Large GPU deals stall on the down payment: the buyer won't prepay a year of capacity, the owner won't carry the buyer's risk. Paralon sits between them. The owner connects the machines, we verify and reserve them for one buyer, the buyer pays by the minute, and the invoicing on both sides runs through us.
Owner
Datacenter GPU servers
connected with one command
Paralon
the layer the deal runs on
Buyer
Your workload
pays only for the minutes it uses
The owner
Wants a commitment, and a large share of it paid in advance.
Idle hardware is expensive to own, and a buyer who stops paying leaves it idle.
The buyer
Wants to pay for capacity as it uses it, not a year ahead.
Often a young company: capital tied up in prepaid capacity is capital not spent on the product.
Paralon
Every minute is paid from credits already in the account.
The owner is never exposed to the buyer's credit, and the buyer never prepays more than a rolling balance agreed in the deal.
How it works
Six steps, one layer
We are not a reseller holding capacity of our own, and not a broker who makes an introduction and steps away. We are the layer the deal runs on.
The owner connects the machines
Every card is verified
The capacity is reserved for one buyer
The buyer pays by the minute
The owner is paid from what is used
Invoicing on both sides runs through us
Who it is for
Owners, buyers and the brokers between them
Earn on capacity without selling it at a discount
- A price agreed with the buyer for the hours your machines work
- Paid from prepaid balances: no exposure to the buyer's credit
- The machines stay yours; capacity outside the reservation can serve the open network
- A reputation built from clean rentals, which buyers choose by
Reserved capacity, paid as you use it
- No down payment and no long contract to finance
- Billed to the minute, and nothing for time a machine is down
- Every card measured before it serves, every node with a record
- Rentals by hand or from code through the Rental API
Run your capacity deals on Paralon
- Bring the owner and the buyer; we run the layer underneath
- Access, verification, metering, payments and invoicing in one place
- One arrangement, the same for every deal you bring
- Talk to us about how it fits your own customers
Trust
Why the deal can run through us
Measured, not declared
Memory bandwidth and compute are measured on every card before it serves, and again after a reboot or a driver change.
No billing for downtime
If a node stops answering, billing stops after five minutes. The owner is alerted, and host faults count against the node's reputation.
Isolated rentals
A rental runs in its own container and cannot reach the owner's private networks.
One arrangement, three parties
The reservation, its price and its terms are agreed between the three of us when it is set up, and the platform enforces it from there.
Best fit
Workloads that run within one server
Inference at scale, fine-tuning, evaluation, batch and synthetic data jobs. A rental is a container on one machine with up to all of its GPUs, reached through Jupyter Lab with a terminal, or started and stopped from code through the Rental API.
Reserved capacity, answered
Can I reserve GPU capacity without paying upfront?
Yes. There is no down payment. You top up credits and they are spent per minute on the nodes reserved for you. The usual arrangement is a rolling balance, for example a month of expected usage kept in credits, instead of prepaying a quarter or a year.
Who owns the GPUs?
The owner keeps them. Paralon does not buy or hold the hardware: we run the layer between owner and buyer, which is access, verification, metering, payments and invoicing.
Can anyone else rent capacity reserved for me?
No. Reserved nodes are visible and rentable only by the entity they are reserved for. Nobody else on the network can see or take them.
Who sets the price of reserved capacity?
The owner and the buyer agree it between them when the reservation is set up, per GPU-hour, and Paralon bills it by the minute. Nothing ties a reservation to a list price.
How is the GPU owner paid?
The owner's share of every billed minute, agreed when the reservation is set up, goes to their earnings balance as it accrues and is paid out from there. The owner's side is invoiced through Paralon for what their machines earned.
Which GPUs can be reserved?
Datacenter and workstation GPUs such as B300, B200, H200, H100, A100 and RTX PRO 6000, whatever the owner connects. Every card is measured for memory bandwidth and compute before it serves.
What happens if a reserved machine goes down?
If a node stops answering, billing stops after five minutes, so the buyer does not pay for downtime. The owner is alerted, and host faults count against the node's reputation.
What access does the buyer get on a reserved node?
A container on the machine with up to all of its GPUs, reached through a Jupyter Lab environment with a terminal. Rentals can also be started, listed and stopped from code through the Rental API.
I broker GPU capacity. Can I run my deals on Paralon?
Yes. Bring the owner and the buyer; Paralon runs access, verification, metering, payments and invoicing under the deal. Write to [email protected].