One of the cheapest credible H100 sources with a named Delaware entity and real compliance certifications, but it is no longer standalone (rebranding to Lightning AI), all reliability numbers are self-reported, and the contract is non-refundable with Delaware arbitration.
by the bestcloudmining editorial desk · review version 2 · prose updated 2026-09-24 · facts as of 2026-09-24 · how we verify
Voltage Park is a US-based GPU cloud provider offering on-demand and reserved access to NVIDIA H100 and newer Blackwell-class clusters, now operating under the Lightning AI brand after a January 2026 merger. It targets AI developers and enterprises needing high-performance compute without long-term commitments. While it’s one of the cheapest credible H100 sources with compliance certifications, all uptime and reliability claims are self-reported, and the non-refundable contract includes binding arbitration.
You can rent bare-metal NVIDIA HGX H100, B200, GB300, and Blackwell GPUs on a per-GPU hourly basis, with self-serve on-demand access or reserved configurations for 6+ months. Access is via SSH, Kubernetes, VM, or API, and deployments are available within 15 minutes. You get full control of the instance but do not own the hardware or have proof-of-compute verification. The provider does not support cryptocurrency mining or decentralized storage; this is strictly for AI/ML and high-performance computing workloads.
Contracts are flexible (hourly, no term) or fixed-term (6+ months for large clusters), with no refund or termination rights except as explicitly stated. Fees are non-refundable, charged incrementally to your card, and remain due even if service is suspended. There is no published SLA page, though uptime is claimed at 99.982%. Payouts don’t apply - this is a compute rental, not mining. Minimum purchase is not specified, but account signup and continuous billing are required.
There is no cryptographic proof-of-compute, hardware TEE, or public verification of GPU availability. Trust is based on operator claims: Voltage Park asserts compliance with ISO/IEC 27001, SOC 2 Type II, and HIPAA for eligible workloads, and operates six Tier 3+ US data centers. These certifications and the named Delaware corporation add legitimacy, but all performance metrics - including uptime - are self-reported with no third-party validation.
Voltage Park has a strong reputation as a low-cost, reliable H100 provider with 100+ named customers and enterprise partnerships. No fund-loss or insolvency incidents are reported. However, a Reddit user reported ghost billing on TensorDock (acquired in 2025) with unresponsive support, signaling potential operational risks. The January 2026 merger with Lightning AI means the brand is transitioning, though existing deployments are said to be unaffected. All facts are verified, but ongoing reliability post-merger remains to be independently confirmed.
Pricing starts at $1.99 per GPU per hour for on-demand H100 access, with no hidden ingress, egress, or support fees. Reserved or large-scale clusters (32+ GPUs) require direct contact for pricing. Payments are in USD only via credit card, with continuous incremental billing. KYC status is unknown, but enterprise deals involve sales onboarding. No minimum purchase is stated, but usage is billed hourly and non-refundable, increasing cost if instances are left running unintentionally.
Voltage Park (now Lightning AI) is a credible, low-cost option for developers needing immediate H100 or Blackwell access without long-term contracts. However, the non-refundable billing, lack of independent verification, and merger-related rebranding mean you should start with a small test deployment. Not recommended for users who require provable compute or exit flexibility.
| Documented price | $1.99 per GPU/hr |
|---|---|
| Notes | Observed 2026-09-24: H100 on-demand 'Starting at $1.99/hr', self-serve, up within 15 min, no long-term contract (https://voltagepark.com/ and https://voltagepark.com/pricing). Reserved/complex configurations ('6+ month', 32-8000+ H100, 3200 Gbps InfiniBand) are 'Contact for pricing'. The pricing page advertises no hidden ingress, egress or support costs. |
Run the long math before paying: cloud mining calculator →
Ready to try Voltage Park? Start small, verify one payout, and only scale up once the hashrate or compute actually matches what you paid for.
Based on our checks as of 2026-09-24: Voltage Park scores 7.8/10 in our weighted review. We logged no red flags. User reports and operator background are documented with dated sources in the meta-review below.
You get temporary, full access to a bare-metal NVIDIA GPU (H100 or newer) in a US data center, accessible via SSH, Kubernetes, or API. You do not own the hardware or receive cryptographic proof of its operation.
You pay $1.99/hour per GPU (starting rate) with no hidden fees for ingress, egress, or support. Billing is continuous and non-refundable; you’re charged per hour as long as the instance runs.
No minimum is stated, and you can deploy on-demand instantly. There’s no free trial, but you can start small to test performance before scaling.
Payouts don’t apply - this isn’t mining or staking. You’re renting compute power, not earning cryptocurrency. You pay for usage; you don’t receive payments.
If operations cease, your instances and data could be lost with no recourse. The contract includes binding arbitration and no refunds, so you’d have limited legal options.
Buying hardware gives you full control but requires upfront cost and maintenance. This is cheaper and faster for short-term AI workloads, but unlike mining, you’re paying for compute, not earning crypto - there’s no revenue potential.
Centralized open-model inference and GPU-cluster provider with published per-token and per-GPU rates; fiat-only and not decentralized.
Developer GPU cloud with published per-second rates across Community and Secure Cloud, plus serverless endpoints and clusters.
US curated GPU cloud and OpenAI-compatible inference; H100 $3.19/hr, H200 $3.99/hr on-demand with hourly billing and $5 never-expiring credits.
No comments yet - be the first. Posting runs a short proof-of-work in your browser (anti-spam), no account needed.