FP4 inference per NVL72
to priority allocation
Limited racks remaining
Q3 2026
Capacity available
Q4 2026
Limited racks remaining
Q4 2026
Capacity available
H1 2027
GB300 NVL72
72 Blackwell Ultra GPUs and 36 Grace CPUs fused into a single liquid-cooled rack-scale domain.
Quantum-2 InfiniBand
Non-blocking 400G/800G fabric with NVLink Switch for full-bandwidth all-to-all collectives.
High-perf storage
Parallel NVMe tiers delivering multi-TB/s throughput, scaled to your checkpoint cadence.
Liquid-cooled power
Direct-to-chip cooling with redundant power delivery for sustained dense training loads.
Dedicated tenancy
Single-tenant clusters with confidential computing and SOC 2 Type II controls.
99.9% SLA
Operated by CNEX with proactive node replacement and committed availability targets.
01
Plan your cluster
Walk through a guided wizard to size GPUs, region, fabric, storage, and term.
02
Model your budget
See live monthly and total contract estimates with discounts applied.
03
Sign your LOI
Generate a Letter of Intent and sign it electronically via Dropbox Sign.
04
Lock priority allocation
CNEX countersigns and holds your configuration while the definitive MSA is finalized.
