Have idle H100/H200s?
We rent spare capacity from operators to serve Qwen 3.8‑27B. You run our serving image; we send traffic and pay per GPU‑hour.
- H100 80GB or H200 141GB, 2+ GPUs per node with NVLink
- ≥ 10 Gbit uplink, public IPv4, Linux with recent NVIDIA drivers
- Predictable idle windows — at least ~40 h/week — or fully dedicated nodes
- Paid monthly in USD; no minimum term
Don't have exactly that? Tell us anyway; we're also curious about MI300X and B200.