Voltage Park
GPU clusters by the hour or on reserve, owned by a nonprofit endowment rather than by investors.
Highlights
- On-demand NVIDIA HGX H100 nodes, self-serve and bootable in about 15 minutes, from $1.99/hr
- NVIDIA Blackwell and Grace Blackwell GPUs available alongside Hopper
- Bare-metal access for direct hardware control during large training runs
- Up to 3,200 Gbps aggregate bandwidth over NVIDIA Quantum-2 InfiniBand
- Dedicated Reserve contracts (12+ months) for committed, long-term capacity
- Company-owned hardware across multiple US data centers rather than resold capacity
- Tier 3+ data center architecture with a stated 99.982% uptime
- No long-term commitment required for on-demand usage
External link — opens voltagepark.com in a new tab. Voltage Park is a third-party product; we are not affiliated with it.
About Voltage Park
What it is
Voltage Park rents NVIDIA H100 and Blackwell clusters for AI training and inference, either on demand by the hour or on longer dedicated-reserve contracts. Its ownership is unusual: it is funded by the Navigation Fund, a nonprofit endowment, rather than by conventional venture capital.
Why it's different
The ownership is not trivia. A GPU cloud backed by an endowment rather than by investors expecting a return on a fixed timeline has different pressure on pricing, and Voltage Park has generally priced below the market. For teams whose binding constraint is cost per GPU-hour, that is the reason to look. What to weigh against it: the hyperscalers offer a surrounding ecosystem — storage, networking, managed services, support — that a specialist GPU renter does not, and moving data in and out is your problem. Availability of large contiguous clusters fluctuates across this whole market, so what matters is what you can actually reserve, not the catalogue.
How people use it
It suits training and fine-tuning runs where the work is self-contained and the cost is dominated by GPU time — research teams, startups training their own models, anyone doing periodic large jobs rather than continuous serving. The practical approach is on-demand for experiments and reserved capacity once the workload is predictable, since the discount for commitment is substantial. Check interconnect specifications for multi-node training, where they determine everything.
Written by the n3os team. We are not affiliated with Voltage Park.
This listing was written from public information, without Voltage Park’s involvement. If you own it and something here is wrong — or you would rather not be listed at all — email us and we will correct or remove it.
Get the ones worth knowing about
We write one of these for every tool worth the trouble. Get the new ones, plus what we have found genuinely useful lately.
Your address goes to Buttondown, who send the emails on our behalf. One click unsubscribes, and the list is never sold or shared.