Compute.cheap
A GPU compute marketplace offering NVIDIA H100 and H200 GPUs at cost-optimized rates, with both on-demand and spot pricing for AI training and inference workloads.
🔗 Visit Compute.cheap
Description
Renting the GPUs needed to train or run large AI models is expensive, and pricing at the big cloud providers is often padded to cover their own idle capacity and long-term contracts. Compute.cheap tries to strip that padding out by turning GPU rental into a real marketplace instead of a fixed price list.
Compute.cheap is a GPU compute marketplace where capacity owners compete on price for NVIDIA H100 and H200 GPUs, offered at $1.981 and $2.841 per GPU-hour on-demand respectively, or $1.090 and $1.685 per GPU-hour on the spot market. There are no long-term commitments — you pay only for active compute, by the GPU-hour, and the marketplace mechanism is meant to push pricing down over time as more capacity providers compete for the same demand.
💬 Our review
The short version: Compute.cheap's pitch is straightforward — undercut the big clouds on raw H100/H200 pricing by running a competitive marketplace instead of a fixed rate card — and on paper its numbers are genuinely competitive with other GPU-rental specialists.
Specialist GPU clouds like RunPod, Vast.ai, and Lambda Labs already compete hard on H100/H200 pricing, so Compute.cheap isn't creating a new category, it's entering a crowded one where the pitch is nearly identical across providers: no long-term contracts, pay by the GPU-hour, spot pricing for the price-sensitive. The marketplace model — providers competing on price rather than a single fixed rate card — is a real mechanism for keeping prices honest over time, but it also means reliability and support quality can vary more than with an established single-vendor provider. For a research team or startup chasing the lowest per-hour H100/H200 rate and comfortable with some variance in provider quality, it's worth comparing live rates here against RunPod and Vast.ai before committing; for production workloads where uptime guarantees matter more than shaving cents off the hourly rate, an established provider with a longer track record is the safer default.
💰 Pricing
📊 Global score
🤖 AI-enriched data
H100 à la demande 1,981 $/h ; H100 spot 1,090 $/h ; H200 à la demande 2,841 $/h ; H200 spot 1,685 $/h. Aucun engagement long terme.
Pros
Tarifs compétitifs sur H100/H200 vs clouds généralistes (AWS, GCP, Azure)
Aucun engagement long terme, facturation à l'heure GPU
Option spot pour réduire encore le coût
Mécanisme de marché qui pousse les prix vers le bas avec la concurrence des fournisseurs
Cons
Pas de palier gratuit, coût dès la première heure
Fiabilité et disponibilité dépendent des fournisseurs tiers du marketplace
Moins d'historique et de réputation que RunPod ou Lambda Labs