A complete, running GB200 cluster offered as a dedicated 5-year rental: 2× NVIDIA GB200 NVL72 racks (liquid cooled) with 144 Blackwell GPUs and 72 Grace CPUs, a full 72-GPU NVLink domain per rack validated in production, and 14 control/management servers. No build, no wait — the hardware is live in the Seattle area today, with handover roughly 30–45 days from signing. Pricing is $4.86/GPU-hr on a 60-month term with 40% prepaid at signing.
Storage and fabric are included: a VAST all-flash cluster with 1.16 PB usable, InfiniBand compute fabric on 6× NVIDIA QM9700 switches, 400G spine/leaf Ethernet with dual edge routers, and an out-of-band management network with LTE and cable backup. The facility runs direct liquid cooling with liquid-to-air sidecars (2× 200 kW), redundant 4-makes-3 power feeds, and UPS plus generator backup.
Managed operations are part of the deal: 24/7 NOC and SRE with on-site engineers, break/fix and RMA under OEM support, a software stack from bare metal to Kubernetes (Slurm, drivers, observability, patching), and an availability SLA with service credits.
Room to grow: the site can expand by 10 more GB300 NVL72 racks — 12 racks total including the 2× GB200 — so a buyer can land on production-validated capacity now and scale on the same floor. The public tile shows the verified headline specification; the complete data room — full spec sheet, exact location, counterparty, and contract terms — is shared once both sides agree to connect.
Every Ax3 listing is vetted by the broker desk before going live: asset existence, control of the capacity, and realistic timelines are confirmed directly with the owner.
