Menu

Nebius GPU prices: L40S, H100, H200, B200 per hour

We may earn commissions when you shop through the links below. Learn more

AI cloud renting NVIDIA GPU VMs per GPU-hour, with managed Kubernetes, Slurm and Token Factory inference.

GPU cloudInference hostingManaged AI hostingPrivate AI hosting

GPU-focused cloud · Headquarters: Amsterdam, Netherlands

From published sources, not hands-on tested

From
$1.55≈ €1.38 (estimated conversion from $1.55)/GPU-hrL40S 48GB · On-demandChecked docs.nebius.com
GPU families
4
B200 · H200 · H100 · +1 more · 4 models
Regions
10
4 in the EU
Billing
Per second
Hourly, spot, reserved, serverless, usage-based

2 of 9 prices older than 7 days · Prices checked 53 sourcesSources (53)How we check

In short

Nebius (Nasdaq: NBIS, headquartered in Amsterdam) runs a self-serve AI cloud that rents NVIDIA GPU virtual machines per GPU-hour, billed per second, including 1-GPU presets.

Read full summary

Since 1 Oct 2026 the on-demand list prices are $4.50 (HGX H100), $5.40 (H200), $8.50 (B200), $9.50 (B300) and $1.80 (RTX PRO 6000) per GPU-hour (compute pricing docs, checked 1 Oct 2026); before that they were $3.85, $4.50, $7.15 and $7.85 for the first four. Public regions are in Finland, France, Spain, the UK, the US and Israel. Egress traffic, public IPs and virtual networks are free, Managed Kubernetes and Soperator (Slurm) carry no platform fee, and Nebius Token Factory sells OpenAI-compatible inference for open models.

From

$1.55≈ €1.38 (estimated conversion from $1.55)/GPU-hrL40S 48GB · On-demandChecked docs.nebius.com
Visit Nebius Opens nebius.com in a new tab · Check the final plan, price and terms there before you order.Compare with alternatives to Nebius
Setup guides for Nebius
Billing increment
Per second
Egress
No egress feesSource
Persistent storage
Yes
Managed service
Yes

Pricing

2 of 9 prices older than 7 days · Prices checked How we check

Billing: per second, hourly, spot, reserved, serverless, usage-based. Lowest pay-as-you-go GPU price: $1.55≈ €1.38 (estimated conversion from $1.55) per GPU-hour (L40S 48GB, on-demand). Other price types start lower: $0.749≈ €0.669 (estimated conversion from $0.749) per GPU-hour (L40S 48GB, spot).

Prices in USD, as listed, excluding VAT.≈ Estimated conversion from the listed USD price at the ECB reference rate of 23 Oct 2026 (1 EUR = 1.1193 USD). Billing currency not verified: providers may bill in USD, and your bank's rate and fees differ. Prices exclude VAT.

GPU offers10

Nebius GPU offers: 10 offers grouped by GPU; 7 verified prices per GPU-hour, 3 not verified, excl. VAT. Availability: limited for every offer.

NVIDIA B200 SXM 180GB180 GB VRAM · 2 offers · from $8.50≈ €7.59 (estimated conversion from $8.50) per GPU-hour

  • On-demand20 vCPU · 224 GB RAM$8.50≈ €7.59 (estimated conversion from $8.50)/GPU-hrVaries (us-central1, me-west1) · per secondChecked docs.nebius.com
  • Spot20 vCPU · 224 GB RAMNot verifiedVaries (us-central1, me-west1) · per second

    Price changed Dynamic spot price since (no fixed price; not comparable with fixed prices)

    Checked · Staledocs.nebius.com

NVIDIA H200 SXM 141GB141 GB VRAM · 2 offers · from $5.40≈ €4.82 (estimated conversion from $5.40) per GPU-hour

  • On-demand16 vCPU · 200 GB RAM$5.40≈ €4.82 (estimated conversion from $5.40)/GPU-hrVaries (eu-north1, eu-west1, us-central1) · per secondChecked docs.nebius.com
  • Spot16 vCPU · 200 GB RAMNot verifiedVaries (eu-north1, eu-west1, us-central1) · per second

    Price changed Dynamic spot price since (no fixed price; not comparable with fixed prices)

    Checked · Staledocs.nebius.com

NVIDIA H100 SXM 80GB80 GB VRAM · 2 offers · from $4.50≈ €4.02 (estimated conversion from $4.50) per GPU-hour

  • On-demand16 vCPU · 200 GB RAM$4.50≈ €4.02 (estimated conversion from $4.50)/GPU-hrEU (eu-north1) · per secondChecked docs.nebius.com
  • Spot16 vCPU · 200 GB RAMNot verifiedEU (eu-north1) · per second

    Price changed Dynamic spot price since (no fixed price; not comparable with fixed prices)

    Checked · Staledocs.nebius.com
Show all 10 offers (from $0.749≈ €0.669 (estimated conversion from $0.749) per GPU-hour)

NVIDIA L40S 48GB48 GB VRAM · 4 offers · from $0.749≈ €0.669 (estimated conversion from $0.749) per GPU-hour

  • Spot8 vCPU · 32 GB RAM$0.749≈ €0.669 (estimated conversion from $0.749)/GPU-hrEU (eu-north1) · per secondChecked · Staledocs.nebius.com
  • Spot16 vCPU · 96 GB RAM$0.884≈ €0.789 (estimated conversion from $0.884)/GPU-hrEU (eu-north1) · per secondChecked · Staledocs.nebius.com
  • On-demand8 vCPU · 32 GB RAM$1.55≈ €1.38 (estimated conversion from $1.55)/GPU-hrEU (eu-north1) · per secondChecked docs.nebius.com
  • On-demand16 vCPU · 96 GB RAM$1.82≈ €1.62 (estimated conversion from $1.82)/GPU-hrEU (eu-north1) · per secondChecked docs.nebius.com

Prices are per single GPU-hour, excluding VAT. Multi-GPU nodes are divided by their GPU count; the node size is shown with the price type. Each row shows when the price was observed and links to the page it was read from.

Research notes for these offers10
NVIDIA B200 SXM 180GB · On-demand
NVIDIA HGX B200 (180 GB HBM3e; Nebius lists 8 TB/s), preset 1gpu-20vcpu-224gb; the 8gpu-160vcpu-1792gb preset is billed at the same per-GPU rate. Platforms gpu-b200-sxm (us-central1, Intel Xeon Platinum 8580) and gpu-b200-sxm-a (me-west1, Israel, Xeon Platinum 8570), both $8.50 per GPU-hour in the compute pricing docs from 1 Oct 2026 ($7.15 before). Default quota is 0 B200 GPUs in both regions, so on-demand use needs a quota increase or a reservation. Not offered in an EU region.
NVIDIA B200 SXM 180GB · Spot
Fixed preemptible VM rate through 7 Oct 2026, from the compute pricing docs ($3.95 per GPU-hour on both B200 platforms). From 8 Oct 2026 this becomes a dynamic spot price (updated up to every 15 minutes, capped one cent below on-demand); nebius.com/prices showed 'from $0.99' per GPU-hour for B200 on 26 Sep 2026. The live spot price was not verified (console only). As of 20 Oct 2026 nebius.com/prices lists the preemptible HGX B200 'from $0.99' per GPU-hour, with the footnote that spot prices are dynamic, can change as often as every 15 minutes, are shown only in the price calculator in the Nebius console, and that the highest preemptible rate is always one cent below the current on-demand rate; the compute pricing docs give the fixed $3.95 for the preemptible B200 NVLink only 'from October 1 until October 8, 2026' and 'From $0.99' from 8 Oct 2026. A 'from' figure is the floor of a dynamic spot price, not a comparable hourly price, so it is not this offer's price; the $3.95 price field keeps its 7 Oct 2026 date because that rate has ended.
NVIDIA H200 SXM 141GB · On-demand
NVIDIA HGX H200 (platform gpu-h200-sxm, 141 GB HBM3e, Intel Xeon Platinum 8468), preset 1gpu-16vcpu-200gb; the 8gpu-128vcpu-1600gb preset is billed at the same per-GPU rate. $5.40 per GPU-hour in the compute pricing docs from 1 Oct 2026 ($4.50 before), covering the preset's vCPUs and RAM. Default quotas for regular VMs without reservations: 32 GPUs in eu-north1, 8 in eu-west1, 0 in us-central1 (quota request needed). Also listed in the private eu-north2 region (existing deployments only).
NVIDIA H200 SXM 141GB · Spot
Fixed preemptible VM rate through 7 Oct 2026, from the compute pricing docs ($2.45 per GPU-hour). Interruptible, with no guaranteed availability. From 8 Oct 2026 this becomes a dynamic spot price (updated up to every 15 minutes, capped one cent below on-demand); nebius.com/prices showed 'from $0.79' per GPU-hour for H200 on 26 Sep 2026. The live spot price was not verified (console only). As of 20 Oct 2026 nebius.com/prices lists the preemptible HGX H200 'from $0.79' per GPU-hour, with the footnote that spot prices are dynamic, can change as often as every 15 minutes, are shown only in the price calculator in the Nebius console, and that the highest preemptible rate is always one cent below the current on-demand rate; the compute pricing docs give the fixed $2.45 for the preemptible H200 NVLink only 'from October 1 until October 8, 2026' and 'From $0.79' from 8 Oct 2026. A 'from' figure is the floor of a dynamic spot price, not a comparable hourly price, so it is not this offer's price; the $2.45 price field keeps its 7 Oct 2026 date because that rate has ended.
NVIDIA H100 SXM 80GB · On-demand
NVIDIA HGX H100 (platform gpu-h100-sxm, 80 GB HBM3, Intel Xeon Platinum 8468), preset 1gpu-16vcpu-200gb (RAM in GiB). The 8gpu-128vcpu-1600gb preset, used for InfiniBand GPU clusters, is billed at the same per-GPU rate. $4.50 per GPU-hour in the compute pricing docs from 1 Oct 2026 ($3.85 before); the GPU-hour price covers the preset's vCPUs and RAM. Billed per second while the VM runs; boot disk and other volumes are billed separately. eu-north1 only; default quota 32 GPUs for regular VMs without reservations. Docs warn of 'Not enough resources' errors when demand exceeds supply.
NVIDIA H100 SXM 80GB · Spot
Fixed preemptible VM rate through 7 Oct 2026, from the compute pricing docs ($2.15 per GPU-hour, preset 1gpu-16vcpu-200gb). Preemptible VMs can be stopped at any time (SIGTERM 60 seconds before stop); data on attached volumes is kept and dynamic public IPs are released. From 8 Oct 2026 preemptible GPU VMs move to a dynamic spot price per GPU type and region that can change every 15 minutes, capped one cent below on-demand; nebius.com/prices showed 'from $0.79' per GPU-hour for H100 on 26 Sep 2026. The live spot price is visible only in the web console and was not verified. Default preemptible VM quota: 8 VMs per region. As of 20 Oct 2026 nebius.com/prices lists the preemptible HGX H100 'from $0.79' per GPU-hour, with the footnote that spot prices are dynamic, can change as often as every 15 minutes, are shown only in the price calculator in the Nebius console, and that the highest preemptible rate is always one cent below the current on-demand rate; the compute pricing docs give the fixed $2.15 for the preemptible H100 NVLink only 'from October 1 until October 8, 2026' and 'From $0.79' from 8 Oct 2026. A 'from' figure is the floor of a dynamic spot price, not a comparable hourly price, so it is not this offer's price; the $2.15 price field keeps its 7 Oct 2026 date because that rate has ended.
NVIDIA L40S 48GB · Spot
Preemptible L40S Intel at the smallest preset 1gpu-8vcpu-32gb: GPU $0.65 + 8 x $0.006 per vCPU-hour + 32 x $0.0016 per GiB-hour = $0.7492 per hour from the compute pricing docs. nebius.com/prices shows 'from $0.74', which does not match this sum (it would round to $0.75); $0.74 matches a $0.005 vCPU rate, which the docs give for the AMD platform. See the L40S AMD preemptible note. Per the pricing page footnote, L40S preemptible instances are excluded from dynamic spot pricing and keep the flat rate on and after 8 Oct 2026.
NVIDIA L40S 48GB · Spot
Preemptible L40S AMD at the smallest preset 1gpu-16vcpu-96gb, from the docs' component prices: GPU $0.65 + 16 x $0.005 per vCPU-hour + 96 x $0.0016 per GiB-hour = $0.8836 per hour. nebius.com/prices shows 'from $0.90', which matches a $0.006 vCPU rate (0.65 + 16 x 0.006 + 96 x 0.0016 = $0.8996), the docs' Intel rate. Both L40S preemptible 'from' figures on the pricing page are therefore consistent with the two platforms' preemptible vCPU rates being swapped. The docs values are used: the docs list every component in USD and ILS, and each preemptible component is about half the on-demand rate for the same platform ($0.012 to $0.006 Intel, $0.01 to $0.005 AMD). Which page is wrong could not be verified. Excluded from dynamic spot pricing (flat rate also on and after 8 Oct 2026). The default L40S quota is shared with the Intel platform.
NVIDIA L40S 48GB · On-demand
NVIDIA L40S PCIe with Intel Ice Lake (platform gpu-l40s-a, Xeon Gold 6338). L40S is priced by component: GPU $1.35 per hour + $0.012 per vCPU-hour + $0.0032 per GiB-hour. This figure is the smallest preset, 1gpu-8vcpu-32gb: 1.35 + 8 x 0.012 + 32 x 0.0032 = $1.5484 per hour (nebius.com/prices: 'from $1.55'). Larger presets go up to 1gpu-40vcpu-160gb, all with 1 GPU. Unchanged in the 1 Oct 2026 column. eu-north1 only; default quota 2 L40S GPUs.
NVIDIA L40S 48GB · On-demand
NVIDIA L40S PCIe with AMD EPYC Genoa (platform gpu-l40s-d, EPYC 9654). Component pricing: GPU $1.35 per hour + $0.01 per vCPU-hour + $0.0032 per GiB-hour. This figure is the smallest preset, 1gpu-16vcpu-96gb: 1.35 + 16 x 0.01 + 96 x 0.0032 = $1.8172 per hour (nebius.com/prices: 'from $1.82'). Presets go up to 4gpu-192vcpu-1152gb (2- and 4-GPU presets are billed per component). eu-north1 only; the default L40S quota (2 GPUs) is shared with the Intel platform.

Compare these prices with other providers’: H100 prices comparedH200 prices comparedL40S prices compared.

Plans5

Serverless endpoint1 plan · price not verified

  • Nebius Token Factory (serverless per-token inference and dedicated endpoints)Varies · ManagedChecked · Stalenebius.comNot verified

GPU instance2 plans · from $1.80≈ €1.61 (estimated conversion from $1.80) per hour

  • NVIDIA RTX PRO 6000 Blackwell Server Edition, 1 GPU (1gpu-24vcpu-218gb)24 vCPU · 218 GB RAM · Varies · NVIDIA RTX PRO 6000 Blackwell Server Edition 96GBChecked docs.nebius.com$1.80≈ €1.61 (estimated conversion from $1.80)/hr
  • NVIDIA HGX B300, 1 GPU (gpu-b300-sxm, 1gpu-24vcpu-346gb)24 vCPU · 346 GB RAM · VariesChecked docs.nebius.com$9.50≈ €8.49 (estimated conversion from $9.50)/hr

CPU instance2 plans · price not verified

  • Non-GPU AMD EPYC Genoa VM, 4 vCPU / 16 GiB (cpu-d3, 4vcpu-16gb)4 vCPU · 16 GB RAM · GlobalChecked · Staledocs.nebius.comNot verified
  • Non-GPU Intel Ice Lake VM, 2 vCPU / 8 GiB (cpu-e2, 2vcpu-8gb)2 vCPU · 8 GB RAM · EUChecked · Staledocs.nebius.comNot verified
Research notes for these plans5
Nebius Token Factory (serverless per-token inference and dedicated endpoints)
OpenAI-compatible API (base URL https://api.tokenfactory.nebius.com/v1/) for open models; the product page features DeepSeek V4 Pro, Kimi K2.7 Code and MiniMax M3. Offers serverless inference and dedicated endpoints, which Nebius says give more control over performance, scaling, model configuration and regional placement, plus post-training (fine-tuning). Per-model token prices are shown only after login (tokenfactory.nebius.com/organization/prices opens a login screen) and were not verified. New sign-ups get a $1 trial credit valid for 30 days; a bank card is required. The Token Factory 'Sign up' links in Nebius blog-post footers point to studio.nebius.ai (the Nebius AI Studio address), while the product page links to tokenfactory.nebius.com.
NVIDIA RTX PRO 6000 Blackwell Server Edition, 1 GPU (1gpu-24vcpu-218gb)
Nebius documents the Server Edition (96 GB GDDR7, 1.6 TB/s, PCIe Gen5), so the plan is listed as the RTX PRO 6000 Blackwell Server Edition. It appears as a plan only, so its on-demand and preemptible rates are not in the GPU offer tables. On-demand $1.80 per GPU-hour, unchanged in the 1 Oct 2026 column; preemptible $0.95 per GPU-hour through 7 Oct 2026, then dynamic spot pricing from 8 Oct 2026 (pricing page: 'from $0.79'). An 8gpu-192vcpu-1744gb preset is also listed but is not compatible with InfiniBand GPU clusters. Regions: us-central1 (gpu-rtx6000, default quota 0), uk-south2 and eu-south1 (gpu-rtx6000-a, default quota 32 each). Billed per second.
NVIDIA HGX B300, 1 GPU (gpu-b300-sxm, 1gpu-24vcpu-346gb)
B300 (Blackwell Ultra, 288 GB HBM3e, 10 TB/s per Nebius docs) is not among the GPUs this site compares, so it is listed only as a plan. On-demand $9.50 per GPU-hour from 1 Oct 2026: the pricing page listed this new price on 26 Sep 2026 with that effective date ($7.85 until 30 Sep 2026); not re-checked since it took effect. Preemptible $4.30 per GPU-hour through 7 Oct 2026, then dynamic spot pricing from 8 Oct 2026 (pricing page: 'from $0.99'). Also sold as 8gpu-192vcpu-2768gb for InfiniBand GPU clusters (InfiniBand listed as 800 Gbps, 8x ConnectX-8). Regions: uk-south1, eu-west2 (France) and us-north1, default quota 32 GPUs each. The 8-GPU preset can also be created with 6 x 3.84 TB of local NVMe SSD (ephemeral, unencrypted, $0.065 per GiB-month); B300 8-GPU VMs are the only VMs that offer local SSD disks. Billed per second; no monthly price is published.
Non-GPU AMD EPYC Genoa VM, 4 vCPU / 16 GiB (cpu-d3, 4vcpu-16gb)
No price is shown from 1 Oct 2026: until 30 Sep 2026 the docs' component prices gave 4 x $0.012 per vCPU-hour + 16 x $0.0032 per GiB-hour = $0.0992 per hour (nebius.com/prices: AMD EPYC Genoa 4-64 vCPUs 'from $0.10'); from 1 Oct 2026 the pricing page lists 'from $0.13', and the new component prices were not re-checked. No monthly plan exists. Available in all public regions. The VM-types doc also lists a 2vcpu-8gb Genoa preset (computed $0.0496 per hour) that the pricing page's 4-64 vCPU range does not show.
Non-GPU Intel Ice Lake VM, 2 vCPU / 8 GiB (cpu-e2, 2vcpu-8gb)
No price is shown from 1 Oct 2026: until 30 Sep 2026 the docs' component prices gave 2 x $0.012 per vCPU-hour + 8 x $0.0032 per GiB-hour = $0.0496 per hour (nebius.com/prices: 'from $0.05'); from 1 Oct 2026 the pricing page lists 'from $0.06' for this family, and the new component prices were not re-checked. No monthly plan exists. eu-north1 only. Boot disk billed separately (Network SSD $0.071 per GiB-month). Default quota: 12 non-GPU VMs and 200 vCPUs per region.

Storage and data transfer

Network SSD disk (erasure coding)
$0.066≈ €0.059 (estimated conversion from $0.066) per GB-month
Block storage · PersistentPrice per GiB per 730 hours, billed per byte-second on the provisioned size whether or not the disk is attached. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
MoreEncrypted by default (AES-256). nebius.com/prices calls this 'Block volumes with erasure coding'. Published as USD 0.071 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Network SSD Non-replicated disk
$0.049≈ €0.044 (estimated conversion from $0.049) per GB-month
Block storage · PersistentPer GiB per 730 hours. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
MoreStored without data replication ('Block volumes without data replication' on nebius.com/prices); encryption must be enabled manually. The Compute SLA pays a 10% credit for each failed non-replicated placement group. Published as USD 0.053 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Network SSD IO M3 disk (3x mirroring)
$0.11≈ €0.098 (estimated conversion from $0.11) per GB-month
Block storage · PersistentPer GiB per 730 hours ('Block volumes with 3x mirroring' on nebius.com/prices). Encryption must be enabled manually. Published as USD 0.118 per GiB-month; stored per GB (1 GiB = 1.073741824 GB). Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
Disk snapshots
$0.066≈ €0.059 (estimated conversion from $0.066) per GB-month
Block storage · PersistentPer GiB per 730 hours. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
MorePoint-in-time copies of boot and secondary disks (fully available since 1 Jul 2026 per the Aether 3.6 release). The default snapshot quota is 0 in every public region except eu-south1 (32 snapshots, 1 TiB), so a quota request may be needed. Published as USD 0.071 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Local SSD disk
$0.061≈ €0.054 (estimated conversion from $0.061) per GB-month
Local NVMe · Not persistentPer GiB per 730 hours (compute pricing docs; 'Block volumes - local SSD disk' on nebius.com/prices). Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
MoreLocal SSD disks are NVMe drives physically attached to the host. They are ephemeral: data is lost whenever the VM is stopped or deleted, regular or preemptible, and when the host or disk fails (a reboot from inside the OS keeps it). Available only on the NVIDIA B300 platform (gpu-b300-sxm) with the 8-GPU preset 8gpu-192vcpu-2768gb in uk-south1, eu-west2 and us-north1, as a fixed 6 x 3.84 TB configuration that can only be enabled when the VM is created (local SSD docs). Local SSD disks are not encrypted (storage types docs). The compute pricing docs advise writing no more than the disk's capacity per day (TBW endurance). Published as USD 0.065 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Shared Filesystem SSD
$0.075≈ €0.067 (estimated conversion from $0.075) per GB-month
File storage · PersistentPer GiB per 730 hours, billed on provisioned size. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
MoreMountable on multiple VMs, Kubernetes pods and Serverless AI deployments. Encrypted at the infrastructure level. Published as USD 0.08 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
WEKA filesystem
$0.093≈ €0.083 (estimated conversion from $0.093) per GB-month
File storage · Persistent$0.1000 per GiB-month on nebius.com/prices. Pricing page: NVIDIA GPU Pricing | Nebius AI Cloud (opens in a new tab)
MoreEncryption is optional and must be chosen when the filesystem is created (XTS-AES-256). Published as USD 0.1 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Object Storage, Standard class
$0.014≈ €0.012 (estimated conversion from $0.014) per GB-month
Object storage · PersistentEgress to the internet or another Nebius region costs $0.015 per GiB (free within a region). Pricing page: Pricing in Object Storage (opens in a new tab)
MoreRequests: class A (PUT, POST, DELETE) $0.0034 and class B (GET, HEAD) $0.00028 per 1,000; bucket operations are free. Not available in uk-south1, uk-south2, eu-south1, eu-west2 or us-north1, where only the Enhanced Throughput class is offered (storage classes docs). Standard and Intelligent upload and download are limited to 20 GBps per tenant per region. Published as USD 0.0147 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Object Storage, Enhanced Throughput class
$0.102≈ €0.092 (estimated conversion from $0.102) per GB-month
Object storage · PersistentNo request charges and no egress charges. Pricing page: Pricing in Object Storage (opens in a new tab)
MoreAvailable in all regions, and the only Object Storage class in uk-south1, uk-south2, eu-south1, eu-west2 and us-north1. Upload and download are not throttled (storage classes docs). Published as USD 0.11 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Object Storage, Intelligent class (Warm tier)
$0.028≈ €0.025 (estimated conversion from $0.028) per GB-month
Object storage · PersistentNo request charges. Pricing page: Pricing in Object Storage (opens in a new tab)
MoreEgress is free up to three times the monthly average stored volume; above that Nebius may limit bandwidth or charge $0.05 per GiB. Objects not accessed for 30 days move to the Cold tier automatically; objects smaller than 180 KiB always stay Warm (storage classes docs). Not available in uk-south1, uk-south2, eu-south1, eu-west2 or us-north1. Published as USD 0.03 per GiB-month; stored per GB (1 GiB = 1.073741824 GB).
Object Storage, Intelligent class (Cold tier)
$0.0093≈ €0.0083 (estimated conversion from $0.0093) per GB-month
Object storage · PersistentSame request and egress rules as the Warm tier. Not available in uk-south1, uk-south2, eu-south1, eu-west2 or us-north1. Published as USD 0.01 per GiB-month; stored per GB (1 GiB = 1.073741824 GB). Pricing page: Pricing in Object Storage (opens in a new tab)
Egress (data out)
$0.00≈ €0.00 (estimated conversion from $0.00) per GB
nebius.com/prices lists 'Networking: Egress/Ingress traffic' and 'Networking: Public IP address' as Free, and the Virtual Networks pricing doc says the service is free of charge. Pricing page: NVIDIA GPU Pricing | Nebius AI Cloud (opens in a new tab)
MoreObject Storage egress is priced separately: Standard class $0.015 per GiB to the internet or another region, Enhanced free, Intelligent free up to 3x the monthly average stored volume and then $0.05 per GiB.

Before you compare

From Nebius’s record: its first stated uses and limitation, how its prices are set, and the providers compared below.

Suited to
  • Single-GPU to multi-node work on NVIDIA HGX H100, H200, B200 and B300, rented per GPU-hour with per-second billing and 1-GPU presets
  • Teams that want managed Kubernetes or Slurm (Soperator) with no platform fee, plus InfiniBand GPU clusters built from 8-GPU VMs
All 5 uses it suits
Main limitation
On-demand list prices rose on 1 Oct 2026 (compute pricing docs): H100 $3.85 to $4.50, H200 $4.50 to $5.40, B200 $7.15 to $8.50, B300 $7.85 to $9.50 per GPU-hour (RTX PRO 6000 and L40S unchanged); CPU 'from' prices also rose
All 10 limitations
Price basis
GPU prices per GPU-hour, billed per second; egress included; 11 storage options listed
All prices, storage and egress
Alternatives
DigitalOcean provider pageScaleway provider pageModal provider pageTogether AI provider page
Compare side by side with NebiusHow they differ

Who it fits

Based on Nebius’s public documentation, checked . We don’t score providers. How we build records

Suited to

  • Single-GPU to multi-node work on NVIDIA HGX H100, H200, B200 and B300, rented per GPU-hour with per-second billing and 1-GPU presets
  • Teams that want managed Kubernetes or Slurm (Soperator) with no platform fee, plus InfiniBand GPU clusters built from 8-GPU VMs
  • GPU workloads that must stay in EU regions (Finland, France, Spain) with a provider that states it acts as a GDPR data processor
  • Serving open models through Token Factory's OpenAI-compatible API or self-hosted vLLM, SGLang or Ollama endpoints
  • Data-heavy pipelines that benefit from free VM egress and ingress traffic

Good fit for

  • TrainingStrong fit8-GPU HGX H100, H200, B200 and B300 VMs can be joined into InfiniBand GPU clusters (InfiniBand listed as 400 Gbps with 8x ConnectX-7, or 800 Gbps with 8x ConnectX-8 for B300), with managed Slurm (Soperator) and Kubernetes at no platform fee and commitment discounts of up to 35%. On-demand capacity is subject to quotas and supply, and B200 defaults to a quota of 0.
  • Fine-tuningStrong fit1-GPU presets are sold at the per-GPU rate (H100 $4.50, H200 $5.40 per GPU-hour from 1 Oct 2026) with per-second billing. Docs include a Serverless AI fine-tuning tutorial with Axolotl, and Token Factory offers managed LoRA and full fine-tuning.
  • Private LLM hostingStrong fitCustomers get full VMs with sudo in EU regions (Finland, France, Spain). Nebius says it operates as a GDPR data processor with regional data residency, and lists SOC 2 Type II and ISO 27001/27701. Network SSD disks, shared filesystems and buckets are AES-256 encrypted by default (optional for Non-replicated and IO M3 disks; local SSD disks are not encrypted), KMS follows a customer-managed key model, and Token Factory offers dedicated endpoints.
  • Batch inferenceSuitableThrough 7 Oct 2026 preemptible VMs had fixed rates well below on-demand (H100 $2.15 preemptible vs $4.50 on-demand per GPU-hour from 1 Oct 2026), and Serverless AI jobs run to completion. From 8 Oct 2026 preemptible HGX and RTX PRO 6000 prices are dynamic (they can change every 15 minutes), and preemptible VMs can be stopped when capacity is reclaimed.
Show 3 more
  • Real-time inferenceSuitableServerless AI endpoints serve OpenAI-compatible models (a vLLM tutorial is provided), SGLang and Ollama have catalog apps, and Token Factory offers serverless and dedicated endpoints. Autoscaling or scale-to-zero for Serverless AI endpoints is not described, and on-demand GPU capacity is not guaranteed.
  • AI agentsSuitablen8n and Flowise catalog templates exist, CPU VMs start at $0.0496 per hour (2 vCPU / 8 GiB, eu-north1), and Hermes Agent supports Token Factory as a model provider. There is no managed agent-hosting product, and OpenClaw support is community-maintained.
  • RAG applicationsSuitableThe catalog includes Qdrant, Milvus, Weaviate and Redis, docs include a RAG tutorial (JupyterLab, Serverless AI and Managed PostgreSQL), and Object Storage offers three storage classes. Stacks are assembled from these parts; there is no managed RAG service.

Look elsewhere for

  • RenderingPartial fitL40S and RTX PRO 6000 Blackwell Server Edition GPUs are available, but Nebius positions them for AI inference, simulation and physical AI, and no rendering-specific guides or templates were found.

Not sure Nebius fits what you want to build?Help me choose

Strengths and limitations

Summarised from the research record for Nebius. We don’t score providers.

Strengths

  • Self-serve sign-up for individuals and companies, with the same per-GPU-hour list prices on nebius.com/prices and in the compute pricing docs
  • 1-GPU presets for H100, H200, B200, B300 and RTX PRO 6000 at the same per-GPU rate as 8-GPU presets; the GPU-hour price includes the preset's vCPUs and RAM
  • Per-second billing for running VMs; stopped VMs are not charged for GPUs, vCPUs or RAM (volumes still are)
  • Free egress and ingress traffic, free public IP addresses and free Virtual Networks; Managed Kubernetes and Managed Soperator have no platform fee
Show 4 more
  • EU public regions in Finland (eu-north1), France (eu-west1, eu-west2) and Spain (eu-south1); the Trust Center lists SOC 2 Type II (with HIPAA), SOC 3, ISO 27001, 27018, 27701 and 42001 among others
  • Applications catalog of 62 apps as of 26 Sep 2026, including Ollama, SGLang, n8n, Flowise, Jupyter, Open WebUI, LiteLLM and vector databases
  • Free technical support with published first-response targets (30 minutes for Critical, 1 business day for Low) and a 99.5% per-VM Compute SLA with service credits
  • Commitment discounts of up to 35% below on-demand for companies that reserve capacity for multiple months

Limitations

  • On-demand list prices rose on 1 Oct 2026 (compute pricing docs): H100 $3.85 to $4.50, H200 $4.50 to $5.40, B200 $7.15 to $8.50, B300 $7.85 to $9.50 per GPU-hour (RTX PRO 6000 and L40S unchanged); CPU 'from' prices also rose
  • Preemptible GPU VMs had fixed rates through 7 Oct 2026 (H100 $2.15, H200 $2.45, B200 $3.95 per GPU-hour); from 8 Oct 2026 Nebius charges a dynamic spot price that can change every 15 minutes, and the live price is shown only in the web console (L40S preemptible keeps its flat rate)
  • Capacity is not guaranteed: the docs describe 'Not enough resources' errors when demand exceeds supply, and default quotas are 0 GPUs for B200 (us-central1, me-west1), H200 in us-central1 and RTX PRO 6000 in us-central1
  • H100 is offered only in eu-north1 (Finland) and B200 only in us-central1 (US) and me-west1 (Israel); no EU region lists B200
Show 6 more
  • GB200 and GB300 NVL72 are contact-sales only, and commitment discounts require a company account and a sales agreement
  • Object Storage Standard egress is billed at $0.015 per GiB, although VM network egress is free
  • Token Factory per-model token prices are shown only after login and could not be verified
  • No bare-metal server product was found: GPUs are delivered as VMs (Nebius says GPUs and network interfaces are not virtualized)
  • Disk-snapshot and custom-image quotas default to 0 in every public region except eu-south1
  • Object Storage's cheaper Standard ($0.0147 per GiB-month) and Intelligent classes are not offered in uk-south1, uk-south2, eu-south1, eu-west2 or us-north1 (every B300 region and the non-US RTX PRO 6000 regions); only Enhanced Throughput ($0.11 per GiB-month, free egress) is available there

Infrastructure

Regions

10 regions across 7 countries; 4 are recorded as EU regions.

Europe (7)
eu-north1 (Finland) (EU); eu-north2 (Iceland, private region); eu-south1 (Madrid, Spain) (EU); eu-west1 (France) (EU); eu-west2 (France) (EU); uk-south1 (United Kingdom); uk-south2 (United Kingdom)
North America (2)
us-central1 (Kansas City, Missouri, US); us-north1 (Woodbury, Minnesota, US)
Asia (1)
me-west1 (Israel)

Source: Docs: Nebius AI Cloud regions (opens in a new tab)

Region notes10
eu-north1 (Finland)
Public region. GPU platforms: H200 NVLink, H100 NVLink, L40S Intel and L40S AMD; CPU platforms: AMD EPYC Genoa and Intel Ice Lake. All services in the regions table are available, including Serverless AI, Managed Soperator, Managed Kubernetes and Standalone Applications. Default GPU quotas for regular VMs without reservations: H200 32, H100 32, L40S 2. Has a component group on status.nebius.com.
eu-west1 (France)
Public region. GPU platform: H200 NVLink (default quota 8 GPUs); CPU: AMD EPYC Genoa. Managed Service for MLflow and Standalone Applications are not available here. Has a component group on status.nebius.com.
eu-west2 (France)
Public region. GPU platform: B300 NVLink (default quota 32 GPUs); CPU: AMD EPYC Genoa. Standalone Applications are not available here. Has a component group on status.nebius.com. Object Storage offers only the Enhanced Throughput class here (Standard and Intelligent are not available). Local SSD disks are available here on the B300 8-GPU preset.
eu-south1 (Madrid, Spain)
Public region. GPU platform: RTX PRO 6000 (gpu-rtx6000-a, default quota 32 GPUs); CPU: AMD EPYC Genoa. Serverless AI, Managed PostgreSQL, Managed MLflow and Standalone Applications are not available here. The only public region whose default quota includes disk snapshots and custom images. No separate component group on status.nebius.com as of 26 Sep 2026. Object Storage offers only the Enhanced Throughput class here (Standard and Intelligent are not available).
uk-south1 (United Kingdom)
Public region in Europe but outside the EU. GPU platform: B300 NVLink (default quota 32 GPUs); CPU: AMD EPYC Genoa. Standalone Applications are not available here. Has a component group on status.nebius.com. Object Storage offers only the Enhanced Throughput class here (Standard and Intelligent are not available). Local SSD disks are available here on the B300 8-GPU preset.
uk-south2 (United Kingdom)
Public region in Europe but outside the EU. GPU platform: RTX PRO 6000 (gpu-rtx6000-a, default quota 32 GPUs); CPU: AMD EPYC Genoa. Managed Soperator, PostgreSQL, MLflow, Standalone Applications and Nebius Tunnels are not available here. No separate component group on status.nebius.com as of 26 Sep 2026. Object Storage offers only the Enhanced Throughput class here (Standard and Intelligent are not available).
us-central1 (Kansas City, Missouri, US)
Public region. GPU platforms: B200 NVLink (gpu-b200-sxm), H200 NVLink and RTX PRO 6000 (gpu-rtx6000); CPU: AMD EPYC Genoa. Default quotas for regular VMs without reservations are 0 for B200, H200 and RTX PRO 6000, so a quota increase is needed. Has a component group on status.nebius.com.
us-north1 (Woodbury, Minnesota, US)
Public region. GPU platform: B300 NVLink (default quota 32 GPUs); CPU: AMD EPYC Genoa. Managed Soperator, Serverless AI, PostgreSQL, MLflow, Standalone Applications and Nebius Tunnels are not available here. No separate component group on status.nebius.com as of 26 Sep 2026. Object Storage offers only the Enhanced Throughput class here (Standard and Intelligent are not available). Local SSD disks are available here on the B300 8-GPU preset.
me-west1 (Israel)
Public region. GPU platform: B200 NVLink (gpu-b200-sxm-a, Intel Xeon Platinum 8570; default quota 0 GPUs); CPU: AMD EPYC Genoa. Standalone Applications are not available here. Companies from Israel are billed in ILS; all other customers in USD. Has a component group on status.nebius.com.
eu-north2 (Iceland, private region)
Private region, available only to users who already have deployments there (default quotas are 0). Lists the H200 NVLink platform. Iceland is an EEA member, not an EU member state. Has a component group on status.nebius.com.

Frameworks

How Nebius works with common AI serving, agent and training frameworks. Support levels, strongest first: an official template, documentation, a community guide, or manual setup on a general-purpose machine.

Hosting

Support for running the framework itself on Nebius.

Flowise
Official template
Catalog app with Flowise on VM and Flowise on Kubernetes (app free; compute billed). The docs also cover a Flowise Standalone Application. Docs: Flowise | Nebius applications (opens in a new tab)
n8n
Official template
Catalog app with one deployment method, n8n on VM (self-managed, manual scaling; app free, compute billed). Docs: n8n | Nebius applications (opens in a new tab)
MoreThe page also offers a 'Deploy with a solutions architect' option through sales (terms not stated).
Ollama hosting providers
Official template
Catalog app with two deployment methods, Ollama on VM and Ollama on Kubernetes (app free; compute billed). The official 'Host an LLM' quickstart also installs Ollama and Open WebUI on an H100 VM. Docs: Ollama | Nebius applications (opens in a new tab)
PyTorch
Official template
The Jupyter catalog app ships JupyterLab and JupyterHub with PyTorch and GPU support. PyTorch is also listed on the software compatibility page, and GPU boot images include NVIDIA drivers and CUDA. Docs: AI & ML applications catalog | Nebius (opens in a new tab)
SGLang
Official template
SGLang catalog app ('Start serving LLMs and VLMs in under 5 minutes with one-click deploy', free; compute billed). Also listed under supported inference runtimes. Docs: AI & ML applications catalog | Nebius (opens in a new tab)
vLLM hosting providers
Documented setup
Official tutorial deploys a Serverless AI endpoint from the vllm/vllm-openai image (OpenAI-compatible API, bearer-token auth) on an L40S preset. vLLM is listed under supported inference runtimes. Docs: Deploying a large language model and chatting with it by using Serverless AI endpoints (opens in a new tab)
MoreNo vLLM app is in the current catalog (there is no page at nebius.com/applications/vllm).
OpenClaw hosting providers
Community guide
No Nebius template or docs guide. Secondary source: opencolin/openclaw-nebius: OpenClaw + Nebius provider plugin, deployment UI and skill (opens in a new tab)
MoreA personal GitHub repository (opencolin/openclaw-nebius, 12 stars) provides deploy scripts, a deploy UI and a ClawHub Token Factory provider plugin. Nebius's blog has an OpenClaw security and hardening guide (5 Mar 2026) that recommends Token Factory as the model provider. The repository's deploy UI and scripts install OpenClaw on Nebius infrastructure (per its README as of 27 Sep 2026), so it is shown as Community guide under Hosting.
Dify
Manual setup
Dify is not in the 62-app catalog or the Token Factory integrations list, and no Nebius guide was found. It could be self-hosted with Docker on a VM (sudo access), but this was not tested. Docs: AI & ML applications catalog | Nebius (opens in a new tab)
LangGraph
Manual setup
Token Factory's integrations list includes LangChain but not LangGraph, and no Nebius LangGraph guide was found. LangGraph is a Python library that can run on any VM or container. Docs: Integrations overview | Nebius Token Factory documentation (opens in a new tab)
Text Generation Inference
Manual setup
TGI is not listed among supported inference runtimes (TensorRT, SGLang, vLLM) and has no catalog app. Docs: Software compatibility and supported runtimes in Nebius AI Cloud (opens in a new tab)
MoreIts container could run on a container VM or a Serverless AI endpoint with a custom image, but this was not tested.

Model API

Nebius serves the models the framework calls; the framework itself runs elsewhere.

Hermes Agent
Documented setup
Hermes Agent's own provider docs list Nebius Token Factory as a built-in provider (provider nebius-token-factory, NEBIUS_API_KEY). Secondary source: Hermes Agent docs: LLM and Model Providers (opens in a new tab)
MoreThis covers inference only: no Nebius guide or template exists for hosting the Hermes Agent process, which would be a manual install on a VM.

Nebius is not listed on the Hermes Agent hosting page. That page lists providers that host the framework with an official template, documentation or a community guide.

Operations

Managed service: yes. Each feature below states what the provider documents, with the source it was checked against. “Not verified” means no reliable source was found.

Features

Compute and access

Custom containers (Docker)
Yes
Container VMs deploy Nebius-provided or custom Docker images from public registries. Serverless AI takes any container image, and Container Registry is available in every region. Docs: Container virtual machines (opens in a new tab)
Root access
Yes
Full VMs: the official quickstart creates a cloud-init user with passwordless sudo. Docs: Getting started with Compute: Host an LLM on Nebius AI Cloud (opens in a new tab)
MoreThe 'root' and 'admin' usernames are reserved and cannot log in over SSH. Catalog app pages describe the VM option as 'Root access & custom setup'.
SSH access
Yes
SSH key-based login to VMs; SSH is also available to Serverless AI deployments. Docs: Getting started with Compute: Host an LLM on Nebius AI Cloud (opens in a new tab)
API access
Yes
Resources can be managed through the web console, CLI, Terraform provider and Go, Python and JavaScript SDKs; Token Factory has an OpenAI-compatible REST API. Docs: Preemptible virtual machines (opens in a new tab)
Jupyter notebooks
Yes
The Jupyter catalog app provides JupyterLab and JupyterHub with PyTorch and GPU support. There is also a JupyterLab Standalone Application and a JupyterLab quickstart for Serverless AI Devlabs. Docs: AI & ML applications catalog | Nebius (opens in a new tab)
Templates
Yes
Applications catalog with 62 apps as of 26 Sep 2026 (VM, Kubernetes or partner-provided deployment). 57 are listed as free and 5 as 'On request'; the underlying compute is always billed. Docs: AI & ML applications catalog | Nebius (opens in a new tab)
Multi-GPU instances
Yes
8-GPU presets for B300, B200, H200, H100 and RTX PRO 6000, and 2- or 4-GPU L40S AMD presets. Docs: Types of virtual machines and GPUs in Nebius AI Cloud (opens in a new tab)
MoreInfiniBand GPU clusters accept only the 8-GPU B300, B200, H200 and H100 presets; the docs list InfiniBand as 400 Gbps (8x ConnectX-7) for H100, H200 and B200 and 800 Gbps (8x ConnectX-8) for B300.
Bare metal
No
GPUs are delivered as VMs. Nebius claims 'bare-metal level performance' because its virtual instances do not virtualize GPUs or network interfaces, but no bare-metal server product was found. Docs: Compute: GPU and CPU infrastructure for AI on Nebius (opens in a new tab)
Kubernetes
Yes
Managed Service for Kubernetes is listed as free on nebius.com/prices and is available in every region. Pricing page: NVIDIA GPU Pricing | Nebius AI Cloud (opens in a new tab)
MoreWorker nodes are Compute VMs and count toward Compute quotas. The Managed Kubernetes pricing docs list only node charges (Compute VM prices) and no control-plane fee.
Show all 21 features

Management

Managed service
Yes
Managed Kubernetes, Managed Soperator (Slurm), Managed PostgreSQL, Managed MLflow, Serverless AI, Standalone Applications and Token Factory managed inference. Docs: Nebius AI Cloud regions (opens in a new tab)
Serverless
Yes
Serverless AI runs containers as Devlabs, jobs and endpoints without managing VMs, billed per second at Compute rates while running (stopped endpoints are not billed). Docs: About Serverless AI (opens in a new tab)
MoreAutomatic scale-to-zero is not described in the docs. Token Factory provides per-token serverless inference.

Storage and data

Persistent storage
Yes
Network SSD disks, shared filesystems and Object Storage persist independently of VMs; data on volumes survives preemption. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
Snapshots
Yes
Disk snapshots cost $0.071 per GiB-month (fully available since 1 Jul 2026). The default snapshot quota is 0 in every public region except eu-south1. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
Managed backups
Partial
Disk snapshots are positioned for backup and recovery, and Managed PostgreSQL has backups. No scheduled backup service for VMs or volumes was found. Docs: Nebius AI Cloud 'Aether 3.6': Operating production AI with more control, efficiency, and confidence (opens in a new tab)

Billing

Per-second billing
Yes
Billing unit is 1 second for GPUs, vCPUs and RAM of running VMs (pricing unit 1 hour). Volumes are billed per byte-second. Pricing page: Compute pricing in Nebius AI Cloud (opens in a new tab)
Spot pricing
Yes
Preemptible GPU VMs had fixed rates through 7 Oct 2026 and have a dynamic spot price per GPU type and region from 8 Oct 2026 (updated up to every 15 minutes, capped one cent below on-demand) with optional max-price policies. Docs: Spot pricing and pricing policies in Nebius AI Cloud (opens in a new tab)
MoreL40S preemptible stays at a flat rate.
Reserved pricing
Yes
Commitment discounts are 'up to 35% less than on-demand' per nebius.com/prices. Docs: Commitment discounts in Nebius AI Cloud (opens in a new tab)
MoreThey are generally prepaid, cover GPU, vCPU, RAM, disks and shared filesystems, are available only to company accounts through sales, and specific rates are not published.

Regions and compliance

EU region
Yes
Public EU regions: eu-north1 (Finland), eu-west1 and eu-west2 (France), eu-south1 (Madrid, Spain). Docs: Nebius AI Cloud regions (opens in a new tab)
SOC 2 report
Yes
The Trust Center lists SOC 2 Type II (with HIPAA) and SOC 3, plus ISO 22301, 27001, 27018, 27032, 27701, 27799 and 42001, NIS 2 and DORA alignment, and CSA STAR Level 1. Docs: Trust center | Nebius (opens in a new tab)
MoreThe reports' contents are not verified.
GDPR data processing agreement
Yes
The legal index lists a Data Processing Agreement and a sub-processor list. Docs: Nebius legal documents (opens in a new tab)
MoreThe Trust Center says Nebius operates under GDPR as an EU data processor and supports regional data residency, and the support docs cover data subject requests under GDPR and the EU Data Act. The DPA's contents are not verified.
Customer-managed keys (BYOK)
Partial
Key Management Service follows a customer-managed key (CMEK) model: customers create, rotate and destroy symmetric and asymmetric keys, including for envelope encryption. Docs: Nebius AI Cloud 'Aether 3.6': Operating production AI with more control, efficiency, and confidence (opens in a new tab)
MorePer the encryption docs, keys for disks, shared filesystems and buckets are issued and managed by the platform, and importing external key material is not documented.

SLA and support

Uptime SLA
99.5% uptime
Compute Cloud service level: 99.50% for a single VM (published 25 Sep 2024). Docs: Service Level for Compute Cloud (opens in a new tab)
MoreCredits are 10% of the service fee for the reporting period below 99.5%, 15% below 99.0% and 30% below 95%; unavailability means loss of external connectivity or the boot disk. Universal SLA terms (new version published 15 Sep 2026, effective 28 Sep 2026): claims within 14 days, credits only against future fees, planned node maintenance and emergency works excluded up to 8 hours per reporting period, and the SLA applies only to fee-paying use. Managed Kubernetes, Object Storage, VPC, PostgreSQL, MLflow and Standalone Applications have their own levels (not verified).
Support channels
2 listed
  • Support tickets in the web console support center (free technical support)
  • Email support@nebius.com (when the console or support center is unavailable)
Time to first non-automated response by ticket priority: Critical 30 minutes, High 60 minutes, Medium 4 hours, Low 1 business day (Netherlands calendar). Docs: Support tickets: sending questions, bug reports and feature requests (opens in a new tab)
MoreSupport is provided in English per the Technical Support Regulation. Incidents are posted on status.nebius.com (email, Atom and RSS subscriptions). No paid support plan is published.

Sources and verification

53 sources, accessed between 26 Sep 2026 and 20 Oct 2026. Official provider pages are listed first; the 2 secondary sources are labelled as such.

How we verify

Prices, regions, storage, features, framework support, SLA and support channels link to the page they were read from, and every price shows when it was checked. Fit labels, strengths and limitations summarise the same research record. The Nebius record was last checked on 26 Sep 2026. A check date older than 7 days is marked “Stale” with an amber dot, and one older than 30 days “Needs refresh”.

Last verified
26 Sep 2026, 19:30 UTC
Checked · Stale
Method
Source-by-source review by AI agents, independently re-checked
Confidence
High
Checked by
AI research agent; every value re-checked against its source by a separate AI agent

Verification notes

On 26 September 2026 we checked every price, preset and region in this listing against Nebius's live price list, documentation and status page. The price list and the docs disagree on the preemptible L40S prices; the listing keeps the docs values. On-demand prices rose on 1 October 2026 and preemptible GPU prices became dynamic on 8 October 2026; live capacity and Token Factory token prices (behind a login) were not verified. Later corrections are listed on the methodology page.

Commercial relationship

AIHostCompare has no active affiliate link with Nebius; the Visit provider link goes to its official website. Commissions never change ordering, alternatives, fit labels or conclusions. Read the affiliate disclosure.

Pricing

  1. 1.Compute pricing in Nebius AI Cloud (opens in a new tab) · Official pricing page · accessed
  2. 2.NVIDIA GPU Pricing | Nebius AI Cloud (opens in a new tab) · Official pricing page · accessed
  3. 3.Pricing in Managed Service for Kubernetes (opens in a new tab) · Official pricing page · accessed
  4. 4.Pricing in Managed Service for Soperator (opens in a new tab) · Official pricing page · accessed
  5. 5.Pricing in Object Storage (opens in a new tab) · Official pricing page · accessed
  6. 6.Pricing in Virtual Networks (opens in a new tab) · Official pricing page · accessed
  7. 7.Standalone Applications pricing in Nebius AI Cloud (opens in a new tab) · Official pricing page · accessed

Documentation

  1. 8."Not enough resources" error for virtual machines in Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  2. 9.About Nebius (opens in a new tab) · Official docs · accessed
  3. 10.About Serverless AI (opens in a new tab) · Official docs · accessed
Show all 53 sources

Documentation

  1. 11.AI & ML applications catalog | Nebius (opens in a new tab) · Official docs · accessed
  2. 12.Billing & Consumption | Nebius Token Factory documentation (opens in a new tab) · Official docs · accessed
  3. 13.Commitment discounts in Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  4. 14.Compute: GPU and CPU infrastructure for AI on Nebius (opens in a new tab) · Official docs · accessed
  5. 15.Container virtual machines (opens in a new tab) · Official docs · accessed
  6. 16.Deploying a large language model and chatting with it by using Serverless AI endpoints (opens in a new tab) · Official docs · accessed
  7. 17.Deploying and deleting applications for Managed Service for Kubernetes (opens in a new tab) · Official docs · accessed
  8. 18.Encryption in Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  9. 19.Flowise | Nebius applications (opens in a new tab) · Official docs · accessed
  10. 20.Getting started with Compute: Host an LLM on Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  11. 21.How to sign up for Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  12. 22.Integrations overview | Nebius Token Factory documentation (opens in a new tab) · Official docs · accessed
  13. 23.Introducing spot pricing for preemptible VMs (opens in a new tab) · Official docs · accessed
  14. 24.Key types and operations in Key Management Service (opens in a new tab) · Official docs · accessed
  15. 25.n8n | Nebius applications (opens in a new tab) · Official docs · accessed
  16. 26.Nebius AI Cloud 'Aether 3.6': Operating production AI with more control, efficiency, and confidence (opens in a new tab) · Official docs · accessed
  17. 27.Nebius AI Cloud documentation index (opens in a new tab) · Official docs · accessed
  18. 28.Nebius AI Cloud regions (opens in a new tab) · Official docs · accessed
  19. 29.Nebius legal documents (opens in a new tab) · Official docs · accessed
  20. 30.Nebius Partner Program (opens in a new tab) · Official docs · accessed
  21. 31.Nebius Token Factory (opens in a new tab) · Official docs · accessed
  22. 32.Nebius Token Factory documentation index (opens in a new tab) · Official docs · accessed
  23. 33.NVIDIA HGX H100 on Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  24. 34.Ollama | Nebius applications (opens in a new tab) · Official docs · accessed
  25. 35.OpenClaw security: architecture and hardening guide (opens in a new tab) · Official docs · accessed
  26. 36.Preemptible virtual machines (opens in a new tab) · Official docs · accessed
  27. 37.Pricing and quotas in Serverless AI (opens in a new tab) · Official docs · accessed
  28. 38.Quotas in Compute (opens in a new tab) · Official docs · accessed
  29. 39.Software compatibility and supported runtimes in Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  30. 40.Storage classes in Object Storage (opens in a new tab) · Official docs · accessed
  31. 41.Support tickets: sending questions, bug reports and feature requests (opens in a new tab) · Official docs · accessed
  32. 42.Switch to Token Factory | Nebius Token Factory documentation (opens in a new tab) · Official docs · accessed
  33. 43.Technical Support Regulation (opens in a new tab) · Official docs · accessed
  34. 44.Trust center | Nebius (opens in a new tab) · Official docs · accessed
  35. 45.Types of storage volumes in Compute (opens in a new tab) · Official docs · accessed
  36. 46.Types of virtual machines and GPUs in Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  37. 47.Using local SSD disks on Compute virtual machines (opens in a new tab) · Official docs · accessed

Terms & policies

  1. 48.Service Level Agreement | Nebius AI Cloud (opens in a new tab) · Official docs · accessed
  2. 49.Service Level for Compute Cloud (opens in a new tab) · Official docs · accessed
  3. 50.Spot pricing and pricing policies in Nebius AI Cloud (opens in a new tab) · Official docs · accessed

Status

  1. 51.Nebius Status (opens in a new tab) · Official status page · accessed

Secondary sources

  1. 52.Hermes Agent docs: LLM and Model Providers (opens in a new tab) · Secondary source · accessed
  2. 53.opencolin/openclaw-nebius: OpenClaw + Nebius provider plugin, deployment UI and skill (opens in a new tab) · Secondary source · accessed

Alternatives to Nebius

Providers that share a hosting category with Nebius, the most relevant first, spread so that every provider is listed on several provider pages. Not a ranking.

Also sharing categories: Together AI profile

Compare all 4 side by side

Related comparisons and guides

Categories that list Nebius