The GPU catalogue.

NVIDIA B300, B200, H200 and H100 nodes in Abu Dhabi and Dubai. Eight GPUs per node, tested before you sign. All rates, on-demand and reserved, are quoted per request. We confirm your rate, down payment and availability before you commit.

NVIDIA B300 Blackwell Ultra

Memory
288 GB HBM3e per GPU
Node
8 GPUs per node
Terms
Spot to 5 years
Rate
On request

The largest memory per GPU in the catalogue, for training runs and large-scale inference that need to stay on one node.

Request a B300 quote

NVIDIA B200 Blackwell

Memory
192 GB HBM3e per GPU
Node
8 GPUs per node
Terms
Spot to 5 years
Rate
On request

Blackwell capacity for clusters: the fit for a multi-node training run or a reserved inference fleet.

Request a B200 quote

NVIDIA H200 Hopper

Memory
141 GB HBM3e per GPU
Node
8 GPUs per node
Terms
Spot to 5 years
Rate
On request

A fit for fine-tuning and memory-heavy inference on proven Hopper capacity.

Request an H200 quote

NVIDIA H100 Hopper

Memory
80 GB HBM3 per GPU
Node
8 GPUs per node
Terms
Spot to 5 years
Rate
On request

The workhorse: benchmarks, proofs of concept and production workloads that want Hopper availability without a long commitment.

Request an H100 quote

Terms, for every GPU.

Spot
Tests, benchmarks, short bursts
1 month
A proof of concept
3 months
A training run
1 year
Production inference
2 years
Reserved capacity
3 years
Reserved capacity
5 years
Long-term reserved capacity

Every rental: acceptance test on your workload, test report before you sign, data resident in the UAE, one contract and one invoice from Oasis Compute FZCO.

Tell us what you need.

Get a quote
WhatsApp