Create aks/nebius get started scenarios

docs/content/getting-started/build-the-platform.md

@@ -22,6 +22,14 @@ against this capacity without knowing which cluster it runs on.
 - A GCP account with permissions to create GKE clusters, VPCs, and IAM roles
 - A GCP service account JSON key
 {{< /tab >}}
+{{< tab "AKS" >}}
+- An Azure account with permissions to create AKS clusters and managed identities
+- An Azure service principal JSON with `clientId`, `clientSecret`, `subscriptionId`, and `tenantId`
+{{< /tab >}}
+{{< tab "Nebius" >}}
+- A Nebius account with permissions to create clusters
+- A Nebius service account JSON key and your project ID
+{{< /tab >}}
 {{< /tabs >}}
 
 ## Set up the InferenceGateway
@@ -99,6 +107,46 @@ curl -fsSL {{< manifest-url "getting-started/clusterproviderconfig-gke.yaml" >}}
 ```
 {{< /editCode >}}
 {{< /tab >}}
+
+{{< tab "AKS" >}}
+Create a Kubernetes secret from your service principal JSON:
+
+{{< editCode >}}
+```bash
+kubectl create secret generic azure-credentials \
+  --from-file=credentials.json=$@</path/to/azure-sp>$@.json \
+  -n crossplane-system
+```
+{{< /editCode >}}
+
+Apply the `ClusterProviderConfig` referencing your secret:
+
+{{< manifests "getting-started/clusterproviderconfig-azure.yaml" >}}
+{{< /tab >}}
+
+{{< tab "Nebius" >}}
+Create a Kubernetes secret from your service account JSON:
+
+{{< editCode >}}
+```bash
+kubectl create secret generic nebius-credentials \
+  --from-file=credentials.json=$@</path/to/nebius-sa>$@.json \
+  -n crossplane-system
+```
+{{< /editCode >}}
+
+Apply the `ClusterProviderConfig`, setting `projectID` to your Nebius project:
+
+{{< manifests path="getting-started/clusterproviderconfig-nebius.yaml" apply="false" >}}
+
+{{< editCode >}}
+```bash
+curl -fsSL {{< manifest-url "getting-started/clusterproviderconfig-nebius.yaml" >}} \
+  | sed 's/project-e00example/$@<your-nebius-project>$@/' \
+  | kubectl apply -f -
+```
+{{< /editCode >}}
+{{< /tab >}}
 {{</tabs>}}
 
 ## Publish hardware and register the cluster
@@ -136,13 +184,33 @@ Modelplane provisions the cluster. This takes about 15 minutes:
 kubectl wait --for=condition=Ready ic/starter --timeout=20m
 ```
 {{< /tab >}}
+
+{{< tab "AKS" >}}
+{{< manifests "getting-started/aks/platform.yaml" >}}
+
+Modelplane provisions the cluster. This takes about 15 minutes:
+
+```bash
+kubectl wait --for=condition=Ready ic/aks-westeurope --timeout=20m
+```
+{{< /tab >}}
+
+{{< tab "Nebius" >}}
+{{< manifests "getting-started/nebius/platform.yaml" >}}
+
+Modelplane provisions the cluster. This takes about 15 minutes:
+
+```bash
+kubectl wait --for=condition=Ready ic/nebius-eu-north --timeout=20m
+```
+{{< /tab >}}
 {{< /tabs >}}
 
 {{< hint "note" >}}
 Modelplane is reconciling the infrastructure against the source of truth, the
 manifest you just applied.
 
-While you wait, Modelplane is creating the EKS or GKE cluster and its GPU node
+While you wait, Modelplane is creating the cloud cluster and its GPU node
 pool, then installing the inference stack with LeaderWorkerSet for multi-node
 serving, llm-d for inference-aware routing, Envoy Gateway for traffic
 management, and the storage class for model weights. This is the same reconciliation loop Crossplane uses to configure other 

docs/content/getting-started/deploying-a-model.md

@@ -27,6 +27,12 @@ The device selector matches against the capacity declared in the
 {{< tab "GKE" >}}
 {{< manifests "getting-started/gke/model-deployment.yaml" >}}
 {{< /tab >}}
+{{< tab "AKS" >}}
+{{< manifests "getting-started/aks/model-deployment.yaml" >}}
+{{< /tab >}}
+{{< tab "Nebius" >}}
+{{< manifests "getting-started/nebius/model-deployment.yaml" >}}
+{{< /tab >}}
 {{< /tabs >}}
 
 Wait until `REPLICAS` shows `1`:

docs/content/getting-started/scale-the-model.md

@@ -11,13 +11,13 @@ across two regions.
 ```mermaid
 graph LR
     subgraph fleet ["Fleet"]
-        IC1["region-a\nL4"]
-        IC2["region-b\nlarger GPU"]
+        IC1["cluster-a\nsmall GPU"]
+        IC2["cluster-b\nlarger GPU"]
     end
 
     subgraph ml ["ML team"]
         MD1["ModelDeployment\nqwen-demo"]
-        MD2["ModelDeployment\nqwen-west\nclusterSelector: region-b"]
+        MD2["ModelDeployment\nqwen-west\ntargets cluster-b"]
         MS["ModelService qwen\n/ml-team/qwen/v1/..."]
     end
 
@@ -27,10 +27,12 @@ graph LR
     MD2 --> MS
 ```
 
-## Deploy to a second region
+## Add a second deployment
 
-The new deployment uses a `clusterSelector` to pin its replica to the `us-west`
-cluster you added in the last step, and selects the larger GPU there:
+The new deployment targets a larger-GPU cluster you added in the last step. On
+EKS, GKE, and AKS it pins to a second region with a `clusterSelector`. On Nebius,
+which runs one region per project, the capacity selector alone routes it to the
+`H100` tier:
 
 {{< tabs >}}
 {{< tab "EKS" >}}
@@ -39,6 +41,12 @@ cluster you added in the last step, and selects the larger GPU there:
 {{< tab "GKE" >}}
 {{< manifests "getting-started/gke/model-deployment-west.yaml" >}}
 {{< /tab >}}
+{{< tab "AKS" >}}
+{{< manifests "getting-started/aks/model-deployment-west.yaml" >}}
+{{< /tab >}}
+{{< tab "Nebius" >}}
+{{< manifests "getting-started/nebius/model-deployment-scale.yaml" >}}
+{{< /tab >}}
 {{< /tabs >}}
 
 Wait until its replica is `Ready`, then check placement. You now have one replica

docs/content/getting-started/scale-the-platform.md

@@ -4,10 +4,10 @@ weight: 40
 description: Grow from one cluster to a multi-region fleet.
 ---
 
-You have one L4 cluster with a running model. In this guide, you'll add two
-larger-GPU clusters in different regions to grow the fleet available to the ML team.
+You have one small-GPU cluster with a running model. In this guide, you'll grow
+the fleet with larger-GPU capacity so the ML team has more to schedule against.
 
-Provisioning two more clusters takes about 10 to 15 minutes.
+Provisioning takes about 10 to 15 minutes.
 
 ## Register more clusters
 
@@ -45,9 +45,32 @@ is a few dollars for this tour. Clean up when you're done (see [Clean
 up]({{< ref "getting-started/clean-up.md" >}})).
 {{< /hint >}}
 {{< /tab >}}
+{{< tab "AKS" >}}
+Register two more clusters with a bigger hardware class: `A100` (`80 GB`) in
+`eastus` and `southcentralus`:
+
+{{< manifests "getting-started/aks/platform-scale.yaml" >}}
+
+{{< hint "note" >}}
+`Standard_NC24ads_A100_v4` runs ~$3.70/hr on demand. Two of them plus the `A10`
+from earlier is a few dollars for this tour. Clean up when you're done (see [Clean
+up]({{< ref "getting-started/clean-up.md" >}})).
+{{< /hint >}}
+{{< /tab >}}
+{{< tab "Nebius" >}}
+Nebius projects are bound to one region, so you grow the fleet by GPU tier rather
+than geography. Register a bigger `H100` (`80 GB`) cluster in the same region:
+
+{{< manifests "getting-started/nebius/platform-scale.yaml" >}}
+
+{{< hint "note" >}}
+The `H100` cluster costs more per hour than the `L40S` from earlier. Clean up
+when you're done (see [Clean up]({{< ref "getting-started/clean-up.md" >}})).
+{{< /hint >}}
+{{< /tab >}}
 {{< /tabs >}}
 
-Modelplane provisions both clusters in parallel:
+Modelplane provisions the new clusters in parallel:
 
 ```bash
 kubectl wait --for=condition=Ready ic --all --timeout=20m
@@ -62,4 +85,4 @@ its deployment changes in a way that no longer fits where it runs.
 
 ## Next step
 
-The fleet now spans three clusters across three regions. The ML team is next. [Scale the model]({{< ref "getting-started/scale-the-model.md" >}}) to serve it from two regions behind a single endpoint.
+The fleet has grown with larger-GPU capacity. The ML team is next. [Scale the model]({{< ref "getting-started/scale-the-model.md" >}}) to serve it across the fleet behind a single endpoint.

docs/manifests/getting-started/aks/model-deployment-west.yaml

@@ -0,0 +1,37 @@
+# A second ModelDeployment, pinned to a different region with clusterSelector.
+# It carries its own deployment name, so the ModelService can front it alongside
+# qwen-demo and serve both from one endpoint.
+apiVersion: modelplane.ai/v1alpha1
+kind: ModelDeployment
+metadata:
+  name: qwen-west
+  namespace: ml-team
+spec:
+  replicas: 1
+  # clusterSelector filters which InferenceClusters this deployment can land on.
+  # This pins the replica to the eastus cluster you added in "Scale the fleet".
+  template:
+    spec:
+      clusterSelector:
+        matchLabels:
+          modelplane.ai/region: eastus
+      engines:
+      - name: qwen
+        members:
+        - role: Standalone
+          nodeSelector:
+            devices:
+            - name: gpu
+              count: 1
+              selectors:
+          # A100 (81920Mi) qualifies; A10 (24564Mi) does not.
+              - cel: |
+                  device.capacity["gpu.nvidia.com"].memory.compareTo(quantity("35Gi")) >= 0
+          template:
+            spec:
+              containers:
+              - name: engine
+                image: vllm/vllm-openai:v0.23.0
+                args:
+                - --model=Qwen/Qwen2.5-0.5B-Instruct
+                - --dtype=half

docs/manifests/getting-started/aks/model-deployment.yaml

@@ -0,0 +1,30 @@
+apiVersion: modelplane.ai/v1alpha1
+kind: ModelDeployment
+metadata:
+  name: qwen-demo
+  namespace: ml-team
+spec:
+  replicas: 1
+  template:
+    spec:
+      engines:
+      - name: qwen
+        members:
+        - role: Standalone
+          nodeSelector:
+            devices:
+            - name: gpu
+              count: 1
+              selectors:
+          # Any L4 satisfies >= 20Gi. The selector matches against the capacity
+          # declared in the InferenceClass, not the pod's resource requests.
+              - cel: |
+                  device.capacity["gpu.nvidia.com"].memory.compareTo(quantity("20Gi")) >= 0
+          template:
+            spec:
+              containers:
+              - name: engine
+                image: vllm/vllm-openai:v0.23.0
+                args:
+                - --model=Qwen/Qwen2.5-0.5B-Instruct
+                - --dtype=half

docs/manifests/getting-started/aks/platform-scale.yaml

@@ -0,0 +1,62 @@
+apiVersion: modelplane.ai/v1alpha1
+kind: InferenceClass
+metadata:
+  name: a100-1x
+spec:
+  description: "AKS Standard_NC24ads_A100_v4, 1x NVIDIA A100 80GB"
+  provisioning:
+    provider: AKS
+    aks:
+      vmSize: Standard_NC24ads_A100_v4
+      diskSizeGb: 100
+      accelerator:
+        type: nvidia-a100
+        count: 1
+  devices:
+  - name: gpu
+    claim: DRA
+    driver: gpu.nvidia.com
+    deviceClassName: gpu.nvidia.com
+    count: 1
+    attributes:
+      architecture: { string: Ampere }
+      cudaComputeCapability: { version: "8.0.0" }
+    capacity:
+      memory: { value: "81920Mi" }
+---
+# verify: Standard_NC24ads_A100_v4 quota/availability in eastus and southcentralus.
+apiVersion: modelplane.ai/v1alpha1
+kind: InferenceCluster
+metadata:
+  name: aks-eastus
+  labels:
+    modelplane.ai/region: eastus
+spec:
+  cluster:
+    source: AKS
+    aks:
+      location: eastus
+  nodePools:
+  - name: gpua100
+    className: a100-1x
+    nodeCount: 1
+    minNodeCount: 1
+    maxNodeCount: 1
+---
+apiVersion: modelplane.ai/v1alpha1
+kind: InferenceCluster
+metadata:
+  name: aks-southcentralus
+  labels:
+    modelplane.ai/region: southcentralus
+spec:
+  cluster:
+    source: AKS
+    aks:
+      location: southcentralus
+  nodePools:
+  - name: gpua100
+    className: a100-1x
+    nodeCount: 1
+    minNodeCount: 1
+    maxNodeCount: 1

docs/manifests/getting-started/aks/platform.yaml

@@ -1,16 +1,18 @@
 apiVersion: modelplane.ai/v1alpha1
 kind: InferenceClass
 metadata:
-  name: a100-1x
+  name: a10-1x
 spec:
-  description: "AKS Standard_NC24ads_A100_v4, 1x NVIDIA A100 80GB"
+  description: "AKS Standard_NV36ads_A10_v5, 1x NVIDIA A10 24GB"
   provisioning:
     provider: AKS
     aks:
-      vmSize: Standard_NC24ads_A100_v4
-      diskSizeGb: 200
+      # verify: Azure has no L4 SKU; NVadsA10v5 is the small-GPU option (>=20Gi).
+      # Confirm it runs the DRA / GPU-operator path in your subscription.
+      vmSize: Standard_NV36ads_A10_v5
+      diskSizeGb: 100
       accelerator:
-        type: nvidia-a100
+        type: nvidia-a10
         count: 1
   devices:
   - name: gpu
@@ -20,9 +22,9 @@ spec:
     count: 1
     attributes:
       architecture: { string: Ampere }
-      cudaComputeCapability: { version: "8.0.0" }
+      cudaComputeCapability: { version: "8.6.0" }
     capacity:
-      memory: { value: "81920Mi" }
+      memory: { value: "24564Mi" }   # verify: A10 reported VRAM (~24GB)
 ---
 apiVersion: modelplane.ai/v1alpha1
 kind: InferenceCluster
@@ -36,8 +38,8 @@ spec:
     aks:
       location: westeurope
   nodePools:
-  - name: gpua100
-    className: a100-1x
+  - name: gpua10
+    className: a10-1x
     nodeCount: 1
     minNodeCount: 1
     maxNodeCount: 1