01
Inference demand.
Make suitable capacity available to teams serving models. Define runtime access, memory, network and service requirements before a placement.
Liquid Compute / Spot capacity
For operators, builders and holders of reserved capacity. Bring available GPUs to inference teams, resellers and burst workloads through the compute desk.
Define what is available, for how long, and when you need it back. Spot placements are agreed against real workload requirements.
Your capacity
GPU infrastructure
Site · Hardware · Fabric · Delivery
Idle hours or released capacity
An available window for another workload
Liquid Compute
Spot demand, through the desk
Match the block to inference, reseller or burst requirements, with availability and reclaim terms agreed up front.
Where spot capacity goes
01
Make suitable capacity available to teams serving models. Define runtime access, memory, network and service requirements before a placement.
02
Bring available blocks to buyers placing capacity with their own customers. Agree the delivery scope, commercial terms and responsibilities up front.
03
Offer capacity for a defined window when a buyer needs extra room. Match the workload’s duration and interruption tolerance to the capacity you can release.
How Liquid Compute works
A cluster does not have to be fully idle to have capacity to sell. Bring the blocks or hours you can release, together with the constraints that make a placement workable.
Share GPUs, quantity, fabric, region, available dates and the notice you need to reclaim capacity.
The desk reviews demand against the block, then coordinates the rate, duration, runtime requirements and delivery conditions.
The operator delivers the capacity. The buyer uses it within the agreed window, with usage, settlement and any reclaim handled under the placement terms.
The placement, in writing
Spot describes how capacity is made available. The placement still needs a clear scope, price basis, delivery window and operating responsibilities.
Operators, builders and holders of reserved GPU capacity with blocks or hours they can make available to other workloads.
Bring the newly available capacity to the desk as a spot opportunity. We review its technical fit and availability against replacement demand. A replacement buyer, utilization or revenue is not guaranteed.
Reclaim rights and notice are agreed before the placement. Some workloads can tolerate interruption; others need a fixed window. The buyer and capacity holder need to agree which applies.
Placements are coordinated through the desk. Share your capacity or workload, and the team confirms fit, availability and terms before a placement is agreed.
The operator continues to run the infrastructure and deliver the agreed service levels. The desk coordinates requirements, matching and commercial terms.
Start with your capacity
Share a block, a window of idle hours, or capacity released by an off-taker. The desk reviews the fit and follows up on the next step.
Looking for capacity? Bring your spot requirement.