Nebius pricing: GPU rental rates
Updated October 2026 · updated 2026-10-06
The lowest published Nebius rate is $1.35/GPU-hour for the L40S (on-demand; No reservation; Not reported, updated 2026-10-06). Nebius AI Cloud is a full-stack AI cloud with self-service GPU VMs, managed Kubernetes, Slurm and Soperator workflows, shared filesystems and object storage. It fits teams that want published GPU pricing, preemptible options, serious datacenter GPUs and a startup-credit path. Compute is billed with second-level granularity against hourly pricing units, and stopped VMs stop compute charges while storage continues separately.
Compare the full daily GPU market
GPU prices compared with the market
| GPU | Provider lowest firm rate | Market median | Difference |
|---|---|---|---|
| H100 | $4.50 | $3.39 | 32.6% |
| H200 | $5.40 | $4.76 | 13.5% |
| B200 | $8.50 | $6.84 | 24.3% |
| B300 | $9.50 | $9.24 | 2.9% |
| L40S | $1.35 | $1.57 | -14.1% |
| RTX PRO 6000 | $1.80 | $2.17 | -16.9% |
The market median gives each provider equal weight. GPU variants, node size, included CPU and memory can differ. A lower rate does not establish equivalent performance.
Live Nebius prices (2026-10-06)
Provider, memory & availability
| GPU / provider | Billing | $/GPU-hour | GPUs / total per hour | Availability / details |
|---|---|---|---|---|
| L40SNebius48 GB VRAM | on-demand | $1.35per GPU-hour+0.0% over 30 days | Node minimum and total not reported | Not reportedView provider2026-10-06Details
|
| RTX PRO 6000 (unspecified)Nebius96 GB VRAM | on-demand | $1.80per GPU-hour+0.0% over 30 days | Node minimum and total not reported | Not reportedView provider2026-10-06Details
|
| H100 (unspecified)Nebius80 GB VRAM | on-demand | $4.50per GPU-hour+16.9% over 30 days | Node minimum and total not reported | Not reportedView provider2026-10-06Details
|
| H200 (SXM)Nebius141 GB VRAM | on-demand | $5.40per GPU-hour+20.0% over 30 days | Node minimum and total not reported | Not reportedView provider2026-10-06Details
|
| B200Nebius180 GB VRAM | on-demand | $8.50per GPU-hour+18.9% over 30 days | Node minimum and total not reported | Not reportedView provider2026-10-06Details
|
| B300Nebius288 GB VRAM | on-demand | $9.50per GPU-hour+21.0% over 30 days | Node minimum and total not reported | Not reportedView provider2026-10-06Details
|
Try another provider or reduce the minimum memory.
Tables rank by price, lowest first. Commercial relationships never affect ordering. How we collect and rank prices.
Monthly view assumes 730 hours of continuous use. Daily view assumes 24 hours.
Cheapest listed per-GPU-hour rates verified at the source. Before storage, egress and networking.
Nebius access and company facts
| Check | Published information | Updated |
|---|---|---|
| Smallest recorded configuration | GPU count not reported for collected rates | 2026-10-06 |
| Published regions in this comparison | No per-configuration region recorded | 2026-10-06 |
| Minimum account funding | not published | 2026-07-07 |
| Headquarters | Amsterdam, Netherlands | 2026-09-27 |
| Data location terms | multi-region | 2026-07-07 |
A company fact or certification is not a guarantee for every product or region. Unrecorded fields are left unknown. The smallest recorded node does not establish immediate stock or account approval.
Strengths and trade-offs
Computed from the verified fees and platform matrix (cells updated 2026-07-07 at Nebius's own documentation; verified items link to their source). Full matrices: fees · enterprise.
Pricing model
Nebius publishes detailed Compute pricing for virtual machines with GPUs, vCPUs and RAM. The docs state that running VM compute is charged with a one-second billing unit and hourly pricing unit, and that charges for smaller units are calculated proportionally. Preemptible GPU pricing is published alongside on-demand pricing for supported platforms, while long-term commitment discounts are described on the public prices page.
Capacity tiers
Nebius offers regular and preemptible GPU VMs, self-service access to multi-GPU systems, managed Kubernetes, Slurm and Soperator. Its self-service page says customers can access NVIDIA Hopper, Blackwell and Blackwell Ultra GPUs without talking to sales up to the published self-service limit, while larger committed clusters can use commitment discounts or sales-led arrangements.
GPU lineup
Nebius documents NVIDIA B300, B200, H200, H100, RTX PRO 6000 and L40S platforms in Compute pricing and VM type docs. The VM type docs describe Blackwell Ultra B300, Blackwell B200, Hopper H200 and H100 SXM systems with InfiniBand connectivity. Region availability varies by platform, and the pricing docs call out specific regions for several GPUs.
Storage, egress and hidden costs
Nebius separates compute from storage. The Compute pricing docs state that stopped VM compute is not charged, but storage volumes still are. Public pricing covers shared filesystem, WEKA filesystem, object storage and block volume options. Object Storage egress varies by storage class: Standard egress to the internet or another Nebius region is charged, while Enhanced Throughput and same-region egress are listed as free in the object storage pricing docs.
Reliability and support posture
Nebius publishes a Service Level Agreement and separate service-level pages for individual services such as Object Storage and Standalone Applications. The SLA says availability commitments apply to specific paid services subject to quotas, limits and other restrictions. Support docs say Nebius AI Cloud provides free technical support through support tickets for bugs, quota requests and service questions.
Free credits and referral programs
Nebius publishes two startup-credit paths. Nebius for Startups lists introductory credits and compute savings, while the AI Lift announcement with NVIDIA says eligible NVIDIA Inception members can apply for larger Nebius cloud credits plus technical expertise and startup benefits. A general public referral program was not found in the official public pages reviewed.
Who should pick Nebius
Pick Nebius if you want published GPU prices, preemptible options, strong NVIDIA datacenter hardware and a more complete AI cloud stack than a bare VM marketplace. It is a strong fit for teams that need managed Kubernetes, Slurm-style workflows, shared filesystems or startup-credit support. For storage-heavy workloads, model object-storage egress and persistent volume costs before comparing only GPU rates.
Similar providers
RunPod
Secure Cloud (datacenter, reliable) and Community Cloud (peer hosts, cheaper). Per-second billing, serverless option.
from $0.25/hr secure cloud
6 GPU families in common
DataCrunch
European GPU cloud, per-minute billing, spot options.
from $0.198/hr on-demand
6 GPU families in common
Ranked by overlap in the GPU models both providers list today; not an endorsement and never affected by commercial relationships.
FAQ
Nebius Compute docs state that running VM compute has a one-second billing unit and hourly pricing unit.
Yes. Nebius publishes preemptible pricing for supported GPU VM platforms.
It depends on the storage class and destination. Standard Object Storage egress to the internet or another Nebius region is charged, while some same-region or enhanced-throughput paths are listed as free.
Yes. Nebius publishes startup-credit programs, including Nebius for Startups and AI Lift for eligible NVIDIA Inception members.
Profile updated 2026-07-06. Sources: nebius.com · nebius.com · nebius.com · docs.nebius.com · docs.nebius.com · docs.nebius.com · docs.nebius.com · docs.nebius.com · nebius.com · nebius.com