Design partner places are open
New beginnings, together.

Compute that knows which hall it's in.

Provision, meter and bill GPU and cloud tenants on the same rack, power and customer objects as everything else in the building. One invoice, one record, no second stack.

Talk to us →
Aire · Cloud

From tenant to invoice.

Six moves. The pod, the rack and the kilowatt are the same object the facilities team and finance see.

AireTenants / Halcyon AID

Compute, on the floor.

Pod 4 · Hall 3 · Contract HA-204

NOV 2026
GPUs9612 nodes
Utilisation87%Pod 4
Power404 kWof 420 kW
Pod 4 / GPU nodes12 nodes live
Committed power404 / 420 kW
✓61,440 GPU hours · on the tenant’s invoice↗
Usage metered into AyanaHA-204
Interactive preview · sample dataSelect a record to explore

01

Tenant on the record.

A cloud tenant is a customer object in DIL, with its contract in Ayana and its pods on real racks in Neev. Scope and roles follow from that, not from a second directory.

  • Customer object
  • Contract
  • Real racks
Tenant · Halcyon AI
POD 4 / PHYSICAL COMPUTEISOMETRIC VIEW
Liquid cooling · 31 °C loop inletGPU 01–02 · 16 GPUs · 89% utilisedGPU 07–08 · 16 GPUs · 85% utilisedGPU 03–04 · 16 GPUs · 87% utilisedFeeder 3 · 156 kW headroomLeaf D-07 · 8 ports freeGPU 09–10 · 16 GPUs · 88% utilisedGPU 05–06 · 16 GPUs · 88% utilisedGPU 11–12 · 16 GPUs · 87% utilisedGPU 01–02 · 16 GPUs · 89% utilised
GPU 01–02live
16 GPUs · 89% utilised

Two GPU nodes on the same physical rack. Their power, utilisation and metered hours roll up to Halcyon AI's contract HA-204.

Pod 4 · Halcyon AI · HA-204
Select a unit to see its shared record.Sample equipment view
Pod 4 · Hall 312 nodes · 96 GPUslive
Contract HA-204committed 420 kW
Network2 × 400G · VRF 204
Open requests1 · +4 nodescapacity check

02

Provision against capacity.

GPU nodes, VMs and bare metal are provisioned only where Neev shows power, cooling and network headroom. A request that the hall cannot carry is caught before a node is racked.

  • GPU
  • Bare metal
  • VM
  • Headroom check
Request · +4 GPU nodeschecked vs floor
TenantHalcyon AI · HA-204
PodPod 4 · Hall 3 · row D
PowerFeeder 3 · 156 kW headroom · needs 44
CoolingRow D · liquid loop · ok
NetworkLeaf D-07 · 8 ports free
Lead time6 days · WO created in Neev
Approve · provision

03

Network and storage.

VRFs, VLANs, cross-connects and storage volumes attached to the tenant and the rack. The cross-connect on the invoice is the one in the patch panel.

  • VRF / VLAN
  • Cross-connects
  • Volumes
Pod 4 · fabric
ObjectAttached toState
VRF 204Halcyon AIUp
Leaf D-07 · ports 1–12Nodes 01–12Up
Cross-connect XC-3-118Carrier A · MMR 3Up
Cross-connect XC-3-121Carrier B · MMR 3Provisioning
Volume vol-204-a · 1.2 PBNodes 01–12Up

04

Meter the kilowatt and the hour.

GPU hours, node hours and kilowatt-hours are metered from the same PDUs and the same scheduler, per tenant, per pod. Usage lands on the contract line in Ayana.

  • GPU hours
  • kWh
  • Per pod
Usage · Halcyon · November
GPU hours
61,440
Node hours
7,680
Power · kWh
284,120
Committed kW
404 of 420
Overage
0

05

Bill through Ayana.

Committed capacity, metered usage, network and storage roll into the tenant's single invoice alongside any colocation they hold. Finance reads it from the ledger they already run.

  • One invoice
  • Committed + metered
Invoice · HA-2026-11$ 166k
LineAmountSource
Committed capacity · 420 kW$ 33,264contract
GPU hours · 61,440$ 122,880Aire meter
Storage · 1.2 PB$ 9,600Aire meter
Cross-connects · 2$ 500contract
SLA credit$ 0
Total$ 166,244to SAP

06

Observe like a tenant, operate like the hall.

The tenant sees their pods, utilisation and power. Facilities see the same nodes as load on a feeder. Vitana sees both when something breaks.

  • Tenant view
  • Hall view
  • Shared incidents
Pod 4 · live
GPU utilisation87%96 GPUs
Power404 kWFeeder 3
Inlet · loop31 °Cok
Incidents · 30 d1P3 · link flap

What makes Aire different.

It is a cloud platform that was born inside the data centre record, not bolted onto it.

01

Capacity-aware provisioning

Nothing is provisioned where Neev shows no power, cooling or network headroom.

  • Headroom checkPer feeder, row and leaf.
  • Work ordersRacking and cabling created in Neev.
  • Lead timeHonest, from the floor.

02

One meter, one invoice

GPU hours and kilowatt-hours are metered from the same sources and billed on the same invoice as colocation.

  • Per tenant, per podUsage lands on contract lines.
  • Committed + meteredBoth bases, one bill.
  • Finance stays putPosts to the ledger.

03

Two views of the same node

A node is load on a feeder to facilities and a GPU to the tenant. Incidents are shared across both.

  • Tenant scopeTheir pods, nothing else.
  • Shared incidentsVitana, with the rack attached.
  • AuditWho provisioned what, when.

Hosting a neocloud, or becoming one? Bring your scheduler and your PDU list and we will show one tenant metered end to end.

Speak to our experts →

Works with the stack you already run.

Aire orchestrates what you run and meters what the floor sees. Bring your hypervisor, your scheduler and your fabric.

Virtualisation & orchestration

Where workloads are placed.

KubernetesSlurmVMware vSphereOpenStackProxmoxTerraform

Accelerators & servers

What the pods are made of.

NVIDIAAMDDellSupermicroHPEsoonLenovosoon

Network & storage

The fabric and the volumes.

CiscoAristaJuniperNVIDIA SpectrumCephNetAppVAST Datasoon

Billing & identity

Through Ayana and your SSO.

AyanaSAPOracle NetSuiteOktaKeycloak
RedfishIPMISNMPPrometheus exportersOpenTelemetryREST

Design partner places are open

One tenant, metered end to end.

Design partner places are open for operators hosting GPU or cloud tenants.

We reply personally. Nothing goes on a list. Privacy