# Nebius AI Cloud

Canonical: https://inetgeek.com/gpu/nebius/

GPU cloud with NVIDIA L40S to B300 instances billed per second, commitment discounts, spot-priced preemptible VMs and free network egress.

## Who it suits

Per-second GPU rental at published on-demand rates, from L40S at $1.55 to B300 at $7.85 per GPU-hour (H100 $3.85 until October 1, 2026), with free egress, nine public regions in Europe, Israel and the US, and SOC 2 Type II and ISO 27001 in its Trust Center.

Consider something else if You want today's rates to hold or a free trial: H100 rises to $4.50 and B200 to $8.50 on October 1, 2026, commitment discounts of up to 35% go through sales, there is no free offer, and the first payment is at least $25.

## iScore

56 / 100 (80% interval 50–62), rank 5 of 13 in gpu — interval overlaps the next rank, so this is an ordering, not a finding.

inetGeek's own reading, computed from sourced facts against category peers by disclosed rules. Not a provider claim. Intervals are 80%; a provider with few documented facts is pulled toward the category middle rather than scored on what little is known.
Method: https://inetgeek.com/iscore/. Full output with every criterion's basis: https://inetgeek.com/scores.json. Every gpu peer ranked: https://inetgeek.com/gpu/nebius/alternatives/

- Price: 43 (33–53), weight 50%, 3 of 10 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: budget_input_price, flagship_input_price, flagship_output_price, free_card_required, free_offer_type, free_tier, storage_price
- Capacity and limits: 54 (41–66), weight 24%, 2 of 3 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: context_window
- Capabilities: 100 (94–100), weight 3%, 0 of 1 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: prompt_caching
- Trust and compliance: 74 (68–79), weight 13%, 9 of 12 criteria scored; scored for peers, not here — undocumented or stated without a number or a yes/limited/no: dedicated_infrastructure, pci_dss, support_included
- Evidence: 89 (89–89), weight 11%, 4 of 4 criteria scored

## What the documentation says

### Pricing

- Entry paid plan: No plan tiers: usage is paid as you go in USD, and the first payment has a $25 minimum. ([source](https://nebius.com/prices), read 2026-09-27)
- Pricing model: On-demand rates per GPU-hour, combinable with commitment discounts; preemptible GPU instances are charged at dynamic spot prices that can change as often as every 15 minutes. Token-priced inference is priced separately under Nebius Token Factory. ([source](https://nebius.com/prices), read 2026-09-27)
- Egress over allowance: Free. Network egress and ingress traffic is not charged. Egress from Standard-class object storage is $0.015 per GiB; Enhanced-class object storage egress is free. ([source](https://nebius.com/prices), read 2026-09-27)

### Infrastructure

- Regions: Nine public regions: Finland, Madrid (Spain), two in France, Israel, Kansas City and Woodbury (US), and two in the United Kingdom. A tenth region, eu-north2 in Iceland, is open only to users already deployed there. GPU models differ by region. ([source](https://docs.nebius.com/overview/regions), read 2026-09-27)
- Uptime SLA: 99.50% for Compute Cloud, with service credits of 10%, 15% or 30% of the service fee when a single VM's uptime falls below 99.50%, 99.00% or 95.00% in a reporting period. ([source](https://docs.nebius.com/legal/sla-levels/compute), read 2026-09-27)

### Storage

- Storage price: $0.08 per GiB per month for the Nebius shared filesystem; block volumes run from $0.053 without replication to $0.118 with 3x mirroring, and Standard object storage is $0.0147 per GiB per month. ([source](https://nebius.com/prices), read 2026-09-27)

### GPU

- H100 SXM, per GPU-hour: $3.85 per GPU-hour on demand for NVIDIA HGX H100, listed with 16 vCPUs and 200 GB RAM per GPU. The pricing page shows a new rate of $4.50 effective October 1, 2026. H200 is $4.50, rising to $5.40 on the same date. ([source](https://nebius.com/prices), read 2026-09-27)
- Cheapest GPU, per hour: From $1.55 per GPU-hour on demand for NVIDIA L40S with Intel CPU (8-40 vCPUs, 32-160 GB RAM), unchanged after October 1, 2026. RTX PRO 6000 is $1.80. ([source](https://nebius.com/prices), read 2026-09-27)
- Largest GPU offered: 288GB HBM3e per GPU on NVIDIA B300 (Blackwell Ultra, SXM); B200 has 180GB and H200 141GB. ([source](https://docs.nebius.com/compute/virtual-machines/types), read 2026-09-27)
- Billing granularity: Per second for GPUs on running VMs, with prices shown per hour; computing resources of stopped VMs are not charged, but their volumes are. ([source](https://docs.nebius.com/compute/resources/pricing/), read 2026-09-27)
- Reserved pricing: Up to 35% below on-demand rates for reserving large-scale clusters for multiple months. Commitments are arranged with sales and require signing up as a company; no per-GPU commitment prices are published. ([source](https://nebius.com/prices), read 2026-09-27)

### Compliance

- SOC 2: SOC 2 Type II is listed in the Trust Center, with the badge reading "With HIPAA"; a SOC 3 summary of the SOC 2 Type II audit is published for public consumption. ([source](https://nebius.com/trust-center), read 2026-09-27)
- ISO 27001: ISO 27001 is listed in the Trust Center, alongside ISO 27701, 27018, 22301, 27799 and ISO 42001. ([source](https://nebius.com/trust-center), read 2026-09-27)
- HIPAA: PHI/ePHI use requires a signed BAA. Under it, Object Storage is approved for persistent ePHI storage, Compute instances and Managed Kubernetes for transient processing only; other AI Cloud services are not eligible by default. ([source](https://docs.nebius.com/legal/hipaa), read 2026-09-27)
- GDPR / data residency: Operates as an EU data processor under GDPR and states that customer data stays within the geographic region the customer selects for storage and processing. ([source](https://nebius.com/trust-center), read 2026-09-27)
- Encryption at rest: AES-256 by default across public storage services, with keys managed through Nebius KMS. Network SSD disks are encrypted by default; Non-replicated and IO M3 disks need encryption enabled manually, and WEKA filesystems must opt in at creation. ([source](https://docs.nebius.com/security/encryption), read 2026-09-27)
- Customer-managed keys: Key Management Service, in preview, lets customers issue and store their own symmetric and asymmetric keys to encrypt and decrypt data, sign hashes or apply envelope encryption in their applications. ([source](https://docs.nebius.com/kms/index), read 2026-09-27)
- Role-based access control: Four general roles (auditor, viewer, editor, admin) tied to default groups, plus service roles for Compute, Object Storage, Audit Logs and SecretStash; custom groups get roles through access permits scoped to a tenant, project or single resource. ([source](https://docs.nebius.com/iam/authorization/roles), read 2026-09-27)
- Private networking / VPC: Virtual Networks: resources in the same network connect over IP and are isolated from resources outside it; each project is given default networking resources on creation. ([source](https://docs.nebius.com/vpc/overview), read 2026-09-27)