diff --git a/docs/plans/2026-02-25-iac-pipeline-design.md b/docs/plans/2026-02-25-iac-pipeline-design.md
new file mode 100644
index 000000000..3b53e39c4
--- /dev/null
+++ b/docs/plans/2026-02-25-iac-pipeline-design.md
@@ -0,0 +1,131 @@
+# IaC Pipeline & Tfstate Migration Design
+
+**Goal:** Migrate Terraform state from nl-ams to fr-par, close IaC drift gaps, and add GitLab CI/CD for infrastructure management.
+
+**Architecture:** Terragrunt run-all with S3 backend in fr-par. MR-based plan/apply workflow. Weekly drift detection via scheduled pipeline.
+
+**Tech Stack:** Terragrunt 0.72.6, OpenTofu/Terraform 1.11.2, Scaleway S3, GitLab CI/CD, glab CLI
+
+---
+
+## 1. Tfstate Migration (nl-ams → fr-par)
+
+1. Create `foxhunt-tfstate` bucket in fr-par via `scw` CLI (bootstrap, not TF-managed)
+2. Copy all state files: `aws s3 sync` cross-region using Scaleway S3 endpoints
+3. Update `infra/live/production/root.hcl`:
+   - `region = "fr-par"`
+   - `endpoints.s3 = "https://s3.fr-par.scw.cloud"`
+4. Run `terragrunt init -migrate-state` in each module directory (kapsule, dns, object-storage, registry, block-storage, secrets)
+5. Verify with `terragrunt plan` — should show no changes
+6. Delete old nl-ams bucket
+
+**Risk:** Low. State is JSON files being copied. `-migrate-state` handles backend reconfiguration. Old bucket remains as fallback until explicitly deleted.
+
+## 2. Resource Import — Close IaC Gap
+
+### Audit Results
+
+| Resource | Actual State | In Terraform? | Action |
+|----------|-------------|---------------|--------|
+| `foxhunt-sccache` bucket (fr-par) | exists | No | Add to `object-storage` module + import |
+| `foxhunt-ci` registry namespace (fr-par) | exists | No | Add to `registry` module + import |
+| `grafana` A record (fxhnt.ai → 100.90.76.85) | exists | No | Add to `dns` module + import |
+| `prometheus` A record (fxhnt.ai → 100.90.76.85) | exists | No | Add to `dns` module + import |
+| `ci-runner/` live dir | orphan `.terragrunt-cache` | No terragrunt.hcl | Delete |
+| `foxhunt-ibkr-*` secrets | exist | Yes (applied) | No action |
+| K8s PVCs (dev-home, postgres, etc.) | exist | N/A | Leave as kubectl/Helm managed |
+| `foxhunt-tfstate` bucket (fr-par, new) | bootstrap | N/A | Not TF-managed (circular dep) |
+
+### Module Changes
+
+**`infra/modules/object-storage/main.tf`** — Add sccache bucket:
+```hcl
+resource "scaleway_object_bucket" "sccache" {
+  name   = "foxhunt-sccache"
+  region = var.region
+
+  lifecycle_rule {
+    enabled = true
+    prefix  = ""
+    expiration { days = 14 }
+  }
+  versioning { enabled = false }
+}
+```
+
+**`infra/modules/registry/main.tf`** — Add foxhunt-ci namespace:
+```hcl
+resource "scaleway_registry_namespace" "foxhunt_ci" {
+  name      = "foxhunt-ci"
+  region    = var.region
+  is_public = false
+}
+```
+
+**`infra/modules/dns/main.tf`** — Add grafana + prometheus records:
+```hcl
+resource "scaleway_domain_record" "grafana" {
+  dns_zone = var.dns_zone
+  name     = "grafana"
+  type     = "A"
+  data     = var.git_ip
+  ttl      = 300
+}
+
+resource "scaleway_domain_record" "prometheus" {
+  dns_zone = var.dns_zone
+  name     = "prometheus"
+  type     = "A"
+  data     = var.git_ip
+  ttl      = 300
+}
+```
+
+After all imports, `terragrunt run-all plan` should show zero changes.
+
+## 3. GitLab CI/CD Pipeline for IaC
+
+### CI Image
+
+`infra/docker/Dockerfile.infra-runner` — lightweight image with:
+- OpenTofu/Terraform 1.11.2
+- Terragrunt 0.72.6
+- `scw` CLI
+- `glab` CLI
+
+Built via Kaniko, pushed to SCW registry (`rg.fr-par.scw.cloud/foxhunt-ci/infra-runner`).
+
+### Pipeline Jobs
+
+**`infra-plan`** (MR gate):
+- Stage: `check`
+- Image: `${REGISTRY}/infra-runner:latest`
+- Tags: `kapsule` (gitlab pool)
+- Trigger: MR pipelines when `infra/**` changes
+- Script: `terragrunt run-all plan` across all modules
+- Output: plan summary as job artifact
+
+**`infra-apply`** (merge to main):
+- Stage: `deploy`
+- Same image + tags
+- Trigger: push to `main` when `infra/**` changed
+- Script: `terragrunt run-all apply -auto-approve`
+- Environment: `production/infrastructure`
+
+**`infra-drift-check`** (weekly scheduled):
+- Stage: `check`
+- Same image + tags
+- Trigger: scheduled pipeline (weekly, Sunday night)
+- Script: `terragrunt run-all plan -detailed-exitcode`
+  - Exit 0 = no drift
+  - Exit 2 = drift → create GitLab issue via `glab issue create`
+- No apply, detection + alerting only
+
+### Credentials
+
+Jobs use existing CI/CD variables (protected + masked):
+- `SCW_ACCESS_KEY`, `SCW_SECRET_KEY`, `SCW_DEFAULT_PROJECT_ID`
+
+### Separation
+
+Infra jobs are independent from Rust build pipeline — different triggers (`infra/**`), different image, different pool (gitlab vs gpu-training). No interference.