GPU rental · India, the EU and the US
Rent NVIDIA GPU servers, from H100 to B300
Dedicated 8-GPU servers for training and inference, sourced across more than ten partner providers. One quote, with no markup from us; you run the stack, we keep the hardware running.
From $2.95 per GPU-hour · One-year term · Quote usually within one working day
What a server costs
GPU server prices per hour and per month
Every server is a full node of 8 GPUs, so the node price is 8 times the GPU-hour rate. Floor rates on our longest term; shorter terms cost more, and your rate is confirmed in writing before you commit.
H100 server: $23.60 an hour
8 x NVIDIA H100, 80 GB HBM3 per GPU, at $2.95 per GPU-hour. About $17,200 a month at 730 hours.
H200 server: $26.80 an hour
8 x NVIDIA H200, 141 GB HBM3e per GPU, at $3.35 per GPU-hour. About $19,600 a month at 730 hours.
B200 server: $30.00 an hour
8 x NVIDIA B200, 180 GB HBM3e per GPU, at $3.75 per GPU-hour. About $21,900 a month at 730 hours.
B300 server: $36.00 an hour
8 x NVIDIA B300, 288 GB HBM3e per GPU, at $4.50 per GPU-hour. About $26,300 a month at 730 hours.
Which GPU to rent
Match the GPU server to the job
Memory decides most of it: the model, its context and its batch have to fit. If you are not sure, tell us the model and we will size it with you.
Training or fine-tuning the largest models: B300
Our largest memory available today, 288 GB of HBM3e per GPU, for frontier-scale training, very long context and the biggest mixture-of-experts models. Allocation is the tightest, so ask early.
Large-model training and FP4 inference: B200
Blackwell at a lower rate than the B300, with 180 GB per GPU. A strong fit for pre-training and heavy fine-tuning, and for serving large models at FP4.
Long context and models over 80 GB: H200
Hopper with 141 GB of HBM3e per GPU. Choose it when a model or its KV cache outgrows the H100, without moving to Blackwell.
Proven, lowest-cost training and inference: H100
The workhorse, at our lowest rate and usually the shortest lead time. Ideal for fine-tuning and serving models that fit in 80 GB per GPU.
Get a quote
Tell us the GPU, the region and the term
Leave a name and a number or email. We call you back with real server options from across the network, usually within one working day.
Questions
Renting a GPU server: what people ask
A full 8-GPU server costs, per hour at our floor rates: H100 $23.60, H200 $26.80, B200 $30.00, B300 $36.00. Over a month of 730 hours that is about $17,200 for the H100, $19,600 for the H200, $21,900 for the B200, $26,300 for the B300. These are floor rates in USD on our longest term; shorter terms cost more, and your exact rate depends on region, term and node count.
One GPU running for one hour. GPU prices are quoted per GPU-hour so that different GPUs and server sizes are easy to compare. An 8-GPU server running for an hour uses 8 GPU-hours, so its hourly cost is 8 times the GPU-hour rate.
A GPU server is a physical machine with GPUs in it; ours are 8-GPU NVIDIA HGX servers. A GPU cloud instance is usually a virtual machine on a shared server, billed on demand. Bare metal means the whole physical server is yours alone, with root access and no virtualisation layer. FlexiCloud rents dedicated, single-tenant bare-metal servers on a term, not on-demand instances.
Start from memory. Models that fit in 80 GB per GPU run well on the H100, our lowest rate. Long context or models over 80 GB point to the H200 (141 GB). Large-model training and FP4 inference suit the B200 (180 GB), and the biggest models and longest context the B300 (288 GB). Our GPU memory guide shows how to work out what your model needs.
For most teams, over a one- to three-year horizon, yes. Buying means paying for the whole server up front, plus power, cooling, hosting and the engineers to run it, and the hardware loses value as each new generation ships. Renting turns that into a fixed monthly cost with the hardware maintained for you. Our rent, buy or API guide sets out when each one makes sense.
Yes. We source dedicated GPU servers in India for workloads whose data has to stay in the country, as well as in the EU and the US. Capacity is allocated per region, so tell us where the data has to live and we confirm availability and lead time in writing before you commit.
Before you rent
Choosing and sizing a GPU server
How the broker model works, whether your model fits, whether to rent, buy or use an API, and how the GPU generations compare.
Renting GPUs: why going through FlexiCloud beats going direct
We partner with more than ten GPU providers in India, the EU and the US, so you do not have to. Here is what that gets you: better pricing, priority on scarce GPUs, live inventory across datacentres and regions, privacy, and one honest point of contact instead of ten sales teams.
Read itWill your model fit? A practical guide to GPU memory for AI
Whether your model fits is arithmetic, not a benchmark. The bytes-per-parameter maths for inference and training, the levers that shrink it (quantization, LoRA, checkpointing), and a back-of-envelope guide to size any job, mapped to what fits on a B300.
Read itRent GPUs, buy your own, or use an API? A guide for teams building on AI
Pay-per-token API, buy your own GPUs, or rent bare-metal nodes? The right answer is about utilisation and control, not hype. A fair look at all three, and a simple way to decide which your workload actually needs.
Read itHopper vs Blackwell: NVIDIA’s GPU generations compared
NVIDIA’s Hopper (H100, H200) and Blackwell (B200, B300) generations compared: memory, architecture, interconnect, power and price, and a practical guide to which one actually fits your AI workload.
Read it
Rent a GPU server through one quote
Tell us the GPU, how many servers and the region. We come back with real options from across our network, with the price and lead time in writing before you commit.