not a default
Picture the setup most teams actually have. A repository full of .tf files, state in an S3 bucket, terraform apply running from CI, and nobody paying HashiCorp anything. It has worked for years and it needs no defending. The question worth asking once a year is whether the tool you run from CI is still the one getting better, because that answer sets the price of staying on it.
This year the answer arrives with evidence attached. HashiConf 2025, the tenth edition and the first as an IBM company, was Terraform heavy in a way the previous conference wasn’t. Sort the announcements by where each one runs and a pattern shows up before you reach the bottom. Stacks: HCP Terraform. Search: HCP Terraform. Hold your own key: HCP Terraform. Pre-written Sentinel policies for NIST: HCP Terraform and Terraform Enterprise. The infrastructure graph everybody wrote about afterwards: announced, with private beta applications open, also HCP.
If the Community Edition is now a stable tool that gains little, staying becomes a choice you re-make every year, and OpenTofu stops being an argument about licensing and turns into a feature comparison you can run with diff. Two items on the list do land in the CLI, and both were available in a beta release about a week before the keynote. That detail matters later, so hold on to it.
Day two work finally gets a block instead of a shell script
Start with the one that touches code you already wrote.
What do Terraform actions replace?
The null_resource with a local-exec provisioner you reach for when an apply needs a follow-up step, like invalidating the CloudFront cache after Terraform pushes a new object into the S3 bucket behind it.
The variants are familiar: a curl to a webhook, a CLI call the provider never covered, a sleep. HashiCorp’s framing is that actions “codify and drive Day 2 infrastructure operations by triggering third-party tools outside of Terraform”. They’re built into providers rather than written by you. In the 1.14 beta, an action_trigger inside a resource’s lifecycle block fires on before_create, after_create, before_update and after_update, or you invoke the action on its own with terraform apply -invoke. Destroy-time triggers are not in this release.
The integration is Terraform actions with the Red Hat Ansible Automation Platform provider, in beta, dispatching an event that triggers Event Driven Ansible. People have been gluing Terraform and Ansible together for a decade, usually while calling the combination Terrible, and after the IBM acquisition the joke became a shipping product. If you run immutable infrastructure you need little of this: the configuration happens in the image build, and the instance gets replaced rather than adjusted. Plenty of estates aren’t immutable, and the servers in them still need patching, draining and checking on a Tuesday. That’s the audience.
Before you plan around it: the action block is a Terraform language feature, not an HCP product. It was already in 1.14.0-beta1, published on September 18, 2025, about a week before HashiConf opened on September 24.
Stacks went GA, and what it is depends on who you ask
Actions are small and concrete. The next item is neither.
What is a Terraform stack, exactly?
A configuration layer above modules: you break infrastructure into components, then define deployments that instantiate those components, with orchestration rules about when plans and applies may run across them.
That’s the documentation’s own framing. Define a networking component once, deploy it to dev, staging and production as three deployments, and stop maintaining three root modules that drifted apart in 2023. The idea isn’t new either. Kief Morris named the infrastructure stack as a unit in the second edition of Infrastructure as Code, and the TACOS vendors have shipped their own version of it for years.
This is where we don’t agree. Mattias doesn’t trust abstractions that decide what you’re allowed to express: the first time someone needs a thing the stack won’t do, they’ll solve it outside the stack and you’ll find out later. He’d rather spend the effort watching what’s actually running. Andrey reads Stacks as a platform engineering play rather than a guardrail: the platform hands you components, you deploy them, and nobody is claiming it stops you. Paulina’s open question is the practical one, whether a deployment can be pinned to the version of the application running on it, or whether that coupling stays in another system.
We’re guessing, and we should be honest about why. Stacks is generally available for HCP Terraform plans, with custom deployment groups and auto approve checks on the Premium tier, and none of us had tried it. The first limit is HashiCorp’s product decision. The second is ours: the same announcement put $500 of credit on the table across all plans.
The resource that costs nothing at AWS still costs you a resource
Which leads to the number that decides whether you ever try it on anything real.
Which resources count toward an HCP Terraform bill?
The managed ones in your state, whatever they cost at the cloud provider, which is why the meter looks nothing like your AWS invoice.
HCP Terraform is priced per resource under management. The billing definitions are specific: a resource counts when mode = "managed" in the state file, null_resource and terraform_data are excluded, and the billed quantity is hourly peak managed resources rather than a headcount taken once a month. A security group is free at AWS and counts as one resource here, and so does each rule you split out of it. An IAM role, its policy and the attachment between them are free too, and that is three more.
On air we put the meter at about a dollar per resource. The real rate is hourly and banded. Take HCP Terraform Standard Edition with Silver support, under a HashiCorp Flex pre-commit contract, in the consumption table effective July 1, 2025: the first 500 managed resources are free, resources 501 to 1,000 cost $0.0001359 each per hour, and the bands step down above that. A thousand billable resources is around $50 a month, well under the flat-rate arithmetic.
That cuts both ways. A team with 400 managed resources pays nothing on Standard, better than any of us described on air. The governance edition is where the picture changes: Plus, the one carrying policy as code, drift detection and audit logging, has no free band and starts at $0.0009511 per resource-hour. The newest features sit on negotiated tiers, so what a mid-sized estate pays sits inside an agreement nobody outside it sees.
HashiCorp has always known who it was aimed at. We once asked Mitchell Hashimoto what made the open source company work where others struggled to monetize: they knew the customer from the start, the top 500, and built products like Vault for their problems. Terraform won the cloud era without an obvious business model attached, and HCP is the model. The friction is that features for the tool most teams run keep landing only on the hosted side.
Bulk import got better, and half of it is in the CLI
Pricing aside, one announcement solves a problem we hit constantly.
What does Terraform Search actually do?
It finds resources that exist in your cloud accounts but not in your state, and helps you import them in bulk instead of one import block at a time.
Everyone meets this. You adopt an account somebody clicked together, you inherit a team, or your credentials expire halfway through an apply and the state ends up describing less than reality. The cleanup has usually meant manual work or a separate discovery tool, which is why it rarely happens. Search is in public beta on HCP Terraform, covers the AWS and Azure providers to start, and builds on the resource identity work that landed in Terraform 1.12.
Our guess on air was that features like this might trickle back to the CLI eventually. The release notes say the primitive already had. Terraform 1.14 adds list resources, declared in *.tfquery.hcl files, and a terraform query command that runs list operations against real infrastructure and can generate import configuration from the results. That is in the Community Edition CLI, in the beta published six days before the conference opened.
So the honest version is split: the discovery primitive runs in the binary you already have, and the console that makes it usable at scale, with a UI and bulk selection across workspaces, is the product.
MCP, hold your own key and Sentinel: the client is free, your context isn’t
The split repeats in every remaining item on the list.
Is the Terraform MCP server worth wiring up?
It’s worth it when it can answer questions about your estate rather than about the public registry, and that’s the part tied to an account.
Everyone shipped an MCP server in 2025, to the point where the joke about your grandmother having one writes itself. The Terraform MCP server has been in public beta since May 2025 and runs locally. What HashiConf added was authentication against HCP Terraform or Terraform Enterprise, so an agent can work against your own account instead of the public registry: your organizations, projects and workspaces, your private module and provider registry, and starting a run. That is the version anyone actually wants, and it needs the account. Whether it can also show you where your infrastructure has drifted is the question we had on air, and the toolset documented at the conference doesn’t answer it. The Azure Copilot integration announced alongside it works through the same server.
The rest follows the shape. Hold your own key went GA in HCP Terraform, so bringing your own encryption key for sensitive data is a hosted feature. That is what a vendor inside a public company does with a feature enterprises will pay for, and HashiCorp has been inside IBM since February 27, 2025. The new pre-written Sentinel policy set for NIST SP 800-53 Rev 5 on AWS is readable by anyone, but Sentinel enforcement runs in HCP Terraform and Terraform Enterprise, so from the CLI it’s reference material rather than a control. A new run task wires Terraform into IBM’s Cloudability governance, which tells you where the integration roadmap now points.
Not every paywall is only a toll booth. Running Vault yourself means unsealing, patching, the PKI and the audit trail, and this year’s Vault, Boundary and Radar list is where paying HashiCorp still looks worth it, at licence prices aimed at banks.
Is it time for OpenTofu?
Put the list back together and the conclusion assembles itself.
Does any of this change the case for OpenTofu?
It moves the case off licensing and onto delivery: how much of each release reaches the binary you actually run.
HashiCorp moved its products to the Business Source License in August 2023, and the arguing finished there. What you run from CI is the source-available Community Edition, and OpenTofu is the MPL fork of the last open source release.
That was the read in the room, and the release notes complicate it. Two substantial features on the HashiConf list, action blocks and terraform query with list resources, were in the Community Edition beta before the keynote and need no subscription. We said on air that we’d correct ourselves once we’d run the new release. This is that correction, and it goes HashiCorp’s way: the orchestration and the consoles are paid, while the language primitives underneath them still ship to everyone. Whether that holds for the next three releases is the thing to watch.
The market has not moved as fast as the argument either. Across our customer work and our community, which skews toward Denmark and toward smaller teams, we haven’t met anyone paying for the HashiCorp platform. That’s a small sample, so treat it as a data point rather than a trend.
A fork is only worth the switch if it ships, and OpenTofu has been shipping. State encryption arrived in 1.7, covering the state and the plan file in the tool itself. Terraform leaves the second one out: state at rest is whatever your backend encrypts, and the command reference says a plan written with terraform plan -out keeps sensitive data “in cleartext in the plan file”. OpenTofu 1.10, from June 2025, added OCI registries for providers and modules. It also added native S3 state locking, but Terraform 1.10 shipped the same thing in November 2024, so put that one in the parity column rather than the reasons column.
One caveat from teams who have done it, Timur Bublik’s migration across 250 to 300 stacks among them: swapping the binary is the easy part. The work is everything already broken that the migration makes you look at.
Project Infragraph is the announcement we would bet on
The last item on the list is the one that isn’t really about Terraform.
Why would an AI agent need a graph of your infrastructure?
Because a model that knows everything about AWS in general still knows nothing about your account, and discovering it one API call at a time is slow, expensive and incomplete.
HashiCorp describes Project Infragraph as a real time infrastructure graph connecting infrastructure, applications, services and ownership, fed by the data their products emit about secrets, identity, access, state and policy compliance. Private beta applications opened at the conference, with the beta expected in December 2025.
The diagnosis is correct, and it’s the one problem on this list we’d call industry wide. Your source code is in one place, your runbooks in another, your logs in a third, your cloud accounts in a fourth, and none of them knows about the others. When a colleague says a server is misbehaving, you answer accurately because you carry the connections in your head. An agent has no such model, so it either asks you or burns tokens rediscovering the estate on every question. Forget the internal developer platform for a moment: what you need first is a queryable graph of your own company for an agent to ground itself in.
The catch is the one from every other section: it’s an HCP capability, built to pull in more of the IBM portfolio, with connectors for the rest. Credit where it’s due though: this is HashiCorp running ahead of the industry, and we expect several companies to arrive at the same answer.
What this means for teams
Nothing here forces a migration this quarter. What changes is the shape of the question. Picking Terraform used to mean picking the tool that got better fastest. Now the parts that get better fastest and the binary you run from CI may not be the same artefact, and that’s worth deciding deliberately rather than by inertia.
This week:
- Tag each announcement you care about with where it runs: CLI, HCP Terraform, Terraform Enterprise, or not shipping yet. The tags, not the feature names, tell you how much of it is yours.
- Count your billable managed resources with
terraform state list, drop the data sources,null_resourceandterraform_dataentries, and walk what’s left through one named plan’s bands. On Standard the first 500 are free, so a 400 resource estate and a 4,000 resource estate are different conversations. - Try the 1.14 beta’s
terraform queryagainst the messiest account you own, the one full of resources nobody imported. It’s the cheapest way to find out whether the Community Edition import story is good enough now. - Write down, in one sentence, what would make you switch to OpenTofu. Agreeing on that before the next release cycle is easier than arguing about it during an incident.
Common questions, answered
Do I need HCP Terraform to use Terraform Stacks?
Yes. Stacks is an HCP Terraform feature, generally available on plans billed by resources under management, with custom deployment groups and auto approve checks on the Premium tier. There is no Community Edition equivalent you can run from the CLI.
Without it, the closest options are the stack style abstractions the Terraform automation vendors built, or a disciplined root module layout with a thin orchestration layer in your pipeline.
Can I use Terraform actions without paying for HCP Terraform?
Yes. The action block is part of the Terraform language in 1.14, so it works from the CLI, and terraform query with list resources ships the same way. You need a provider that implements actions for the resources you care about, and triggers in this release fire around creates and updates only.
The part that needs a licence is on the Ansible side. Triggering an Event Driven Ansible workflow requires the Red Hat Ansible Automation Platform provider and a platform to point it at.
How do I estimate what HCP Terraform would cost us?
Count the billable managed resources in your state files, then walk that count through the bands of one named plan instead of multiplying by a flat rate. On Standard Edition with Silver support, in the HashiCorp Flex consumption table effective July 1, 2025, the first 500 are free and the next 500 are $0.0001359 per resource-hour, so a thousand billable resources is roughly $50 a month.
Billable means mode = "managed" in state, with null_resource and terraform_data excluded. The meter doesn’t care what a resource costs at AWS: a security group, each of its rules and an IAM policy attachment count the same as an RDS cluster.
Related episodes
- Episode 83, OpenTofu vs Terraform: Where We Are Now with Cole Bittel is the feature by feature comparison this one only gestures at, from someone who contributes to OpenTofu.
- Episode 64, From Terraform to OpenTofu: Story From the Trenches covers what a real migration cost across hundreds of stacks, and what broke that had nothing to do with the fork.
- Episode 54, HashiCorp’s BSL Move and OpenTF is where this started: the licence change, the fork, and what we expected would follow.
- Episode 18, HashiConf Special is the 2020 version of this write-up, back when the announcements were Boundary and Waypoint and the tools were still MPL licensed.
Resources
- Scale infrastructure with new Terraform and Packer features at HashiConf 2025, HashiCorp settles the release stage and edition of each item: Stacks GA and hold your own key GA on HCP Terraform, Search and actions in public beta, the $500 trial credit, and HashiCorp’s own definition of what actions do.
- Terraform and Ansible: unifying infrastructure provisioning and configuration management, HashiCorp settles how the Ansible provider dispatches an Event Driven Ansible workflow.
- Terraform v1.14.0-beta1 release notes settle that
actionblocks, list resources in*.tfquery.hclandterraform queryare CLI features, published on September 18, 2025. - Terraform Stacks documentation settles the component and deployment model, the orchestration rules, and where Stacks runs.
- HashiCorp Flex pricing table and terms, effective July 1, 2025 settles the billing model and the bands: hourly peak managed resources, 500 free on Standard Edition, stepped rates above that, and no free band on Plus.
- Strengthen security with Vault, Boundary and Radar features at HashiConf 2025, HashiCorp settles the security side of the list: Vault Enterprise 1.21, the OpenShift CSI provider, Radar Jira scanning and Boundary RDP credential injection.
- HashiCorp previews the future of agentic infrastructure automation with Project Infragraph, IBM Newsroom settles what Infragraph is, that the beta was expected in December 2025, and that this was the tenth HashiConf and the first as an IBM company.
- OpenTofu 1.10.0: a well seasoned release settles what the fork shipped in June 2025, including OCI registries and native S3 state locking.
Since we recorded
Both CLI features we flagged landed for real: Terraform 1.14.0 went GA on November 19, 2025 with action blocks, list resources and terraform query in the Community Edition release, so the day two and discovery primitives are available without an HCP plan. OpenTofu kept its cadence with 1.11.0 on December 9, 2025, adding ephemeral values, write only attributes and an enabled meta-argument for resources, and 1.12.0 on May 14, 2026. Project Infragraph came out of private beta into limited availability on June 30, 2026, for HCP Terraform Standard and Premium customers on US instances only, with connectors for AWS, Azure, GitHub and Terraform Enterprise. We returned to the wider question in episode 101, Infrastructure as Code in 2026, where stacks as a pattern and the role of AI in infrastructure work get a longer hearing.
