Substrate
noun
An underlying layer that supports complex systems.
Composable GPU infrastructure — configure compute by cores, VRAM, memory, and storage.
Configure your instance
16 cores · 48GB VRAM · 128GB RAM · 500GB SSD
~$8.82/hr
Trusted by teams at
How it works
Compute, assembled to your needs
Instead of renting fixed machines, configure exactly the resources your workload requires.
Choose cores
Select the number of GPU cores for your training or inference workload. Substrate allocates dedicated cores, never shared.
Add VRAM
Allocate video memory precisely for model weights, KV caches, and activation tensors. No memory tax from co-tenancy.
Add RAM
Set system memory for data loading, preprocessing, and host-side buffers. Scale independently from GPU resources.
Add storage
Attach NVMe SSD storage for datasets, checkpoints, and artifacts. Provisioned in seconds, billed per GB used.
Why it matters
Infrastructure without fixed shapes
No oversized instances
Only allocate the resources you need. No paying for idle cores or unused memory sitting in a fixed SKU.
Predictable performance
Dedicated cores and memory, not shared bursts. Consistent throughput for every job, every time.
Built for AI workloads
Training, inference, rendering, simulation. Infrastructure shaped for the demands of modern compute.
Per-second billing
Pay for exactly what you run. Billing stops the moment your job ends — no rounding to the nearest hour.
Instant teardown
Instances are fully cleaned and returned to the pool in seconds. No lingering costs, no stale state.
API + Terraform native
First-class Terraform provider and REST API. Integrate into any existing IaC pipeline in minutes.
< 0s
Provisioning time
From API call to SSH-ready
0%
Uptime SLA
Guaranteed availability
0+
GPU configurations
A100, H100, H200 available
Per-second
Per-second billing
No rounding, no minimums
Trust & compliance
Enterprise-grade security, built in
Compliance and encryption are foundational — not bolt-on features. Every layer of Substrate is built for teams handling sensitive data and regulated workloads.
SOC 2 Type II
Audited annually by a Big Four firm. Full audit report available under NDA.
HIPAA Compliant
BAA available on Pro & Enterprise. PHI-safe infrastructure for healthcare AI.
Zero Data Logging
We never inspect, log, or retain your workload data. Your models stay yours.
100% Renewable
Norwegian hydropower. PUE below 1.15. Natural fjord-water cooling.
“Your infrastructure should never be the weakest link in your security posture.”
Infrastructure
Built inside a mountain
Substrate runs on infrastructure housed in Lefdal Mine Datacenter — a former NATO mining facility carved into the mountains of western Norway. Powered by 100% renewable hydroelectric energy and cooled by fjord water, delivering a PUE below 1.15.
Location
Måløy, Norway
Power
100% renewable hydro
Cooling
Fjord water cooling
Security
Underground facility
Cross-section: Lefdal Mine Datacenter, Måløy, Norway
Developer experience
Infrastructure you can script
API-first. CLI-native. Define your compute with declarative configs and manage infrastructure as code.
resource "substrate_compute" "train" {
gpu_cores = 16
vram_gb = 48
ram_gb = 128
storage_gb = 500
}How we compare
Substrate vs. the status quo
| Feature | Substrate | AWS EC2 | GCP Compute | Lambda Cloud |
|---|---|---|---|---|
| GPU allocation | Exact core count | Fixed SKUs | Fixed SKUs | Fixed SKUs |
| VRAM configuration | Per-GB | Tied to instance | Tied to instance | Tied to instance |
| Billing granularity | Per-second | Per-second | Per-second | Per-hour |
| Provisioning time | < 30 seconds | Minutes | Minutes | Minutes |
| Data logging | None | CloudTrail | Cloud Audit | Varies |
| Compliance | SOC 2, HIPAA | SOC 2, HIPAA | SOC 2, HIPAA | SOC 2 |
Who it's for
Built for teams running serious workloads
ML engineers
Fine-tune models with precisely allocated GPU memory and compute.
AI startups
Scale from prototype to production without re-architecting infrastructure.
Research labs
Run experiments with reproducible, precisely configured compute environments.
Rendering studios
Allocate GPU cores for parallel rendering without fixed machine limits.
Simulation teams
Configure memory-heavy instances for physics, climate, or financial models.
FAQ
Frequently asked questions
Ready to build?
Get early access to composable GPU infrastructure. No commitments, no fixed contracts.
Free tier available · No credit card required