H200 nodes now live in baku-1. InfiniBand fabric online.  Reserve capacity →
The Baku Compute Company · Azerbaijan

GPU clusters by the hour, in the Land of Fire.

Vetted H100 and H200 nodes in our Baku datacenter. Need 256 H100s for a week? Done. Eight H200s for a year? Sure. Spin down whenever — you're never locked into a long-term contract.

Rent H100s from $1.85/hr * price varies with market conditions
Independent market reference

Ornn Compute Price Index · H100 SXM

daily close /gpu·hr

Loading Ornn daily index data…

Trailing 90 days Daily settled series Source: Ornn ↗

External global market index in USD per GPU-hour. It is not a Baku Compute quote or an indication of local availability.

Instances

Transparent per-hour pricing.

Spot prices float with demand on the cluster. Reserve for a fixed term to lock a rate.

GPUNodeVRAM / GPUInterconnectFrom (gpu·hr)
H100 SXM80 GB3.2 Tb/s IB$1.85
H200 SXM141 GB3.2 Tb/s IB$2.40
A100 SXM80 GB1.6 Tb/s IB$1.10
L40S48 GBPCIe$0.65

VMs spin up in ~5 min · bare-metal & managed Slurm on request · failed nodes refunded and repacked.

01 / training

Large runs, short terms

Burst to hundreds of interconnected GPUs for a single training run, then hand them back. No annual commitment to amortize.

02 / inference

Steady serving

Hold dedicated nodes at a reserved rate for production inference, with headroom to scale during peaks.

03 / research

Labs & universities

Per-hour billing that fits grant cycles. Built for teams at ADA, the academies, and regional research groups.

Why Baku

A neutral compute hub between Europe, the Gulf and Central Asia.

Baku sits on the fiber routes linking Frankfurt, Dubai and the Caspian corridor — low-latency to three regions at once, on power that's among the cheapest in the neighborhood.

~40 ms
RTT to Frankfurt & Dubai
24/7
On-call cluster engineering
0 ₼
Data egress fees, ever
5 min
Median VM spin-up time
FAQ

The usual questions.

Is it bare metal, VMs, or Kubernetes? +
Standard orders are provisioned as virtual machines. For bare metal, managed Slurm, or Kubernetes, contact us and we'll set up a dedicated cluster.
How fast do nodes spin up? +
About five minutes for VMs. Reserved bare-metal clusters are provisioned ahead of your start time.
Are nodes fully interconnected with InfiniBand? +
H100 and H200 clusters run a 3.2 Tb/s InfiniBand fabric for collective ops. Smaller PCIe instances are for single-node and inference workloads.
What happens when hardware fails? +
It happens — GPU failure rates are far higher than web servers, so we design for it. We refund failed nodes and repack you onto healthy hardware.
Can I pay and contract from Azerbaijan? +
Yes. We bill in USD and support local settlement. Talk to us about terms for academic and enterprise accounts in the region.
Get started

Tell us what you need to run.

Send your workload, GPU count and timeframe — we'll come back with availability and a price.

hello@bakucompute.com