GPU Compute Pricing

We offer simple per-GPU pricing with no hidden fees. Prices include compute, memory, storage (up to 1TB NVMe per GPU), network transfer within Gibraltar, and standard support. You're billed hourly with monthly/annual discounts available. Unlike hyperscalers, we don't charge for API calls, data ingress, or inter-GPU communication.

Four factors: (1) Gibraltar's lower operational costs and favorable tax environment, (2) 100% renewable energy reducing power costs by 30%, (3) Efficient liquid cooling enabling higher utilization, (4) Direct-to-customer model without hyperscaler markup. We pass these savings directly to customers while maintaining enterprise-grade infrastructure.

Yes. 1-year commitments receive 10% additional discount, 3-year commitments get 20% off listed prices. For example, a reserved A100 80GB drops to £729/month (1-year) or £648/month (3-year). We also offer flexible reserved capacity—commit to spending level rather than specific GPU count for burstable workloads.

Minimal extras at Pelagos. GPU prices include 1TB NVMe, 20TB monthly egress, and standard support. Additional costs: Extra storage (£0.15/GB/month for NVMe, £0.03/GB/month for object storage), egress over 20TB (£5/TB vs. £75/TB on AWS), and optional premium support (£500-2,000/month depending on tier).

Coming Q3 2026. We're developing a spot market for unused capacity at 50-70% discounts. Unlike hyperscaler spot instances that can terminate anytime, ours guarantee minimum 1-hour runtime with 5-minute termination notice, making them viable for checkpoint-enabled training workloads. Join waitlist for early access.

Multi-GPU within single node: simple multiplication (8x A100 = £4,860/month). Multi-node clusters: add £200/month per node for high-speed interconnect (InfiniBand or RoCE) enabling efficient distributed training. No data transfer costs between nodes. We help optimize topology for your specific framework (PyTorch DDP, DeepSpeed, Megatron).