RAW Offers Dedicated AI Servers Through a Single API

Summary

RAW is an infrastructure service offering dedicated CPU and NVIDIA GPU servers for AI inference, training, fine-tuning, agents, and vector databases. The service provides root access, CUDA, NVMe storage, public IPs, unlimited bandwidth, and support for tools including vLLM, Ollama, PyTorch, Qdrant, and Milvus. Its API can create, resize, rebuild, and destroy servers, with the company advertising roughly three-second GPU provisioning and per-second billing. Listed prices include CPU instances from $8-$9 per month, a 20 GB VRAM GPU at $304 per month, and 96 GB VRAM configurations at $1,510 and $2,914 per month. RAW says servers are dedicated rather than time-sliced, with users retaining their weights on their own disks. The service lists five regions: Frankfurt, Dublin, Ashburn, Hillsboro, and Singapore, and advertises EU GPU availability for GDPR-oriented inference. RAW also claims zero egress charges and presents comparisons asserting lower costs than AWS, SageMaker, Bedrock, and other cloud providers; those savings are company claims based on the page's pricing comparisons. The page says the platform has SOC 2 Type II coverage, but provides no further certification details.