Capacity available now
We operate our own sites and partner with a trusted network. When you need GPUs, you get them — no waitlist, no lottery, no enterprise sales cycle.
AI Infrastructure
The compute behind intelligence.
Infrastructure that adapts to your workload. B200, H200 and H100 — billed by the hour, or locked in when you're ready to commit.
NO HIDDEN FEES · NO SURPRISE EGRESS CHARGES · SCALE UP OR DOWN ON DEMAND
Workloads
Built for the full AI lifecycle.
Training, inference, agents, scientific simulation, geospatial analytics and more — if it needs GPUs, it runs on Kala.
Why Kala
Not another GPU reseller.
We built the infrastructure ourselves — from the power source to the API. That means lower costs, more capacity, and no surprises on your bill.
NO WAITLIST
NO HIDDEN FEES
Zero egress charges
Egress is free. The price you see on the pricing page is the price you pay — we don't recoup margin through data transfer fees.
ENERGY ADVANTAGE
Lower cost at the source
Our sites run on stranded and curtailed energy that traditional data centres can't access. Lower power costs flow through to competitive GPU rates without sacrificing hardware quality.
LATEST HARDWARE
B200, H200, H100 — now
We stock the current generation. No waiting for hardware allocation cycles or being pushed to older SKUs — the fleet you see in the pricing table is what's available.
Pricing
Choose your compute. Start scaling today.
Transparent hourly rates for powerful GPUs. Scale your workloads without hidden costs or surprise charges.
NVIDIA B200
Get started →
$4.07
PER GPU / HOUR
Built for frontier-scale training, large-scale AI modelling and simulation.
| Memory | 96 GB HBM3e |
| Bandwidth | 8.0 TB/s |
| Tensor cores | 528 |
| Released | 2024 |
MOST POPULAR
NVIDIA H200
Get started →
$3.20
PER GPU / HOUR
Optimised for high-throughput AI training, inference and data analytics.
| Memory | 141 GB HBM3e |
| Bandwidth | 4.8 TB/s |
| Tensor cores | 528 |
| Released | 2024 |
NVIDIA H100
Get started →
$2.00
PER GPU / HOUR
Reliable power for development, fine-tuning and scalable cloud workloads.
| Memory | 80 GB HBM2e |
| Bandwidth | 3.35 TB/s |
| Tensor cores | 456 |
| Released | 2023 |
NEED DEDICATED CLUSTERS, RESERVED CAPACITY OR CUSTOM CONFIGURATIONS? TALK TO OUR TEAM
Pricing models
On-demand or reserved — pay how it suits you.
No single model fits every team. Use on-demand for full flexibility, or commit to reserved capacity for a lower effective rate on sustained workloads.
PAY-AS-YOU-GO
On-demand
Billed hourly with no commitment. Spin up GPUs in minutes and release them the moment your job finishes. Full published rates — exactly what you see in the pricing table above.
- No upfront payment or lock-in
- Access to the full GPU fleet
- Best for experimentation, development, and bursty workloads
- Minimum billing: one hour
COMMITTED
Reserved capacity
Lock in capacity for a set term and receive a lower effective hourly rate in exchange. Best for teams running predictable, sustained workloads — long training runs, batch pipelines, or production inference.
- Discounted rates available on committed terms
- Guaranteed availability — no queuing
- Available across H100, H200, and B200
DEDICATED
Private clusters
Isolated hardware reserved exclusively for your organisation. Uncontended performance for security-sensitive or large-scale production workloads, with full control of the stack.
- Dedicated nodes — no shared tenancy
- Custom cluster sizes and GPU configurations
- Priority support and SLA options
- Pricing on application
CUSTOM BUILD
Bespoke requirements
Need a specific GPU, dedicated hardware, or a configuration we don't list? We'll procure and install exactly what your workload requires — talk to us about a custom deployment.
Need something we don't list?
If you have bespoke requirements — specific GPUs, dedicated hardware, or a particular configuration — we're happy to procure and install exactly what you need.
Compute with its own power source.
Kala Mesh distributes workloads across our energy-site and city-based infrastructure — training runs where power is cheapest, inference close to your users, with geo-aware placement. And if you need a GPU configuration we don’t have free, Kala Mesh sources it from our trusted partner network, so you’re never blocked on hardware. One platform, one API, one bill.
On-demand capacity
Spin up GPUs in minutes for development, testing and short-term jobs. Pay only for what you use.
Dedicated compute
Dedicated GPU resources with guaranteed availability for long-running training and production inference.
Scalable clusters
Expand from single nodes to multi-GPU clusters as demand grows, without re-architecting your systems.
Burst capacity
Absorb traffic spikes and crunch periods with additional compute, available the moment you need it.
Platform
Built for how AI teams actually work.
Standard tools and standard access methods — the interfaces your team already uses, without proprietary lock-in.
ACCESS
Console, API & CLI
Provision and manage compute through a web console, a REST API, or a command-line interface.
FRAMEWORKS
PyTorch, TensorFlow & JAX
Full CUDA access means existing training code runs without modification.
CONTAINERS
Docker & Kubernetes
Containerised workloads via Docker with NVIDIA Container Toolkit, and cluster orchestration via Kubernetes.
INFRASTRUCTURE AS CODE
Terraform
Provision and manage GPU resources declaratively alongside the rest of your infrastructure.
BARE METAL
Root access, no restrictions
Full control of the machine — install any driver, kernel module, or orchestration layer your stack requires.
NETWORKING
High-bandwidth GPU interconnect
Low-latency interconnect between nodes for distributed training at scale.
SECURITY
Isolated tenancy
Secure, isolated compute environments with private networking between nodes.
OBSERVABILITY
Real-time monitoring
Live workload and utilisation visibility through the Kala Mesh platform.
Getting started
From signup to first job in three steps.
01
Create your account
Sign up on the Kala compute platform. It takes a couple of minutes.
02
Choose your compute
Pick your GPU and configuration: on-demand instances, dedicated nodes or a cluster.
03
Launch and scale
Run your workloads. Scale up, scale down or burst as needed, with full visibility into usage and spend.