1–8 GPUs by the week
NVIDIA B300s on NVLink, booked by the week, with PyTorch, NeMo, vLLM or CUDA ready to run. Billed per GPU-hour used.
Get started
Request access to B300 GPUs.
Tell us how many GPUs you need and for how long — 1 to 8 B300s by the
week, or a dedicated node for months — and what
you're running. We'll reply within one business day with availability and a
quote.
Every booking comes with PyTorch, NeMo, vLLM and CUDA images, encrypted
storage with a verifiable erasure certificate, and API access. Agent sandboxes
are in early access — ask to join.
What we offer
Dedicated nodes
A whole DGX B300 — eight GPUs, bare metal, your own operating system and network — for months or years at a fixed monthly rate.
Private inference
Serve open models on your own GPUs with the vLLM image — an OpenAI-compatible endpoint whose data stays in Sydney.
Agent sandboxes (early access)
Isolated, disposable environments for running AI agents' code. Confidential computing (Intel TDX + GPU CC) is being commissioned and not yet enabled.
Request access
A few details about the workload help us come back with the right configuration and price.