Platform · Fast-moving · Intermediate

GPU cloud

Rented accelerator capacity for training and inference, from hyperscalers and specialist providers.

What GPU cloud is

Renting avoids capital expenditure and long lead times, at a higher per-hour cost than owning. Specialist providers often undercut hyperscalers for raw compute without the surrounding services.

How it works

Capacity is booked on demand, by reservation or on interruptible spot terms; workloads must checkpoint frequently to survive preemption.

Why it matters

Compute procurement strategy materially changes the unit economics of an AI product.

Common uses

→Training runs
→Burst inference capacity
→Fine-tuning jobs

Strengths

✓No capital outlay
✓Elastic scale

Watch for

✓Expensive at sustained utilisation
✓Availability constraints on top hardware

Continue exploring

AI Compute GPU for AI AI Infrastructure Inference