Voltage Park review
Updated September 28, 2026 · DeployCue Editorial Team · Facts checked September 26, 2026
We earn a commission when you sign up through links on this site. Partner status can affect where a brand appears in our rankings. Partners with a live offer in your country come first, then every other brand by score.
Our verdict
Voltage Park is a US GPU cloud for teams that need datacenter accelerators, real fabric between nodes and enterprise certifications rather than the cheapest card on the market. Entry GPU pricing starts at $1.99 per hour, above the €0.61 median across the GPU category.
Key facts
- Best for
- Multi-GPU training and inference on HGX silicon, metered by the hour with no egress charge.
- GPU models
- NVIDIA HGX H100, NVIDIA HGX B200, NVIDIA HGX GB200, NVIDIA HGX B300, NVIDIA HGX GB300↗
- GPU hour from
- $1.99↗
- Billing granularity
- Hourly↗
- Egress fee
- No egress fee↗
- Locations
- 6↗
- Interconnect
- 3,200 Gbps NVIDIA Quantum-2 InfiniBand↗
- Compliance certifications
- ISO/IEC 27001, SOC 2 Type II, HIPAA↗
- Legal entity
- Voltage Park, Inc.↗
- Headquarters
- United States↗
- Support channels
- Email, Help-centre, Community↗
- Pricing model
- Hourly billing
- Official website
- voltagepark.com
VPS
Storage
Facts checked September 26, 2026, 9 sources
Apps and profiles
- Elsewhere
- LinkedInXreddit.comyoutube.com
Pros and cons
Pros
- Egress is not a line item: No egress fee on GPU traffic and No egress costs on storage.
- One catalogue covering H100, B200, GB200, B300 and GB300, listed as NVIDIA HGX H100, NVIDIA HGX B200 and NVIDIA HGX GB200.
- An SLA of 99.982% and ISO/IEC 27001, SOC 2 Type II and HIPAA for buyers with a security review.
Cons
- Support runs on email, help centre and community only, so no phone or live chat during a long run.
- Sites sit in the US, so data placement has a single market to work with.
- hourly metering means short jobs still pay for a whole step.
Who it suits
- Teams training large models that need HGX boards and 3,200 Gbps NVIDIA Quantum-2 InfiniBand between nodes.
- Buyers whose security review expects ISO/IEC 27001, SOC 2 Type II and HIPAA.
Who should look elsewhere
- Salad, for anyone after consumer cards such as NVIDIA RTX 3060, NVIDIA RTX 3070 and NVIDIA RTX 3070 Ti at a different price tier.
- ComputePrices.com, if a single index spanning H100 SXM, A100 SXM and H200 decides your shortlist.
- SberCloud, for work on older accelerators like B300, H100 and A100.
Core features
The catalogue is built on NVIDIA HGX silicon: NVIDIA HGX H100, NVIDIA HGX B200 and NVIDIA HGX GB200. Capacity sits across 6 sites, and the nodes are joined by 3,200 Gbps NVIDIA Quantum-2 InfiniBand, so a training job spans real datacenter fabric rather than a shared bus. Every board listed is a datacenter part, which suits serious training and inference work.
The company behind it
The provider is Voltage Park, Inc., headquartered in the US, so contracts, compliance review and data handling all sit in one jurisdiction. Support runs on email, help centre and community, which covers routine infrastructure questions but leaves out a live channel when a training run stalls.
How Voltage Park compares
Voltage Park vs RunPod8.9 vs 9.5
Voltage Park vs ScalaHosting8.9 vs 8.3
Voltage Park vs Contabo8.9 vs 9.2
Voltage Park vs Kamatera8.9 vs 9.5
Voltage Park vs Cloudways8.9 vs 9.2
Voltage Park vs PlanetHoster8.9 vs 8.7
Voltage Park vs Acronis8.9 vs 9.9
Voltage Park vs Aruba8.9 vs 9.3
Voltage Park vs Hostinger8.9 vs 9.1
Voltage Park vs Decodo8.9 vs 9.1
Voltage Park vs MarsProxies8.9 vs 9.0
Voltage Park vs hide.me8.9 vs 8.7
Voltage Park vs Proton Drive8.9 vs 8.3
Frequently asked questions
What does a GPU cost to start with?+
Entry GPU pricing starts at $1.99 per GPU hour, and the entry configuration is 1 x NVIDIA H100 GPU, On-Demand (100 Gbps Ethernet), with $1.99 as the matching entry price. Reserved terms sit alongside on-demand, so steady workloads can be bought ahead rather than metered at the same rate.
Is there an egress charge?+
GPU traffic carries No egress fee, and storage is listed as No egress costs. For jobs that push checkpoints or dataset shards off the platform repeatedly, that removes a line item which many providers price by the gigabyte.
How is usage metered?+
Billing granularity is hourly, and the models on offer are hourly and reserved. A run is paid for by the hour, with reserved capacity for steady workloads. Around 49.1% of the providers in the set bill hourly, so this is the common case rather than an edge.
Which GPUs can I book?+
The catalogue is NVIDIA HGX H100, NVIDIA HGX B200 and NVIDIA HGX GB200, so Hopper and Blackwell parts sit on one platform, including the paired GB200 and GB300 boards used for large-model training. There is no consumer or gaming tier in the list, which keeps the choice narrow but consistent for cluster work.
Where does it run?+
All capacity sits in the US, spread across 6 sites. That suits buyers who want training data and outputs to stay in a single market, and it narrows the options for teams that need another country for residency or latency reasons.
Will a regulated team pass review?+
The security material lists ISO/IEC 27001, SOC 2 Type II and HIPAA, which covers most enterprise questionnaires, and service levels promise 99.982%. Anything beyond that, such as a named regulator or a local licence, is a separate conversation with the provider before a workload moves across.
What else runs on the platform?+
Alongside GPU capacity the platform lists bare metal and vps and object and block. A training job can be paired with general compute and with both object and block storage from one account, which saves bolting a second provider onto the pipeline.
Sources
- voltagepark.com checked September 26, 2026
Backs: Billing granularity, Interconnect, SLA uptime
“Self-serve and spin up nodes in 15 minutes. No long-term contracts. Starting at $1.99/hr.”
- voltagepark.com/about checked September 26, 2026
Backs: Locations, Regions, Regions
“Our six Tier 3+ US data centers strategically located across the US offer significant uptime with redundancy and concurrent maintainability”
- voltagepark.com/blackwells checked September 26, 2026
Backs: GPU models
“NVIDIA B200, GB200, B300 and GB300s for your most demanding AI workloads.”
- voltagepark.com/pricing checked September 26, 2026
Backs: Billing models, Entry plan, Entry price, Free egress
“Self-serve. Up within 15 mins. No minimum terms. ... Long-Term Reserve ... 6+ month contracts with friendly payment terms.”
- voltagepark.com/pricing checked September 26, 2026
Backs: Egress fee, GPU hour from, Pricing model
“No hidden ingress, egress or support costs.”
- voltagepark.com/product/storage checked September 26, 2026
Backs: Storage types
“One unified namespace For file, object and block access”
- voltagepark.com/security checked September 26, 2026
Backs: Compliance certifications, Support channels
“We are compliant with recognized industry standards, including ISO / IEC 27001, SOC 2 Type II, and HIPAA for eligible workloads”
- voltagepark.com/terms-of-service checked September 26, 2026
Backs: Headquarters, Legal entity
“The Platform is operated by us in the United States.”
- voltagepark.com/virtual-machines checked September 26, 2026
Backs: Server types
“Teams choose bare metal performance and control, or VM flexibility and speed”
Compiled from 9 source pages, checked September 26, 2026. We did not open an account.
Related by the facts
Voltage Park
Multi-GPU training and inference on HGX silicon, metered by the hour with no egress charge.