Minutes, not months
Provision accelerators instantly instead of waiting on procurement cycles or capacity tickets.
Platform · GPU Compute
GPU compute for the heavy math, live in minutes.
When training and heavy inference need acceleration, GPUs take over. Launch B200, H200, H100, A100, L40S and RTX 4090 instances across 13 regions in minutes, right beside your CPU data and orchestration tiers. Scale from a single GPU to multi-node clusters, and only pay for time that is actually running.
$600 in launch credit + 20M free inference tokens. No commitment.
Engineered for
- 13 global regions
- CPU & GPU compute
- Self-healing orchestration
- Zero-trust security
- OpenAI-compatible API
- Sovereign & air-gapped options
GPU Compute
Why teams choose it.
B200, H200, H100, A100, L40S and RTX 4090 GPUs for training and heavy inference, prepaid by the hour.
Every class of accelerator
Match each workload to the right GPU, from frontier training on B200 to cost-efficient inference on L40S and RTX 4090.
Prepaid, never wasted
Paid time only counts while your cluster is running, and any unused hours return to your compute balance.
Capabilities
Compute that works the way your teams do.
Every capability is available through the dashboard, the API and the AI ops copilot.
Frontier accelerators
B200 192GB and H200 141GB for the largest models, with H100, A100, L40S and RTX 4090 for everything else, all available from the same console and API.
Multi-node clusters
High-bandwidth interconnects for distributed training and large-batch inference.
Paired with CPU compute
Keep data prep, retrieval and orchestration on CPU instances on the same fabric, so GPUs spend their time on the work only they can do.
Fast attached storage
NVMe scratch and persistent volumes that follow your workloads.
Budgets and quotas
Per-project spend limits and alerts keep experimentation safe.
Self-healing nodes
Unhealthy GPUs are drained and replaced automatically, and checkpointed jobs resume on healthy capacity without anyone filing a ticket.
- GPU classes
- 6
- Max memory per GPU
- 192 GB
- Global regions
- 13
- Unused time returned
- 100%
Figures describe platform capabilities and service-level targets. Contracted commitments are set out in our Service Level Agreement.
Global cloud fabric
GPU Compute, everywhere you operate.
Available across 13 regions on six continents, connected by a private, encrypted fabric with automatic failover and per-region data residency.
- US East · Ashburn
- US West · Hillsboro
- Canada Central · Montréal
- South America · São Paulo
- EU West · Dublin
- EU Central · Frankfurt
- EU North · Stockholm
- Middle East · Dubai
- Africa South · Johannesburg
- Asia South · Mumbai
- Asia Southeast · Singapore
- Asia Northeast · Tokyo
- Australia East · Sydney
Use cases
What teams build with GPU Compute.
Starting points that deliver value in days.
Model training
Pre-train and post-train models on B200 and H200 clusters.
Fine-tuning
Adapt open models to your domain in hours on a single node.
High-throughput inference
Serve models cost-effectively on L40S and RTX 4090.
Rendering and simulation
Burst for VFX, digital twins and physics simulation.
Scientific computing
Accelerate genomics, chemistry and climate research.
Notebooks and dev boxes
Give every engineer a GPU workstation in seconds.
Security & compliance
Secure at every layer.
Zero-trust access, encryption everywhere and immutable audit trails are standard on every product, on every plan.
- Single-tenant GPU instances with isolated networking and encrypted volumes.
- Encryption at rest and in transit, with customer-managed keys available.
- SSO, role-based access control and scoped API keys.
- Your data is never used to train shared models.
- Built to align with SOC 2, ISO 27001, HIPAA and FedRAMP-moderate control frameworks.
Zero-trust by default
Every request authenticated, authorised and encrypted.
Built to align with control frameworks
- SOC 2
- ISO 27001
- HIPAA
- GDPR
- FedRAMP Moderate
- PCI DSS
AES-256
at rest
TLS 1.3
in transit
Immutable
audit trail
FAQ
Frequently asked questions
Straight answers from the team that runs the fabric. Still curious? Our architects reply fast.
Which GPUs are available?
NVIDIA B200, H200, H100, A100 and L40S, plus RTX 4090, across our global regions. Current rates are listed on the Pricing page.
How do I pay for GPU compute?
GPU time is prepaid. Choose your GPUs, how many and how many hours, then pay upfront or from your compute balance. Paid time only counts while your cluster is running, unused time returns to your compute balance, and optional auto-renew keeps long jobs going.
Can I reserve capacity?
Scale and Sovereign customers can reserve dedicated capacity for predictable workloads. Talk to us about your needs.
What happens if a GPU fails mid-job?
Self-healing orchestration detects the fault, drains the node and reschedules your job onto healthy capacity, resuming from your latest checkpoint.
Your next GPU cluster is minutes away.
Start with $600 in launch credit and 20M free inference tokens, double the typical big-cloud free trial. No commitment, live in minutes.
Double the typical big-cloud free trial · Per-second metering · Live in minutes