diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index e59e96a08..2f8f272b2 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -1,12 +1,14 @@
 # GitLab CI/CD — Foxhunt
-# Compile: ci-compile pool (L4, CUDA 8.9)  |  Training: ci-training pool (L40S, 48GB VRAM)
+# Two runners: CPU (compile/build/deploy) and GPU (training)
 # CI builder image: CUDA 12.4 + Rust 1.89 + protoc + sccache (hosted on Scaleway CR)
 # Service images: pushed to Scaleway Container Registry (Kaniko)
 # Note: .gitlab-ci.yml itself is NOT in changes: filters — CI-only edits skip build
 #
-# Node pool routing: Runner default nodeSelector pool=ci-compile.
-# Override via KUBERNETES_NODE_SELECTOR_pool: "pool=<pool-name>" (k=v format required)
-# Pools: ci-compile (L4 GPU), ci-training (L40S GPU), services (DEV1-M), gitlab (GP1-XS)
+# Runner routing (by tags):
+#   kapsule,rust,docker → CPU runner → ci-compile-cpu pool (POP2-HC-32C-64G, 32 vCPU)
+#   kapsule,gpu         → GPU runner → ci-compile (L4) or ci-training (L40S)
+# Override nodeSelector: KUBERNETES_NODE_SELECTOR_pool: "pool=<name>" (k=v format)
+# Pools: ci-compile-cpu (POP2-HC), ci-compile (L4), ci-training (L40S), services, gitlab
 
 # Ensure pipeline is always created — individual job rules handle filtering.
 # Without this, GitLab CE may silently drop push pipelines when changes:
@@ -151,8 +153,7 @@ build-infra-runner:
 
 # Base template for Rust jobs — pre-baked CI builder from Scaleway CR
 # Image pre-exists in SCR; rebuild via build-ci-builder when Dockerfile changes
-# Runs on pool=ci-compile (L4: 24GB VRAM, CUDA CC 89) by default (runner nodeSelector)
-# Training jobs override via KUBERNETES_NODE_SELECTOR_pool: ci-training (L40S: 48GB VRAM)
+# Runs on ci-compile-cpu pool (POP2-HC-32C-64G: 32 vCPU, 64GB RAM) via CPU runner
 .rust-base:
   image: ${CI_BUILDER_IMAGE}
   tags:
@@ -162,11 +163,12 @@ build-infra-runner:
     # Compile uses CUDA stubs from ci-builder image — no nvidia runtime needed.
     # L4 GPU = compute capability 8.9 → candle-kernels generates sm_89 PTX
     CUDA_COMPUTE_CAP: "89"
-    # Give compile pods all available CPU for fast Rust compilation
-    KUBERNETES_CPU_REQUEST: "7000m"
-    KUBERNETES_CPU_LIMIT: "7800m"
-    KUBERNETES_MEMORY_REQUEST: "16Gi"
-    KUBERNETES_MEMORY_LIMIT: "80Gi"
+    # Use all 32 vCPUs on POP2-HC for maximum Rust compilation parallelism
+    CARGO_BUILD_JOBS: "30"
+    KUBERNETES_CPU_REQUEST: "28000m"
+    KUBERNETES_CPU_LIMIT: "31000m"
+    KUBERNETES_MEMORY_REQUEST: "24Gi"
+    KUBERNETES_MEMORY_LIMIT: "56Gi"
   before_script:
     # Mold linker fallback — .cargo/config.toml uses -fuse-ld=mold via clang.
     # clang's -fuse-ld=mold searches for "ld.mold" (not "mold") in its linker paths.
@@ -1154,6 +1156,7 @@ infra-plan:
   image: ${INFRA_RUNNER_IMAGE}
   tags:
     - kapsule
+    - docker
   needs: []
   rules:
     - if: $CI_PIPELINE_SOURCE == "merge_request_event"
@@ -1186,6 +1189,7 @@ infra-apply:
   image: ${INFRA_RUNNER_IMAGE}
   tags:
     - kapsule
+    - docker
   needs:
     - job: infra-plan
       optional: true
@@ -1214,6 +1218,7 @@ infra-drift-check:
   image: ${INFRA_RUNNER_IMAGE}
   tags:
     - kapsule
+    - docker
   needs: []
   rules:
     - if: $CI_PIPELINE_SOURCE == "schedule" && $DRIFT_CHECK == "true"
@@ -1253,9 +1258,9 @@ deploy:
   image: ${INFRA_RUNNER_IMAGE}
   tags:
     - kapsule
+    - docker
   variables:
-    # Deploy runs on ci-compile (nvidia runtime required by runner default).
-    # 200m CPU fits alongside compile since deploy runs in a later stage.
+    # Deploy runs on CPU compile pool (lightweight kubectl apply)
     KUBERNETES_CPU_REQUEST: "200m"
     KUBERNETES_CPU_LIMIT: "500m"
     KUBERNETES_MEMORY_REQUEST: "256Mi"
diff --git a/infra/live/production/kapsule/terragrunt.hcl b/infra/live/production/kapsule/terragrunt.hcl
index 003837fdd..1e3709b53 100644
--- a/infra/live/production/kapsule/terragrunt.hcl
+++ b/infra/live/production/kapsule/terragrunt.hcl
@@ -21,6 +21,11 @@ inputs = {
   ci_compile_type        = "L4-1-24G"
   ci_compile_max_size    = 1
 
+  # CPU compile pool (POP2-HC for cargo check/build — 32 vCPU, no GPU)
+  enable_ci_compile_cpu_pool = true
+  ci_compile_cpu_type        = "POP2-HC-32C-64G"
+  ci_compile_cpu_max_size    = 1
+
   # CI training pool (L40S for hyperopt + training — 48GB VRAM, CUDA CC 89)
   enable_ci_training_pool = true
   ci_training_type        = "L40S-1-48G"
diff --git a/infra/modules/kapsule/main.tf b/infra/modules/kapsule/main.tf
index 17f87e473..c7932b42a 100644
--- a/infra/modules/kapsule/main.tf
+++ b/infra/modules/kapsule/main.tf
@@ -109,6 +109,25 @@ resource "scaleway_k8s_pool" "ci_training" {
   }
 }
 
+# CPU compile pool (POP2-HC — high CPU, no GPU, for cargo check/build)
+resource "scaleway_k8s_pool" "ci_compile_cpu" {
+  count       = var.enable_ci_compile_cpu_pool ? 1 : 0
+  cluster_id  = scaleway_k8s_cluster.foxhunt.id
+  name        = "ci-compile-cpu"
+  node_type   = var.ci_compile_cpu_type
+  size        = 1
+  min_size    = 0
+  max_size    = var.ci_compile_cpu_max_size
+  autoscaling        = true
+  autohealing        = true
+  public_ip_disabled = var.public_ip_disabled
+  region             = var.region
+
+  lifecycle {
+    ignore_changes = [size]
+  }
+}
+
 # CPU pool for remote development (DevPod — autoscales to zero)
 resource "scaleway_k8s_pool" "dev" {
   count                = var.enable_dev_pool ? 1 : 0
diff --git a/infra/modules/kapsule/outputs.tf b/infra/modules/kapsule/outputs.tf
index b0a6582b6..1b1d265da 100644
--- a/infra/modules/kapsule/outputs.tf
+++ b/infra/modules/kapsule/outputs.tf
@@ -29,6 +29,11 @@ output "ci_compile_pool_id" {
   value       = var.enable_ci_compile_pool ? scaleway_k8s_pool.ci_compile[0].id : ""
 }
 
+output "ci_compile_cpu_pool_id" {
+  description = "ID of the CPU compile node pool (POP2-HC)"
+  value       = var.enable_ci_compile_cpu_pool ? scaleway_k8s_pool.ci_compile_cpu[0].id : ""
+}
+
 output "ci_training_pool_id" {
   description = "ID of the CI training node pool (L40S)"
   value       = var.enable_ci_training_pool ? scaleway_k8s_pool.ci_training[0].id : ""
diff --git a/infra/modules/kapsule/variables.tf b/infra/modules/kapsule/variables.tf
index a0d40da8d..d75b687d1 100644
--- a/infra/modules/kapsule/variables.tf
+++ b/infra/modules/kapsule/variables.tf
@@ -75,6 +75,24 @@ variable "ci_training_max_size" {
   default     = 1
 }
 
+variable "enable_ci_compile_cpu_pool" {
+  description = "Create high-CPU node pool for CI compilation (no GPU needed)"
+  type        = bool
+  default     = false
+}
+
+variable "ci_compile_cpu_type" {
+  description = "Instance type for the CPU compile pool (high CPU, no GPU)"
+  type        = string
+  default     = "POP2-HC-32C-64G"
+}
+
+variable "ci_compile_cpu_max_size" {
+  description = "Maximum number of nodes in the CPU compile pool"
+  type        = number
+  default     = 1
+}
+
 variable "enable_dev_pool" {
   description = "Create CPU node pool for remote development (DevPod)"
   type        = bool