Meet Nebius: The Cloud Built for the AI Era

Intro

Every once in a while, a new cloud platform shows up that doesn’t just offer “more compute” . It rethinks what the cloud should look like in an AI-first world. That’s what caught my attention with Nebius, a European-born cloud designed from the ground up for high-performance, AI-centric workloads. One that just closed a $17 billion GPU capacity deal with Microsoft.

In this post, I’ll take you through a hands-on tour of Nebius: their history, GPU catalog, pricing model, stack offering, and CLI setup. We’ll explore what makes Nebius worth evaluating while staying fair and unbiased, focused only on the technical aspects. By the end, you’ll know if this AI Cloud deserves a spot in your strategy.

From Russia’s Google to AI Hyperscaler

1. The Yandex Era (1997-2021) -$31B Tech Empire

Arkady Volozh founded Yandex in 1997 , Russia’s version of Google before Google even entered the market. By 2011, Yandex went public on Nasdaq with a $1.3 billion IPO, becoming a tech powerhouse with search, maps, ride-hailing, and cloud services across Russia. At its peak in late 2021, Yandex N.V. hit a $31 billion valuation.

2. War, Sanctions, Forced Evolution and the $5.4B Exit

Then everything changed. In February 2022, Nasdaq halted trading after Russia’s Ukraine invasion triggered sanctions. In July 2024, Yandex N.V.(based in Amsterdam) sold its Russian assets to local investors for $5.4 billion, the largest corporate exit from Russia since the war began. What remained? A Finnish data center, $2 billion in cash, and thousands of battle-tested engineers who’d spent 20+ years building one of Europe’s largest tech ecosystems.

3. Rebirth as Nebius

In August 2024, the company rebranded as Nebius Group and resumed Nasdaq trading in October 2024 Nebius. The twist? Nebius is one of the only public AI infrastructure companies(debt free), most competitors are private or divisions of tech giants. The company raised $1 billion in September 2024 to fuel AI infrastructure expansion, and in December 2024, raised another $700 million from investors including Nvidia, which acquired a 0.5% stake. A year later, Nebius signs $17.4 billion AI infrastructure deal with Microsoft.

Regions

Nebius operates a rapidly expanding global footprint with availability zones across Europe and North America:

Data Centers (GPU Capacity)

🏷️Combined scale:

This is a combined 400+ MW secured capacity with $1B+ annual revenue potential at full utilization, expanding to 1 GW by end of 2026. For context, 1 million NVIDIA Blackwell GPUs require ~1-1.4 GW of power, Nebius is targeting 3x+ xAI’s Colossus 1 scale at full buildout.

Free Tier & Credits

Unlike hyperscalers, Nebius has no free trial. You must add a credit card deposit with a minimum of $25 to enable billing. However, startups can apply for their programs (from $5k to $150K in credits), though approval requires application review and may take time

Nebius Offering

1. Nebius AI Studio

Managed inference-as-a-service with pre-configured environments, serverless one-click model deployment, and integrations for popular frameworks. Built for teams who want to deploy models fast without infrastructure overhead.

2. Nebius Cloud

Full IaaS with GPU instances, managed Kubernetes, and CLI/API access. For teams seeking control over the full stack.

3. 💥New: Token Factory

Token Factory – is a brand new Serverless inference platform (like Groq, Fireworks, Together AI) with guaranteed uptime, zero-retention data flow, and usage-based pricing, no GPU management required.

⚙️ Core Services

Beyond the usual cloud fundamentals (storage, networking, IAM, monitoring), Nebius includes:

• VM Compute: GPU/CPU virtual machines with customizable presets
• Kubernetes: Managed Kubernetes (MK8s) clusters with GPU support
• HPC: Slurm-managed clusters via Soperator (Slurm on Kubernetes)
• Databases: Managed PostgreSQL clusters
• Inference as a Service: Run open-source AI Models (Llama, Qwen, DeepSeek…) at production speed
• Fine-tuning service: Fine-tune and distill leading open-source models into specialized, domain-expert systems
• ML Tools:
  • Managed MLflow clusters for experiment tracking,
  • Managed Ray clusters for distributed computing

⚡Applications

What sets Nebius apart as an AI cloud: A rich marketplace of one-click, pre-configured AI/ML deployments. no YAML wrestling, no dependency hell.

1. Standalone VMs Apps

Deploy complete AI stacks (inference servers, development environments, Databases) as ready-to-use virtual machines with GPU support baked in.

2. Kubernetes Applications

One click Install of AI frameworks & tools as Helm charts directly into your MK8s clusters, from model serving platforms to MLOps tooling.

🔋GPU Platforms & Presets

1. GPU Platforms

Nebius offers the following GPU platforms across regions:

2. Available Presets

🏷️Pricing & Value

Nebius uses simple hourly pricing across all GPU and CPU instances, with several services included for free.

Included Free

✅ Managed Kubernetes
✅ Managed Slurm (Soperator)
✅ Egress/Ingress traffic
✅ Public IP addresses

CLI Installation & Quick Start

1. Install the CLI

Ubuntu/Debian:

# Add Nebius repository
curl -fsSL https://storage.ai.nebius.cloud/nebius-public-keys/repository.gpg | sudo gpg --dearmor -o /usr/share/keyrings/nebius-archive-keyring.gpg

echo "deb [signed-by=/usr/share/keyrings/nebius-archive-keyring.gpg] https://storage.ai.nebius.cloud/nebius-repo stable main" | sudo tee /etc/apt/sources.list.d/nebius-repo.list

# Install
sudo apt update && sudo apt install nebius-cli

# Add Nebius repository
curl -fsSL https://storage.ai.nebius.cloud/nebius-public-keys/repository.gpg | sudo gpg --dearmor -o /usr/share/keyrings/nebius-archive-keyring.gpg

echo "deb [signed-by=/usr/share/keyrings/nebius-archive-keyring.gpg] https://storage.ai.nebius.cloud/nebius-repo stable main" | sudo tee /etc/apt/sources.list.d/nebius-repo.list

# Install
sudo apt update && sudo apt install nebius-cli

Other OS: See official installation guide for macOS, Windows, and other Linux distributions.

2. Auto completion

nebius completion bash > ~/.nebius/completion.bash.inc
echo 'if [ -f ~/.nebius/completion.bash.inc ]; then source ~/.nebius/completion.bash.inc; fi' >> ~/.bashrc
source ~/.bashrc

nebius completion bash > ~/.nebius/completion.bash.inc
echo 'if [ -f ~/.nebius/completion.bash.inc ]; then source ~/.nebius/completion.bash.inc; fi' >> ~/.bashrc
source ~/.bashrc

3. Configure Your Profile

# Interactive setup (recommended)
$ nebius profile create

profile name: my-profile
Set api endpoint: api.nebius.cloud
Set federation endpoint: auth.nebius.com

# Opens browser for authentication
✔ Profile "my-profile" configured and activated

# Interactive setup (recommended)
$ nebius profile create

profile name: my-profile
Set api endpoint: api.nebius.cloud
Set federation endpoint: auth.nebius.com

# Opens browser for authentication
✔ Profile "my-profile" configured and activated

Or configure manually:

export NB_PROFILE_NAME=<profile-name>
export NB_PROJECT_ID=<project-id>  # From console Project Settings

nebius profile create \
    --profile $NB_PROFILE_NAME --endpoint api.nebius.cloud \
    --federation-endpoint auth.nebius.com --parent-id $NB_PROJECT_ID

export NB_PROFILE_NAME=<profile-name>
export NB_PROJECT_ID=<project-id>  # From console Project Settings

nebius profile create \
    --profile $NB_PROFILE_NAME --endpoint api.nebius.cloud \
    --federation-endpoint auth.nebius.com --parent-id $NB_PROJECT_ID

4. Verify Setup

# Check your identity
nebius iam whoami
...
 user_profile:
  attributes:
    email:  
    family_name: hd
    given_name: Clouddude
    locale: ""
    name: Clouddude
    phone_number: ""
    picture: https://avatars.githubusercontent.com/u
    preferred_username: ""
    sub: "xxxxx"
  federation_info:
    federation_id: federation-e00github
    federation_user_account_id: "xxxxxx"
  id: useraccount-xxx
  tenants:
    - tenant_id: tenant-xx
      tenant_user_account_id: tenantuseraccount-xx

# List profiles
nebius profile list

# View config
cat ~/.nebius/config.yaml

# Check your identity
nebius iam whoami
...
 user_profile:
  attributes:
    email:  
    family_name: hd
    given_name: Clouddude
    locale: ""
    name: Clouddude
    phone_number: ""
    picture: https://avatars.githubusercontent.com/u
    preferred_username: ""
    sub: "xxxxx"
  federation_info:
    federation_id: federation-e00github
    federation_user_account_id: "xxxxxx"
  id: useraccount-xxx
  tenants:
    - tenant_id: tenant-xx
      tenant_user_account_id: tenantuseraccount-xx

# List profiles
nebius profile list

# View config
cat ~/.nebius/config.yaml

Network

Having worked with 6+ cloud providers (including AliCloud and Civo), networking always caught me off guard.
Default pool: Each project has default-network-pool with a region-specific CIDR (e.g., 10.x.0.0/16).

The key difference

• CIDR blocks are defined at the pool level, not the VPC
• Networks reference pools (no CIDR at network level)
• Subnets allocate their CIDRs from those pools—provided they don’t overlap and stay within the pool’s range

How it works:

# 1. Pool defines the CIDR range (e.g., 10.0.0.0/8)
nebius vpc pool create --cidr 10.0.0.0/16

# 2. Network references the pool (no CIDR at network level)
nebius vpc network create --name my-network --ipv4-private-pool-id <pool_id>

# 3. Subnets carve out non-overlapping blocks from the pool
nebius vpc subnet create --network-id <network_id> --cidr 10.0.1.0/24  # Must be within pool range

nebius vpc subnet create --network-id <network_id> --cidr 10.0.2.0/24  # Different block, no overlap

# 1. Pool defines the CIDR range (e.g., 10.0.0.0/8)
nebius vpc pool create --cidr 10.0.0.0/16

# 2. Network references the pool (no CIDR at network level)
nebius vpc network create --name my-network --ipv4-private-pool-id <pool_id>

# 3. Subnets carve out non-overlapping blocks from the pool
nebius vpc subnet create --network-id <network_id> --cidr 10.0.1.0/24  # Must be within pool range

nebius vpc subnet create --network-id <network_id> --cidr 10.0.2.0/24  # Different block, no overlap

Quick Kubernetes Cluster Deploy

# Get subnet ID
export NB_SUBNET_ID=$(nebius vpc subnet list --format json | jq -r '.items[0].metadata.id')

# Create cluster
export NB_CLUSTER_ID=$(nebius mk8s cluster create \
  --name quickstart-cluster \
  --control-plane-subnet-id $NB_SUBNET_ID \
  '{"spec": {"control_plane": {"endpoints": {"public_endpoint": {}}}}}' \
  --format json | jq -r '.metadata.id')

# Create node group (2 nodes, cpu-e2 preset)
nebius mk8s node-group create \
  --name quickstart-nodes \
  --parent-id $NB_CLUSTER_ID \
  --fixed-node-count 2 \
  --template-resources-platform "cpu-e2" \
  --template-resources-preset "2vcpu-8gb" \
  --template-boot-disk-type network_ssd \
  --template-boot-disk-size-bytes 137438953472

# Get kubeconfig
nebius mk8s cluster get-credentials --id $NB_CLUSTER_ID --external
kubectl cluster-info

# Get subnet ID
export NB_SUBNET_ID=$(nebius vpc subnet list --format json | jq -r '.items[0].metadata.id')

# Create cluster
export NB_CLUSTER_ID=$(nebius mk8s cluster create \
  --name quickstart-cluster \
  --control-plane-subnet-id $NB_SUBNET_ID \
  '{"spec": {"control_plane": {"endpoints": {"public_endpoint": {}}}}}' \
  --format json | jq -r '.metadata.id')

# Create node group (2 nodes, cpu-e2 preset)
nebius mk8s node-group create \
  --name quickstart-nodes \
  --parent-id $NB_CLUSTER_ID \
  --fixed-node-count 2 \
  --template-resources-platform "cpu-e2" \
  --template-resources-preset "2vcpu-8gb" \
  --template-boot-disk-type network_ssd \
  --template-boot-disk-size-bytes 137438953472

# Get kubeconfig
nebius mk8s cluster get-credentials --id $NB_CLUSTER_ID --external
kubectl cluster-info

⚡Custom IP management for Kubernetes:
You must ensure your pool has enough CIDR blocks for all cluster requirements (nodes, services, pods), or you’ll hit “Unable to allocate CIDR from the pool” errors. Check available blocks with:

nebius vpc subnet get --id <subnet_ID><br>
# Look for: spec.ipv4_private_pools.pools.cidrs

nebius vpc subnet get --id <subnet_ID><br>
# Look for: spec.ipv4_private_pools.pools.cidrs

Coming Up Next

Now that we’ve explored what makes Nebius stand out, from its GPU-focused architecture to its rich developer experience .It’s time to take things one step further.

In Part 2, we’ll show how to deploy and configure GPU resources using Terraform. from authentication and instance provisioning to network setup. This step brings us closer to integrating Nebius into the official vLLM Production Stack, making open-source inference more portable and cloud-agnostic.

Stay tuned  for Part 2⚡

Run AI Your Way — In Your Cloud

Want full control over your AI backend? The CloudThrill VLLM Private Inference POC is still open — but not forever.

📢 Secure your spot (only a few left), 𝗔𝗽𝗽𝗹𝘆 𝗻𝗼𝘄!

---

Run AI assistants, RAG, or internal models on an AI backend 𝗽𝗿𝗶𝘃𝗮𝘁𝗲𝗹𝘆 𝗶𝗻 𝘆𝗼𝘂𝗿 𝗰𝗹𝗼𝘂𝗱 –
✅ No external APIs
✅ No vendor lock-in
✅ Total data control

Claim YOur FREE VLLM POC

𝗬𝗼𝘂𝗿 𝗶𝗻𝗳𝗿𝗮. 𝗬𝗼𝘂𝗿 𝗺𝗼𝗱𝗲𝗹𝘀. 𝗬𝗼𝘂𝗿 𝗿𝘂𝗹𝗲𝘀…