About the tool
CloudGPU is a GPU cloud and model API service for developers. You can rent a raw GPU with root SSH and JupyterLab, or deploy a template such as ComfyUI, Ollama, LLaMA-Factory or MiniMax H3 and get an HTTP endpoint without setting anything up. Cards range from an RTX 2080 at $0.15 an hour to RTX 5090, A100, H100, B200 and GB300 machines, some of them hosted on third-party data centres and peer-to-peer GPU networks. Billing is per minute or per hour, and the quote shows the minimum charge before you confirm. The AI API is one OpenAI-compatible endpoint for chat, image, speech and transcription models, including Claude, DeepSeek, GLM, Kimi and MiniMax. CloudGPU lists its Claude rates at about half the official list price and states that it is not the official Claude API. A remote MCP server and a local package let Cursor, Claude Code and other AI agents check a task, show a quote and run GPU jobs within a budget the user approves. Everything is paid from one prepaid US dollar balance, topped up from $5 by card, Alipay, Google Pay, Apple Pay, USDT on TRC-20 or bank transfer. The balance does not expire, but unused credit is not refunded in cash. The service is in public beta with no uptime guarantee, is not offered in mainland China, and the website does not name the company behind it.
Key use cases
Building AI-native applications
Create your own AI products, from copilots and assistants to intelligent workflows and dynamic interfaces.
API & developer readiness
CloudGPU's AI API is one OpenAI-compatible endpoint at https://cloudgpu.app/v1 for chat, image, speech and transcription models such as Claude, DeepSeek, GLM and Kimi, billed per token from the shared balance. It works with the OpenAI SDK, LangChain, Cursor and Cline by changing the base URL and key. A REST API and a remote MCP server with OAuth also let agents rent GPUs and run jobs within an approved budget.


