Managed infrastructure.
Hardware you don't share.
Dedicated servers or managed Kubernetes — without hypervisors, shared tenancy, or unpredictable cloud pricing.
64 physical cores, 512 GB RAM, 12 TB NVMe per node. Everything included.
Early access pricing available to charter applicants.
Three things cloud doesn't give you.
Dedicated hardware
No hypervisor. No VMs. No shared tenancy. Your server, your cores, your performance floor — not what's left after other tenants run.
Storage included
12 TB NVMe per node, included at every tier. No separate storage billing, no egress surprises.
Bandwidth included
10 TB outbound per node. No per-GB fees. The number on the spec sheet is the number you get.
Infrastructure we built for ourselves.
Smelt Cloud is a software company and applied AI lab that needed serious compute for its own products. We built proprietary infrastructure to run them — and realized the infrastructure itself was a product worth offering.
You get the same platform we run our own workloads on: dedicated AMD EPYC hardware, immersion-cooled, with full root access or managed Kubernetes on single-tenant infrastructure.
Deployment 001 — South Florida
Charter members lock in a permanent rate discount at this site and every future Smelt Cloud site. Available to early applicants only.
Not ready to commit? Join the waitlist → — free, no payment required.
Questions, answered.
What processors do you use?
Enterprise-grade AMD EPYC multi-core processors. We sell a guaranteed performance floor — 64 physical cores, 512 GB RAM, AMD EPYC — rather than advertising specific SKUs. This lets us source the best-value hardware and guarantee your performance floor regardless of generation.
Where are your servers located?
MIA-1 is our South Florida site (South Florida). Additional sites are on the roadmap — each site’s charter prepayments fund the next buildout.
How is this so much cheaper than cloud?
We’re not a trillion-dollar company with external investors demanding astronomical margins. We’re a software company that got tired of cloud fees, so we built our own infrastructure — and realized other teams were in the same position. We own every layer: hardware acquisition, immersion cooling, facility, and software stack. No landlord markup, no hypervisor tax, no egress fees. We pass the structural savings directly to you.
What network speeds do you offer?
25 Gbps internal per node — NIC speed between nodes on the switch fabric, verifiable with iperf3. Internet upstream is shared and rate-limited per server. Dedicated uplinks are available for customers who need guaranteed external bandwidth. Internal backbone is 25–100 Gbps between nodes.
What access options are available?
Metal — Flatcar Container Linux with full root access and IPMI out-of-band management. Your server, you manage it.
Orchestrated — A fully single-tenant RKE2 Kubernetes cluster with its own dedicated control plane. Not shared infrastructure. Managed upgrades, Prometheus + Grafana monitoring, and ArgoCD GitOps available as add-ons.
What’s the SLA?
99.9% uptime SLA backed by service credits. Charter members also get direct engineering access during the buildout period and IPMI for out-of-band management at any time.
Do you offer GPU nodes?
Not yet. CPU compute (the Ingot S1) is what’s available now. GPU nodes are on the roadmap. If you need GPU inference or training capacity, note it on your application — we’re building the GPU waitlist and will prioritize based on demand.
When is Smelt Cloud NOT the right choice?
We’d rather tell you upfront than waste your time:
- Highly bursty workloads (0 to 10,000 req/s in seconds) — dedicated hardware can’t autoscale. You pay for peak capacity 24/7. For burst, keep cloud autoscaling and use Smelt for your steady-state baseline.
- 10+ global regions today — we’re launching with MIA-1. Use Cloudflare Workers/Pages for edge now; migrate origin compute as we expand.
- HIPAA, PCI DSS, or FedRAMP required now — SOC 2 and compliance certifications are on the roadmap, not at launch. If you need them today, AWS is the right answer.
- Total workload under $500/mo — the Ingot S1 starts at $975/mo. If you’re running a single small app, we’re oversized for you. Hetzner or DigitalOcean are better fits.
If none of those apply, you’re probably a good fit. Apply and we’ll confirm.