Compute Cheap
H100s at $2.04/HR, H200s at $3/HR
Last verified:
What is Compute Cheap?
Compute Cheap is a GPU compute marketplace specializing in low-cost NVIDIA H100 and H200 GPUs for AI training, inference, and research workloads. It operates on a marketplace model where capacity owners compete on pricing, and combines interruptible and reserved instance options to minimize costs.
Compute Cheap pricing
Pricing model: Freemium
H100: $1.15/GPU-hour (interruptible) or $1.19/GPU-hour (reserved); H200: $1.39/GPU-hour (interruptible) or $1.99/GPU-hour (reserved). Pricing updated Sep 3, 2026.
Compute Cheap pros
- Extremely competitive pricing ($1.15/hr for H100 interruptible, historically among the lowest available)
- Marketplace model ensures prices are set by competition rather than fixed tiers
- Both interruptible and reserved options available for different reliability/cost tradeoffs
- Simplified infrastructure focused on GPUs without enterprise service overhead
Compute Cheap cons
- Limited GPU selection (only H100 and H200; no other architectures mentioned)
- Interruptible instances lack guaranteed availability and can be terminated
- Sparse documentation on support, SLAs, compliance, or multi-tenancy security details
- Requires manual request/reservation process rather than instant provisioning
Frequently asked questions about Compute Cheap
Why is Compute Cheap cheaper than other providers?
The marketplace model lets capacity owners compete for workloads instead of using fixed pricing tiers, and idle infrastructure is monetized rather than left unused.
What's the difference between interruptible and reserved pricing?
Interruptible GPUs are cheaper but can be terminated; reserved GPUs cost slightly more but are guaranteed availability once booked.
How do I get started?
Use the 'Request Compute' option to specify your GPU and workload requirements; availability is confirmed before payment.