GPU Hours
NVIDIA compute, sold by the hour.
Rent the GPU capacity you need — for an afternoon or for a year — running in Pruna Trading's own AI data centers.
01
Choose how you buy
Every option is priced per GPU hour. The difference is commitment and exclusivity.
Pay as you go
On-demand
Billed per GPU hour with no minimum term. Start when you need compute and stop when you don't.
- No long-term commitment
- Billed per GPU hour
- Ideal for experiments and bursts
Guaranteed capacity
Reserved
Commit to a term and your GPUs are held for you — at a lower hourly rate than on-demand.
- Capacity held for your team
- Lower effective hourly rate
- Flexible term lengths
Single-tenant
Dedicated cluster
A private GPU cluster in our data centers, built to your node count and networking requirements.
- Single-tenant hardware
- Sized to your workload
- High-speed multi-node fabric
02
Included with every GPU hour
No add-on surprises. The essentials for running serious AI workloads come as standard.
- Latest-generation NVIDIA data center GPUs
- NVIDIA drivers and CUDA ready to go
- High-bandwidth networking for multi-node jobs
- Usage reporting down to the GPU hour
- Monitoring by our 24/7 operations team
- Direct access to infrastructure engineers
03
Built for every stage of AI
Train foundation and custom models
Reserve multi-node capacity for long training runs, with the power, cooling and fabric to keep every GPU busy until the job finishes.
Fine-tune on your own data
Spin up on-demand GPU hours for a fine-tuning run, then release them. You pay for the hours used and nothing more.
Serve models in production
Run inference on reserved or dedicated capacity so your latency and availability don't depend on someone else's queue.
Explore without buying hardware
Give researchers access to serious compute for experiments and prototypes, without a capital purchase.
04
Billing, explained
By the GPU hour: the number of GPUs allocated to you multiplied by the hours they are allocated. Usage reports show the breakdown.
Tell us how many GPU hours you need.
Share your GPU count, duration and timeline — we'll come back with capacity and a quote.