Reserve capacity.
On demand.
Backstop compute capacity for neoclouds.
Darkstack provides standby AI infrastructure to neoclouds and cloud providers, giving them guaranteed overflow capacity when deployments are delayed or existing sites reach capacity.
What we supply
Reserve capacity. On demand.
Reserved megawatts
Committed capacity held on standby against a reservation fee.
Drawn per GPU-hour
When you draw against your reserved GPU capacity, you pay for what you use.
Stay only as long as you need
When your own data center is delayed, full, or demand unexpectedly spikes, draw on Darkstack’s capacity.
Darkstack sells to neocloud providers, not end users. Customers reserve megawatts of capacity by paying a monthly standby fee. If their own data center is delayed, full, or demand unexpectedly spikes, they draw on Darkstack’s capacity. Once activated, they pay based on usage, such as per GPU-hour. The idea is to act almost like insurance for compute capacity: our customers do not necessarily want to use Darkstack permanently; they want to know it is there when their own infrastructure cannot meet demand.
Capacity
Sites available today, with no power constraint.
| Region | Status | Sites |
|---|---|---|
| Italy | Available | 2 |
| Norway | Available | 1 |
| United Kingdom | Available | 2 |
| Southeast Asia | Available | 1 |
| East-coast USA | Available | 1 |
| West-coast USA | Available | 1 |
Six regions available today
Performance
Time to first token.
- We run at the edge, by design
- Round-trip latency runs about 1 ms per 100 km of fibre. Serving a London user from off the UK coast rather than Northern Virginia saves roughly 70 ms before any compute begins.
- Reserved capacity removes the queue
- On a shared cluster a request waits for a free GPU slot before prefill even starts. A held megawatt begins work the moment the request lands.
- Density is the commercial edge
- ~896 GPUs per megawatt at 1.01 PUE against 700–800 in an air-cooled hall, roughly 1.2× the revenue from the same power.
About & contact
The neocloud behind the neoclouds.
Or reach out directly at hello@darkstack.com.
Investor materials
Technical diligence, in one document.
Darkstack technical data sheet
Capacity and regions, deployment model and timeline, reference compute configurations, power and thermal, capacity density, network fabric, latency and siting, and the service model. Every quantitative claim carries its basis.