# Compute Cheap

> Compute Cheap offers the world's cheapest prepaid H100, H200, GB300, and B200 GPU-hours, available as reserved or interruptible capacity in the US and EU.

Compute Cheap is a GPU cloud provider based in San Francisco that positions itself as the lowest-cost source of frontier GPU compute. It offers prepaid H100, H200, GB300, and B200 GPU-hours on a reserved or interruptible basis, with capacity available in the United States and European Union. The service targets researchers, startups, and AI teams that need reliable frontier compute at radically lower prices than the major cloud providers.

## What It Is

Compute Cheap is a GPU-as-a-service platform focused exclusively on delivering frontier NVIDIA GPU capacity — H100, H200, GB300, and B200 — at the lowest available market price. Rather than a self-serve cloud console, it operates on a request-and-confirm model: customers specify their GPU type, mode (reserved or interruptible), and GPU-hour block, and Compute Cheap confirms availability before any payment is taken. The platform publishes a live pricing API and a public leaderboard comparing its prices against other GPU clouds.

## How the Reservation Model Works

The workflow has three steps:
- **Request**: Choose GPU type, reserved or interruptible mode, and the number of GPU-hours needed. The minimum reservation is 2,000 GPU-hours per request (roughly 8 GPUs for 10 days).
- **Confirm**: Compute Cheap verifies capacity availability before charging anything. Once confirmed, the customer prepays 50% of the block upfront.
- **Run**: GPUs are reserved and ready; the remaining 50% is billed as usage completes.

Launch sizes of 1×, 2×, 4×, and 8× GPUs are supported for H100 and H200.

## Reserved vs. Interruptible Capacity

Compute Cheap offers two pricing modes:
- **Reserved**: Capacity is held for the full block at a fixed rate. Suited for workloads that cannot tolerate interruption.
- **Interruptible**: Lower per-GPU-hour cost, but capacity can be reclaimed if demand spikes. Best for fault-tolerant training runs and batch jobs that use standard checkpointing.

The site notes that preemptions on interruptible instances are rare enough that normal checkpointing typically covers them, based on a published customer testimonial from Arcee AI's infrastructure team.

## Pricing Transparency and Market Comparison

Compute Cheap publishes a daily chart comparing its GPU-hour price against a market benchmark for each GPU type and mode over a rolling 31-day window. A public Pricing API delivers live prices as JSON. A leaderboard page ranks Compute Cheap's prices against other GPU cloud providers. The site states there is no separate platform fee layered on top of the GPU-hour rate — pricing is strictly hourly rate × GPU-hours.

## Compliance and Regions

Compute Cheap states it is HIPAA Ready, SOC 2 certified, and GDPR Ready. Capacity is currently available in the United States and European Union, with customers able to specify a preferred region in their request notes before any deposit is taken.

## Current Status

As of the homepage snapshot (dated references through September 2026), H100 SXM 80GB and H200 SXM 141GB capacity is available at published interruptible and reserved rates. GB300 and B200 capacity is listed as sold out. The company publishes a blog with pricing analysis and company updates, including a June 2026 post explaining its cost model and an August 2026 post ranking GPU cloud providers. Compute Cheap Inc. is incorporated and operating from San Francisco, California.

## Features
- H100, H200, GB300, and B200 GPU support
- Reserved and interruptible capacity modes
- Prepaid GPU-hour blocks with 50% upfront deposit
- Minimum 2,000 GPU-hour reservation
- 1×, 2×, 4×, and 8× GPU launch sizes
- US and EU regions
- Live Pricing API (JSON)
- Public leaderboard vs. other GPU clouds
- Daily market benchmark price chart
- No separate platform fee
- HIPAA Ready
- SOC 2 certified
- GDPR Ready
- Uptime SLA

## Integrations
nvidia-smi, PyTorch (torchrun)

## Platforms
WEB, API

## Pricing
Paid

## Links
- Website: https://compute.cheap
- Documentation: https://compute.cheap/docs/api/pricing
- EveryDev.ai: https://www.everydev.ai/tools/compute-cheap