DCompute Cloud
A GPU cloud for H100 jobs with per-second billing, five-minute provisioning, and reserved clusters for larger teams.
This GPU rental service provides on-demand access to NVIDIA H100 80GB SXM GPUs for AI training and inference, with per-second billing and fast provisioning. It is designed for teams that need current-generation NVIDIA hardware without committing to traditional cloud pricing models, long contracts, or sales-led procurement.
A single H100 is listed at $1.99/hour, with billing running only while the instance is active. This makes the service suitable for short experiments, burst workloads, and longer training or inference jobs where avoiding idle GPU time can make a meaningful difference to overall compute costs.
Key Features
NVIDIA H100 80GB SXM
Run workloads on NVIDIA H100 80GB SXM hardware, providing the CUDA-compatible NVIDIA environment teams expect for modern AI training and inference workloads.
The service is focused on giving users direct access to the GPU capacity they need without adding unnecessary procurement overhead.
Per-Second On-Demand Billing
Pay for GPU usage based on active runtime rather than committing to a fixed amount of compute.
A single H100 is listed at $1.99/hour, while the billing meter runs only when the machine is active.
This model is useful for:
- Short experiments
- Training runs
- Burst inference
- Development and testing
- Workloads with variable GPU requirements
Fast Provisioning
GPU instances can be provisioned quickly, with a listed startup time of approximately five minutes to first token.
This reduces the waiting period between deciding to run a workload and getting access to the required hardware.
No Sales Call Required
Get access to GPU capacity without going through a traditional sales process.
The self-serve positioning makes it easier for teams to provision compute when they need it rather than waiting for a sales conversation or custom cloud agreement.
Reserved GPU Clusters
Teams with larger or more predictable workloads can reserve GPU capacity rather than relying exclusively on individual on-demand instances.
Reserved clusters range from:
- 10 to 200 GPUs
- 30-day to 6-month terms
Prepayment can also reduce the effective GPU rate for longer commitments.
Flexible Payment Options
The service supports both invoice and card payments, giving smaller teams and larger organizations different ways to handle GPU procurement.
Built For AI Teams
The platform is designed for teams that care about direct access to NVIDIA GPU hardware and want greater control over how long they pay for it.
It is particularly useful for:
- AI startups running training and inference workloads
- ML engineers experimenting with H100 infrastructure
- Research teams running GPU-intensive experiments
- Developers needing temporary high-performance compute
- Growing AI teams that need multiple reserved GPUs
- Organizations looking for longer-term H100 capacity
Common Use Cases
Model Training
Provision H100 GPUs for training workloads without maintaining permanent hardware.
AI Inference
Run inference workloads on H100 capacity and shut down instances when they are no longer required.
Short Experiments
Use on-demand billing for experiments where purchasing or reserving long-term capacity would create unnecessary cost.
Burst Compute
Scale GPU usage around periods of high demand without keeping expensive hardware active when workloads are idle.
Reserved Training Clusters
Reserve between 10 and 200 GPUs for longer-running projects with terms from 30 days to six months.
Cost-Conscious GPU Workloads
Use active-time billing to avoid paying for GPU capacity while it is sitting unused.
On-Demand vs. Reserved Capacity
The service supports two primary ways to access H100 capacity.
On-demand is designed around per-second billing and short-term flexibility. It works well when GPU requirements change frequently or when you need compute for individual jobs.
Reserved capacity is intended for larger, predictable workloads, with clusters from 10 to 200 GPUs and terms ranging from 30 days to six months. Prepayment can provide a lower rate for teams willing to commit to capacity.
Pricing and Commercials
The listed on-demand price is $1.99 per hour for a single H100 80GB SXM, with billing calculated per second while the instance is active.
For larger requirements, reserved clusters support 10–200 GPUs over 30-day to 6-month terms, with potential prepay discounts. Payment can be made by card or invoice.
Why It Matters
For AI teams, GPU cost is not just about the hourly rate. Paying for capacity while it sits idle can quickly make experiments and variable workloads more expensive.
This service focuses on the hardware itself and a straightforward billing model: provision an H100 when you need it, pay while it is active, and turn it off when the workload ends. For teams with larger predictable requirements, reserved clusters provide another path to securing GPU capacity.
Get H100 Compute Without the Cloud Overhead
Access NVIDIA H100 80GB SXM GPUs with per-second billing, fast provisioning, and flexible on-demand or reserved capacity for AI training and inference workloads.
