GPU instances

Dedicated GPUs for AI agents

Serve models, fine-tune and build embeddings next to your agents on dedicated NVIDIA GPUs, created and managed over the same REST API and MCP server as every other CircuitNotion Cloud instance.

Join the GPU waitlist See sizes and plans

GPU instances are coming waitlist open

No GPUs are online yet. The waitlist decides which ones we bring online first, so tell us what your agents need.

What a GPU instance gives you

Never time-shared

Each GPU, or MIG slice, belongs to one instance for its whole life. No other workload shares its memory or compute.

Same API, same MCP tools

Agents create a GPU instance with create_instance and a gpu-* size, then run commands and deploy like on any other server.

CUDA ready

Start from the Ubuntu 24.04 + CUDA 12.8 template or bring your own image. nvidia-smi works inside.

GPU metrics over the API

Utilization, memory, temperature and power for every GPU, from the dashboard or GET /instances/{id}/metrics.

Tell us what you need

The waitlist decides which GPUs we bring online first. Pick a GPU (RTX 4090 / 5090 class, L4 / L40S, A100, H100 / H200, or any NVIDIA GPU), how many, what you will run and when you need it. We contact you as soon as it is ready. Agents can join for you with an API key:

REST · join the GPU waitlist

curl -X PUT "https://cloud.circuitnotion.com/api/v1/gpu-waitlist" \
  -H "Authorization: Bearer $CNC_API_KEY" -H "Content-Type: application/json" \
  -d '{"gpu":"h100","gpu_count":1,"use_case":"inference","timeframe":"month","notes":"Llama 70B with vLLM"}'

GPU questions

When will more GPUs be available?

We add GPUs based on the waitlist: the GPUs people ask for most come first. Join it and we will contact you when the GPUs you asked for are ready.

Are GPUs shared with other customers?

No. An instance gets whole GPUs or a MIG slice of its own and keeps them until it is deleted, also while it is stopped.

How are GPU plans priced?

Like every plan: monthly, 30 days per payment with MoMo or card, with no automatic renewal. The budget you give on the waitlist helps us set prices.

Can my agent use GPU instances?

Yes. GPU sizes appear in GET /api/v1/templates and the list_templates MCP tool, and an agent creates one like any other instance. When no GPU is free, the error's hint points it to the waitlist.

Get on the GPU waitlist

Join the GPU waitlist