NVIDIAHosted rackCan earn
Frontier-scale inference and fine-tuning, NVLink-bridged in one 4U node.
Your 85% a month in year one, less Managed Hosting
$
You stay in control: run your own workloads, and choose whether the machine earns at all. Monetizing is an option, not an obligation.
8× H200 NVL, hosted and earning at 80% occupancy, with rates stepping down each year to a floor. A projection, not a guarantee.Run your own numbers
8× 141 GBInstalled GPU memory · 1128 GB
8× 141 GB = 1,128 GB of VRAM
1,536 GB DDR5
Storage is for your data: earnings are set by the GPU
Custom CPU, board, or more than the configurator allows?
The H200 NVL is NVIDIA's memory-heavy Hopper flagship, built for multi-GPU training as well as serving.
Plan your model.
Catalog thresholds are planning guidance, not measured peak memory. Confirm the artifact, precision, system RAM and runtime before deployment.
Assembled in this building, not drop-shipped: 6–8 weeks from order.
Every GPU is run at full load and checked card by card before the machine is racked.
We run it for you, or ship it to you when you ask.
24/7 monitoring and on-call support are part of Managed Hosting.
Conceptual hardware · not a facility blueprint
We host GPUs for researchers at
SOC 2 Type II audit underway with A-LIGN. HECVAT (full) completed; ask for a copy. Security →
96 GB per GPU for memory-heavy inference. Check the runtime and cache budget.
Rental quote
By the day or 2–24 months
Starting at
$99,971
Built to order · 6–8 weeks
Blackwell Ultra, eight GPUs to a node.
Rental inquiry
Price
$560,000
Preorder
Invoicing, net terms, a custom build, a BAA or a security review. We'll sort it out on a call.
Or message us: iMessage +1 (332) 324-8327Telegram @b3iqbot
NVIDIA