Voltage Park
6.3/10US-based GPU cloud that rents NVIDIA H100 and Blackwell capacity on-demand or on long-term reserved contracts for AI training and inference teams.
Data last updated November 2026 · Our methodology
Overview
Voltage Park is a GPU cloud that rents NVIDIA HGX H100 servers, and Blackwell systems on contract, to teams training and serving AI models. It started in 2023 with backing from a nonprofit linked to Jed McCaleb, which funded an initial order of about 24,000 H100s. In January 2026 it merged with Lightning AI. The combined company trades as Lightning AI, but the Voltage Park site, pricing, and existing contracts are still in place. There are three ways to buy. On-demand Ethernet instances run from 1 GPU, and on-demand InfiniBand clusters start at 8 GPUs; both are self-serve, can be running in about 15 minutes, and need no contract. Long-term reserved clusters start at 32 GPUs on terms of 6 months or more. The only published rate is $1.99 per H100 GPU-hour. InfiniBand, reserved, and Blackwell pricing are quote-only. The company says it charges no ingress, egress, or support fees. Customers get virtual machines and bare metal, managed Kubernetes, a managed Slurm service, attachable storage, and a REST API for deploying and monitoring instances. It suits AI startups, research labs, and enterprises that want cheap H100 hours in US Tier 3+ data centers with ISO 27001, SOC 2 Type II, and HIPAA coverage for eligible workloads. The main drawback is the narrow scope and uneven operations. It offers only GPU compute in the US, with no general web hosting or CDN. The SemiAnalysis ClusterMAX 2.0 review rated it Silver, noting that on-demand capacity was often sold out, Slurm setup was rocky, hardware failures needed manual intervention, and security patches lagged.
Our Rating
AI Software6.2/10
Hosting & Infrastructure6.4/10
Who Voltage Park is best for
Best software for StartupsGood fit
AI startups can rent a single H100 by the hour with no contract, at one of the lowest published rates in the market.
ML engineers get root access to VMs and bare metal, Kubernetes, Slurm, and a REST API. API keys, however, have to be requested from support.
Reserved clusters of 32 to 8,000+ InfiniBand-connected GPUs on 6+ month terms come with ISO 27001 and SOC 2 Type II compliance.
Less suited
Low hourly H100 pricing has made it popular with researchers, but it offers no academic-specific programs that we could confirm.
It sells only GPU compute for AI workloads, with no general web or application hosting.
Pros & Cons
Pros
- Its on-demand H100 rate of $1.99 per GPU-hour is among the lowest SemiAnalysis found, and Voltage Park says it charges no egress or support fees.
- A single GPU can be rented self-serve with no contract and be running in about 15 minutes.
- It scales from 1 GPU to reserved InfiniBand clusters of 8,000+ GPUs in US Tier 3+ data centers.
- SemiAnalysis praised support for responsiveness after engineers rebuilt a login image in under 24 hours.
- Its security page lists ISO/IEC 27001, SOC 2 Type II, and HIPAA for eligible workloads.
Cons
- SemiAnalysis found on-demand H100 capacity regularly sold out during testing.
- Failed nodes on on-demand clusters have to be cycled out manually by operators, unlike the automated recovery at top-tier providers.
- SemiAnalysis found a preinstalled NVIDIA container toolkit about nine months out of date and exposed to critical CVEs.
- Only the entry H100 rate is published. InfiniBand, reserved, storage, and Blackwell pricing all need a sales quote.
- The service runs only in the US and offers only GPU compute, with no CDN, managed databases, or serverless products.
- API keys cannot be self-generated at first and must be requested from support.
Key Features
Not included
- Managed WordPress
- Shared Hosting
Voltage Park Pricing
Long-Term Reserve
- 3200 Gbps InfiniBand
- Contact sales for pricing
- 32 to 8,000+ GPUs in a single configuration
- Contracts of 6 months or more with payment terms
- Blackwell B200, B300, GB200, and GB300 available on contract
On-Demand Ethernet
- No minimum term
- 1 to 1,016 HGX H100 GPUs
- H100 from $1.99 per GPU-hour
- No ingress, egress, or support fees
- Self-serve, running in about 15 minutes
On-Demand InfiniBand
- Price on request
- 3200 Gbps InfiniBand
- 8 to 1,016 HGX H100 GPUs
- Self-serve, no minimum term
Frequently Asked Questions
On-demand H100 SXM instances on Ethernet start at $1.99 per GPU-hour with no contract. InfiniBand clusters, long-term reserved capacity, and Blackwell GPUs are quote-only.
Yes. The merger closed in January 2026 and the combined company is called Lightning AI. Voltage Park's site and pricing page are still live, and the company says existing agreements and deployments are unchanged.
On-demand Ethernet instances start at 1 GPU and on-demand InfiniBand at 8 GPUs, neither with a minimum term. Long-term reserved clusters need at least 32 GPUs and a contract of 6 months or more.
The pricing page says there are no hidden ingress, egress, or support costs.
Its security page lists ISO/IEC 27001, SOC 2 Type II, and HIPAA for eligible workloads, with details in the Trust Center. Under its shared responsibility model, customers secure their own operating systems, data, and workloads.
Yes. A REST API at cloud-api.voltagepark.com deploys, manages, and monitors VMs and bare metal instances using bearer tokens that last one year. First-time API keys have to be requested from support.
Yes. The pricing page lists managed Kubernetes, and SemiAnalysis reviewed a newly launched managed Slurm service and OIDC integration for Kubernetes.
Six Tier 3+ data centers in four US states. Earlier reporting named Texas, Virginia, and Washington.
Similar Tools
See all alternatives
Bare metal cloud provider that offers dedicated servers, GPU servers and virtual machines on demand, billed hourly or monthly, for developers and infrastructure-heavy businesses.
Luxembourg-based edge and cloud provider offering CDN, virtual machines, bare metal, managed Kubernetes, object storage, AI GPU cloud, and DDoS/WAF security for businesses.
CoreWeave is a GPU-focused cloud that rents NVIDIA GPU and CPU instances, managed Kubernetes, and storage to teams training and serving AI models.
Ranked near Voltage Park in AI Software
See Voltage Park in the listVoltage Park is #242 in AI Software by our score.
