Capacity for three kinds of customer, and how an engagement proceeds.
Liquidata serves AI developers and enterprises that need whole machines, regulated and public-sector buyers whose data cannot leave their jurisdiction, and research organizations whose terms follow grant and program cycles. The capacity is the same in each case; what differs is the contract, the controls and the schedule.
Dedicated and managed clusters
Whole Vera Rubin NVL72 racks and multi-rack clusters on commitments of 12 to 36 months, designed to the workload, delivered with fabric and storage, and run by Liquidata around the clock. A dedicated cluster is the right unit when a workload needs the full NVLink domain of a rack, or several racks joined by the scale-out fabric, rather than a share of a machine.
Private deployments under local control
Capacity that stays in-country on a partner site, inside the customer's security boundary, with Liquidata operating under the customer's controls. Built for governments, research institutions and regulated industries whose rules, among them NIS2 and DORA in the European Union, require control over where data and models reside and who operates the systems that hold them [1].
Research Capacity
Clusters for universities and research organizations, including the public research partners of the sites Liquidata deploys on, with terms that fit grant and program cycles rather than commercial commitment lengths.
Which workloads can follow the power, and which cannot.
A training run, a batch inference job or an evaluation suite can be paused and resumed, so it can be scheduled around renewable output, the state of a storage station and electricity prices at sites that have their own generation and storage. A latency-sensitive service cannot, because its users are waiting. Liquidata therefore offers two classes of capacity on such sites: energy-aware capacity, priced to reflect the flexibility the customer gives up, and standard capacity with service levels untouched. The customer decides which of its workloads qualify, and the scheduling software applies that decision; nothing is moved without the customer's rule allowing it.
How an engagement proceeds, from scope to operation.
Four steps, in a fixed order. The order exists so that no hardware is ordered without a customer and no customer is promised a date the site cannot meet. The step names below describe the sequence; the terms of each engagement are provided to customers under confidentiality.
Workload and size
Racks, fabric, storage, region and term, worked out with Liquidata's engineers against the customer's workloads rather than a catalogue.
Commitment for a term
Priced for the term and the module, tied to the site's delivery date, so that the customer knows the date it is contracting against [2].
Module ordered and built
Ordered once the capacity is under contract and the site's energization dates are in a signed connection agreement, then built and delivered against confirmed power.
Operated as a service
Acceptance against the workload, then operations, service levels and support by Liquidata as the provider of record for the life of the commitment.
What the figures on this page rest on.
Manufacturer specifications are as published; regulatory figures are as adopted or proposed on the dates given; estimates are identified as estimates. Where a figure is Liquidata's own design assumption, the text says so.
- Directive (EU) 2022/2555 (NIS2) and Regulation (EU) 2022/2554 (DORA).
- Commitment lengths and the order of contracting are described on the platform page; the detailed terms of an engagement are provided to customers under confidentiality.