TOOLS
Size your AI cluster. See your real TCO.
Workload → cluster → cost → power → carbon. In under 60 seconds.
Loading calculator...
Reference Clusters
Starter
- 2 x H100
- For: inference up to Llama 3 70B
- Audience: smaller teams, dev/staging workloads
$5.59 / hr
Scale
- 8 x H200
- For: production inference, multi-tenant serving
- Audience: prod-scale inference teams
$25.20 / hr
Train
- 8 x B200 NVLink
- For: fine-tuning and small training runs
- Audience: research and ML platform teams
$30.80 / hr
Frontier
- 32 x B200
- For: large training runs
- Audience: foundation model builders, large enterprises
$123.20 / hr
Frequently Asked Questions
Where do your carbon numbers come from?
It is a clean, lower-carbon, modeled estimate based on our blended mix of behind-the-meter solar and grid power. We calculate actual physical electrons rather than relying purely on offsets.
What's the catch on the cost savings?
We own our power generation and land, skipping the utility markup. Those savings are passed directly to our infrastructure rates.
Is this real NVIDIA silicon?
Yes. We deploy authentic, brand-new NVIDIA HGX nodes exactly to NVIDIA reference architecture specifications.
How do migrations from AWS, Azure, or GCP work?
Our clusters run standard Kubernetes, Slurm, and Docker. If it runs on a hyperscaler, it runs here with zero code changes.
Can I run this on a contract or month-to-month?
We offer standard 1-year and 3-year reserved instances for the lowest TCO, as well as flexible month-to-month terms.
Ready to size for real?
Talk to an engineer who can validate the math for your workload.
Book a callOr fill out a detailed form →