HPC-AI Cloud rents B200, H200, H100, and RTX GPUs by the minute for AI training and inference. GPU rates run from $0.48 to $7.50 per GPU hour.
Researched and fact-checked by Bartu Cavusoglu · Updated October 26, 2026
HPC-AI Cloud, available at HPC-AI.COM, is a GPU cloud run by HPC AI Technology Pte. Ltd. of Singapore. It rents NVIDIA GPU servers for AI training, fine-tuning, and inference, billed per minute. The pricing page lists B200, H200, H100, RTX-4090, and RTX-5090 machines, and the navigation also features B300 SXM6.
Beyond raw GPU rental, the platform includes a Model APIs product marked New, a fine-tuning service, reserved clusters, batch jobs, CPU instances, and high-speed storage. The company says its core infrastructure, Colossal-AI, already powers leading enterprises, and that more than 10,000 researchers and engineers use it. These figures are the company's own claims.
Users sign up, open the console, and launch an on-demand instance on a chosen GPU type. The site describes dedicated full-machine rentals with no long-term commitment. The pricing page notes that only full machines with 8 cards are supported, to keep performance at its best. Prices are listed per GPU per hour.
Teams that prefer not to manage servers can use the Fine-Tuning SDK, a Training as a Service feature where the customer brings the code and the platform handles scaling. Reserved clusters can be rented as bare metal or cloud deployments, and a set of APIs lets teams automate jobs and manage GPU resources.
The core feature is on-demand GPU rental across five hardware types, with per-minute billing. Batch jobs provide a cheaper B200 option for work that can wait. CPU instances with 8 vCPUs and 18 GB RAM handle setup and data processing, while high-speed storage is described as low-latency and tuned for high-throughput AI work.
On the software side, the Fine-Tuning SDK and Model APIs cover teams that want hosted training or inference. The site also describes a pre-configured AI pipeline, APIs for automation, and a 99% uptime claim. Membership plans add monthly credits that refill every 30 days, with up to 12% extra credits.
HPC-AI Cloud is a GPU cloud from HPC AI Technology Pte. Ltd. that rents NVIDIA GPU servers for AI training, fine-tuning, and inference. It also offers model APIs, a fine-tuning SDK, reserved clusters, and batch jobs.
On-demand GPUs start from $0.48 per GPU hour for an RTX-4090 and from $7.50 for a B200. H200 starts at $2.68 and H100 at $2.42 per GPU hour. CPU instances cost $0.03 per CPU hour.
The pricing page lists B200 SXM, H200 SXM, H100 SXM, RTX-4090, and RTX-5090. The site navigation also features B300 SXM6.
The homepage says full machines are billed per minute, and prices are shown per GPU per hour. Machines are rented as full servers with 8 cards only.
The source pages do not mention a free plan or free trial. They show pay-as-you-go rates and paid membership plans.
Basic costs $1,000 per month and returns $1,080 in credits from the second month. Pro costs $5,000 for $5,500 in credits, and Enterprise costs $10,000 for $11,200. The first month returns slightly less in each plan.
Yes. Reserved GPU clusters can be rented in bare metal or cloud deployment. Pricing for H200, B200, RTX-4090, and RTX-5090 clusters is available through sales.
It suits small AI teams that need premium GPUs without long commitments, since billing is per minute. The 8-card full-machine requirement can be a barrier for very small workloads.