Nebius is an AI-focused cloud backed by former Yandex engineers, with GPU infrastructure across Europe and the US. Thunder Compute focuses on low-cost, persistent GPU instances for individual developers and small teams.
This comparison covers GPU pricing, infrastructure, and developer experience for both providers.
Key takeaways:
- Nebius H100 NVLink on-demand: $3.85/hr. Preemptible: $2.15/hr.
- Thunder Compute H100 PCIe: $2.19/hr with per-minute billing.
- Nebius operates EU data centers with GDPR compliance. Thunder Compute operates in North America.
- Thunder Compute includes 100GB per GPU, one-click IDE extensions, and no egress fees.
- Nebius offers reserved capacity discounts up to 35% for multi-month commitments.
What Nebius Does
Nebius is the successor to Yandex N.V., rebuilt as an independent AI cloud company headquartered in Amsterdam, with engineering teams in Finland, Serbia, and Israel. The company received a $2B NVIDIA investment in March 2026 and operates H100, H200, and B200 deployments across EU and US regions.
| GPU | vCPUs | RAM (GB) | Preemptible/hr | On-demand/hr |
|---|---|---|---|---|
| NVIDIA HGX B200 | 20 | 224 | $3.95 | $7.15 |
| NVIDIA HGX H200 | 16 | 200 | $2.45 | $4.50 |
| NVIDIA HGX H100 | 16 | 200 | $2.15 | $3.85 |
| NVIDIA RTX PRO 6000 | 24 | 218 | $0.95 | $1.80 |
| NVIDIA L40S (Intel CPU) | 16–192 | 96–1,152 | $0.90 | $1.82 |
| NVIDIA L40S (AMD CPU) | 8–40 | 32–160 | $0.74 | $1.55 |
| Source: Official Nebius pricing. Last update: July 14, 2026. | ||||
Nebius bills per second. Reserved capacity offers up to 35% off on-demand rates for multi-month commitments. Network SSD storage runs approximately $0.071/GiB/month.
What Thunder Compute Does
Thunder Compute takes a fundamentally different approach from enterprise-focused providers like Nebius. It's built for individual developers, researchers, and small teams who need powerful GPU computing without the complexity or cost of traditional cloud services.
Thunder Compute provides on-demand GPU instances with a focus on low prices, persistent storage, and one-click IDE integration. A100 80GB starts at $1.09/hr and H100 PCIe at $2.19/hr, both billed per minute with no egress fees.
Thunder Compute provides on-demand GPU instances with 1-8 GPUs per machine, featuring NVIDIA options like A6000, L40, L40S, A100s, and H100. Cost optimization results from a software-driven approach that maximizes hardware performance, which passes savings directly to users.
Every instance includes 100GB per GPU at no extra charge. Your home directory and installed packages survive restarts automatically. Snapshots cost $0.05/GB/month and let you save and restore your full environment, including when switching GPU configurations.
Pricing Comparison
| GPU | Thunder Compute | Nebius (on-demand) | Nebius (preemptible) |
|---|---|---|---|
| H100 80GB | $2.19/hr (PCIe) | $3.85/hr (NVLink) | $2.15/hr |
| A100 80GB | $1.09/hr | Not listed | Not listed |
| L40S | $0.791 | $1.35/hr | $0.74/hr |
| RTX A6000 | $0.35/hr | Not listed | Not listed |
| 1Thunder Compute only provides NVIDIA L40 GPUs. Last update: July 14, 2026. | |||
Nebius preemptible H100 at $2.15/hr is slightly cheaper than Thunder Compute H100 PCIe at $2.19/hr. Preemptible instances can be interrupted, making them unsuitable for uninterrupted training runs. Thunder Compute's on-demand rate is not subject to interruption.
Infrastructure and Target Audience
Nebius runs its own data centers in Europe (Finland, UK) and the US, with H100, H200, and B200 HGX NVLink configurations. The platform includes InfiniBand networking for multi-GPU distributed training, a self-serve console with per-second billing, Kubernetes support, and a Token Factory inference API. NVIDIA's $2B investment gives Nebius early access to next-generation GPU hardware.
Nebius suits organizations that require EU data residency for GDPR compliance, teams running large-scale training that benefit from reserved capacity discounts, or groups that need an OpenAI-compatible inference API alongside GPU instances.
Thunder Compute targets individual researchers, small teams, and startups that need fast, low-cost access to A100 and H100 instances. Per-minute billing, persistent storage by default, and one-click IDE setup reduce infrastructure overhead to near zero.
Developer Experience
Nebius provides a self-serve console for configuring instances, attaching storage volumes, and managing GPU resources. Per-second billing is precise, and Kubernetes integration supports containerized ML workloads. The interface suits developers who are comfortable with cloud infrastructure management.
Thunder Compute minimizes setup time. The VS Code, Cursor, and Windsurf extensions connect to a running instance in one click, with no SSH key management or CUDA driver installation required. Storage is configured by default, so there is no volume attachment step.
When to Choose Each Provider
Choose Nebius if you need:
- EU data centers and GDPR compliance.
- Reserved capacity discounts for sustained workloads.
- InfiniBand multi-node clusters for distributed training.
- A Token Factory inference API.
Choose Thunder Compute for:
- The lowest on-demand A100 or H100 rate without commitment.
- Per-minute billing.
- Persistent storage and IDE integration included.
- Fast single-instance setup with no enterprise onboarding.
Last Thoughts on Nebius vs Thunder Compute
Nebius and Thunder Compute target different needs. Nebius serves enterprise-scale deployments with EU data residency, large clusters, and reserved capacity. Thunder Compute serves developers who want the most affordable, immediate access to managed A100 and H100 GPUs. Match the provider to your team size, compliance requirements, and whether EU infrastructure is a hard requirement.
FAQ
How Much Does Nebius Charge for H100 GPUs?
Nebius H100 NVLink on-demand is $3.85/hr per GPU as of June 1, 2026, up from the prior $2.95/hr rate. Preemptible instances run $2.15/hr and can be interrupted. Thunder Compute H100 PCIe is $2.19/hr with per-minute billing.
Is Thunder Compute Cheaper Than Nebius for H100?
Yes for single-GPU on-demand access. Thunder Compute H100 PCIe is $2.19/hr versus Nebius H100 NVLink at $3.85/hr. Nebius preemptible H100 at $2.15/hr is slightly cheaper but can be interrupted, so it is not suitable for uninterrupted training runs.
Does Nebius Offer EU Data Centers?
Yes. Nebius operates GDPR-compliant data centers in Finland and the UK. For teams with European data residency requirements, Nebius is a stronger fit than Thunder Compute, which operates in North America.
What Makes Thunder Compute Easier to Use for Individual Developers?
Thunder Compute offers VS Code, Cursor, and Windsurf extensions that connect to a running instance in one click. No SSH setup, no CUDA configuration. 100GB per GPU is included by default. Nebius is better suited to teams managing Kubernetes and cloud storage.
Does Nebius Have Preemptible GPU Instances?
Yes. Nebius preemptible H100 NVLink is $2.15/hr as of June 2026. These instances can be interrupted when Nebius reclaims capacity, so they suit checkpoint-friendly batch jobs. Thunder Compute does not offer preemptible pricing.
Which Provider Is Better for EU Data Residency?
Nebius. It operates data centers in Finland and the UK with GDPR compliance and formal data processing agreements. Thunder Compute's infrastructure is in North America and is not suited to EU data residency requirements.
What Is Nebius's Background and Ownership?
Nebius is the successor to Yandex N.V., rebuilt as an independent AI cloud company after Yandex sold all Russian assets in 2024. It received a $2B NVIDIA investment in March 2026 and is headquartered in Amsterdam with engineering teams in Finland, Serbia, and Israel.
Does Nebius Offer Reserved Capacity Discounts?
Yes. Nebius reserved capacity discounts reach up to 35% off on-demand rates for multi-month commitments. Thunder Compute does not offer reserved pricing; all access is on-demand with per-minute billing and no commitment required.