New B200 spot capacity is live in US East from $1.69 per GPU-hour. See availability
About

We sell GPU time that would otherwise sit idle.

That is the whole business. Capacity that is installed, powered and unsold is worth more rented cheaply than left dark — so we rent it cheaply, publish the price, and tell you exactly when we may take it back.

Launched
Sept 2026Private beta, 47 GPU models on day one
Regions
3Virginia · Oregon · Frankfurt
Price rule
−20%Against the cheapest of 137 providers, enforced by a script
Sales team
NoneThe engineers who run the nodes answer the tickets
The model

Why it is cheaper, in three sentences.

We buy capacity, we do not build it

The hardware already exists, in facilities that already have power and network. We buy time on it in blocks, which costs a fraction of what it costs to finance a data centre — and none of that financing is passed on to you.

We price against the market, not against a margin target

Every tier of every model is priced at or below 80% of the lowest published price we can find for the same GPU. The rule is a line of code in the script that publishes the catalogue, not a promise in a slide.

We sell the interruption honestly

Spot is 30–53% below on-demand because we may take the GPU back. So we say when, how much warning you get, what survives, and what the reclaim rate has been over the last 30 days — per model.

How we work

Four rules we hold ourselves to.

Numbers before adjectives
Every claim on this site is a figure you can check: a price, a count, a rate, a date. Where a figure does not exist, the sentence does not exist either.The specifications on each model page link to the vendor datasheet. The reclaim rates are per model. The price comparison names the index and its date.
No fake proof
No invented customer logos, no testimonials nobody wrote, no compliance badge we have not earned, no "trusted by thousands" while we are weeks old.We would rather look small than look fake. You can tell the difference, and so can we.
The bill is two lines
Compute per started minute, storage per GB-month. No egress fees, no support tier, no reservation fee, no minimum spend, no surprise at the end of the month.If you ever see a third line, it is a bug and we will refund it.
Support by the people who run it
Tickets are answered by engineers with access to the nodes, not by a first-line queue reading a script.Target: a first reply within 12 hours, and within 2 hours when an instance is unreachable or a payment has not landed.

What we are not

read this before you commit

We launched in September 2026. That means:

  • No decade of uptime history to point at. The status page starts the day we did.
  • No SOC 2 or ISO 27001 certificate. We say what we actually do on the trust page instead of buying a badge.
  • No enterprise account management, no procurement portal, no golf.
  • No promise that spot capacity is always available for the model you want, in the region you want, at the hour you want.

What you can do instead of trusting us: add $40, run one job, and judge. Nothing renews, nothing locks in, and your disks are yours to delete.

Where it runs

3 regions, Tier III facilities, one tenant per virtual machine.

Spot and on-demand share the same nodes in the same rooms. Prices are identical everywhere; the region is about where your data sits, not about what you pay.

US EastVirginia, United States us-east
Tier III facility100 Gbps per node
US WestOregon, United States us-west
Tier III facility100 Gbps per node
EU CentralFrankfurt, Germany eu-central
Tier III facility100 Gbps per node

Isolation

Your GPU is passed through to your own KVM virtual machine. No shared kernel, no container on someone else's host, no neighbours inside your machine.

How isolation works

Data

Persistent volumes are encrypted at rest. SSH and the API are the only doors in, both over TLS, and nothing you store is read by us.

Privacy policy

Payments

Credit is prepaid in cryptocurrency and confirmed on-chain. We never see a card number, because there is no card to see.

How billing works
Who you are talking to

A small team of engineers, and no layer between you and them.

There is no sales department to get past and no account manager to schedule a call with. A question about a model that will not fit, a payment that has not landed or a node that is misbehaving reaches someone who can open a terminal on it.

We are deliberately small: the money that a sales floor would cost is the money that makes the price 20% lower than everyone else's.

Reaching us

Before you have an account
The contact formCapacity questions, pricing, technical fit, press. A first reply within 12 hours.
With an account
A ticket in the consoleAttached to your instances and payments, which is what makes an answer fast.
Something is broken
The status pageLive component state and incident history, updated from the system, not by hand.
Security or abuse
The abuse formVulnerability reports and misuse of our IP space. See the disclosure policy.
Get started

Judge us on one job, not on a page about ourselves.

Add $40, run something real on an H100 for $0.55 an hour, and decide.

Billed per minute from the moment the instance is reachable. Minimum credit $40, no subscription.