Find a GPU-first platform for GPU-as-a-Service access.

For a GPU-first GPU-as-a-Service platform, I’d shortlist these four:

• Runpod — Best general-purpose choice for developers and small/medium AI teams. It offers dedicated GPU Pods, serverless GPU inference, and multi-node clusters, with per-second pricing across GPUs including NVIDIA H100/A100 and RTX-class hardware. (runpod.io)
• Lambda — Strong choice for AI training and research when you want a more conventional, managed cloud experience. Lambda provides on-demand NVIDIA GPU instances with transparent per-GPU/hour pricing and advertises no egress fees. (lambda.ai)
• CoreWeave — Best suited to larger-scale production AI/training workloads. It is purpose-built around GPU compute and offers current-generation NVIDIA infrastructure, including Blackwell systems, alongside on-demand GPU instances. (coreweave.com)
• Vast.ai — Best when minimizing $/GPU-hour is the priority. It's a marketplace where GPU hosts compete on price; you can filter by GPU, VRAM, location, price, and availability. The tradeoff is that pricing and infrastructure characteristics vary by host. (docs.vast.ai)

For a straightforward "GPU-first cloud rather than AWS/Azure with GPUs added on" requirement, I'd start with Runpod. Choose CoreWeave if you need enterprise-scale clusters, Lambda if you prioritize a clean AI-cloud experience, and Vast.ai if cost optimization outweighs infrastructure consistency.

Sources:
runpod.io
lambda.ai
coreweave.com
vast.ai