Compute Cheap Inc.
Compute Cheap aims to put frontier GPU compute in the hands of researchers, startups, and open-source teams by offering reliable NVIDIA GPU capacity at radically lower prices. It combines competition among capacity owners, distributed supply across the US and EU, and a lean operating model with no platform fee.
At a Glance
- Researchers
- AI startups
- Frontier-model teams
- Open-source teams
- +3 more
AI Tools by Compute Cheap Inc.
(1)
Compute Cheap
Affordable GPU Cloud for AI Teams
Discussions
No discussions yet
Be the first to start a discussion about Compute Cheap Inc.
Latest News
Compute Cheap featured in a Hacker News Show HN launch: 'The cheapest GPU cloud – H100s at $2.04/HR, H200s at $3/HR.'
Published 'The cheapest GPU cloud providers in 2026, ranked,' comparing 19 providers and positioning Compute Cheap as the lowest published guaranteed-rate provider in the table.
Updated Terms of Service and Uptime SLA for the request-based GPU reservation service.
Published 'Why compute should be cheap,' explaining the company's mission, sourcing model, customer traction, and GPU pricing.
Products & Services
Request-based prepaid access to NVIDIA H100 SXM 80GB, H200 SXM 141GB, and B200 SXM 180GB capacity in 1-, 2-, 4-, or 8-GPU launch sizes across US and EU regions. Customers can choose reserved/non-preemptible or lower-priced interruptible/preemptible capacity; the minimum request is 2,000 GPU-hours.
A public, unauthenticated read-only JSON endpoint at https://compute.cheap/api/v1/pricing that publishes current offer prices, GPU models, capacity type, minimum reservation, deposit percentage, allocation status, and sold-out modes.
Market Position
Compute Cheap positions itself as a low-cost GPU cloud/neocloud and says it has the lowest published guaranteed rates in its August 15, 2026 comparison of 19 GPU clouds. Its differentiation is prepaid, request-only H100/H200/B200 capacity, distributed supply, competition among capacity owners, and no platform fee. Competitors identified in its comparison include Latitude.sh, SF Compute, Voltage Park, Vast.ai, Prime Intellect, Verda, RunPod, Together AI, Hyperbolic, Thunder Compute, Nebius, Modal, Lambda, AWS, CoreWeave, Google Cloud, and Microsoft Azure.
Founding Story
The company says it started from the observation that frontier AI labs can use capital and scale to make intelligence affordable, while the compute underneath remains expensive and inaccessible to most builders. Its initial vision was to make reliable frontier compute available to researchers, startups, and open-source teams at much lower prices by sourcing distributed capacity and shifting pricing power away from large clouds.
Business Model
Revenue Model
Prepaid GPU-hour reservations: customers request a GPU type, capacity mode, GPU-hour quantity, and start date; Compute Cheap confirms availability, collects 50% upfront, and bills the remaining 50% as usage completes. Reserved capacity is fixed-price; interruptible capacity is cheaper and may be reclaimed. The company states it does not add a platform fee.
Pricing Tiers
Preemptible; 2,000 GPU-hour minimum; request-only and availability-confirmed.
Non-preemptible reserved capacity with 99.9% monthly uptime commitment; 2,000 GPU-hour minimum.
Preemptible; 2,000 GPU-hour minimum; request-only and availability-confirmed.
Non-preemptible reserved capacity with 99.9% monthly uptime commitment; 2,000 GPU-hour minimum.
Preemptible; 2,000 GPU-hour minimum; limited availability.
Non-preemptible reserved capacity; 2,000 GPU-hour minimum; limited availability.
Target Markets
- Researchers
- AI startups
- Frontier-model teams
- Open-source teams
- AI companies
- Voice/audio AI companies
- Foundation-model and language-model training
- Model serving and inference
- Audio-model training and experimentation
- Generative-video model training and inference
- Fault-tolerant training and batch jobs using interruptible capacity
- Research, startup, and open-source AI workloads
- Runway
- ElevenLabs
- Midjourney
- fal