Aircloud Compute is now Lightyard. Aircloud continues as a private inference platformAircloud continues for inference

For neoclouds

GPU clusters for neoclouds,
built to your spec.

You have the customers and the team that serves them. We find the site, design, source and build each new cluster to your spec and hand it over wave by wave, or run it for you on our software.

You bring
Customers and a team
We do
Site, design, source, build
Operator
Your team, or us
Customers
Contracted by you
For the executive team

More capacity without a deployment team.

Every new cluster is a project of its own: a site, a design, hardware orders, freight, racking and burn-in. We run that project, so your people stay on customers, product and the cloud you sell.

  • No new hires for each buildThe engineers who design, source and rack the cluster are ours for the length of the project.
  • Dates in the proposalDelivery is stated in weeks from signing, so launches can be planned around it.
  • Your call on operationsAfter handover your team runs the cluster. A managed contract is there if that changes.
Cluster deployment
For your engineers

Built to your spec, on hardware you own.

The alternative to building is buying wholesale capacity from a larger cloud and reselling it. That capacity comes on the provider's terms.

Built to your spec, on hardware you own.
Wholesale capacity from another cloudA cluster we build for you
HardwareWhatever the provider stocks, in its configurationThe NVIDIA platform and configuration you specify
NetworkSet by the provider, sometimes sharedA dedicated RoCE v2 fabric, to your design or ours
Images and driversThe provider's images and toolingWhat you run today, agreed in the proposal before burn-in
AcceptanceThe provider's SLA and the provider's testsYour acceptance tests, run on every node before handover
The hardwareOwned by the providerOwned by you
Your customerServed on someone else's hardware and termsServed on hardware you control

Compared by category, not by provider.

For capacity planning

Capacity that arrives on dates you can plan against.

We hand capacity over in monthly waves of 32 nodes, each one accepted before the next. Your sales team can see when each block of nodes is due and sign customers around the dates in the proposal.

  • A date for every waveEach wave has its own date in the proposal, counted in weeks from signing.
  • Accepted wave by waveYou accept each wave against your own tests before it joins your inventory.
  • Room for later wavesLater waves join the same fabric, because it's designed for the final size from the start.
Illustrative
Software

Tenants, allocations and billing, if you want them.

The stack we built to run our own clusters covers the fleet, the cloud and the platform: inventory and health, tenants, contracts and billing, endpoints and jobs. Use the layers you need next to your own tools, or none of them.

01

Every node tracked from delivery.

Each node enters the inventory as it's racked. Burn-in results, firmware, health and every repair are recorded against it, so a failing node shows up in one place, with its history.

02

Capacity divided between customers.

Customers, contracts and allocations sit on the same inventory as the hardware. Each customer gets its own nodes, and billing follows what it was allocated.

03

What customers run on top.

On their share, customers launch inference endpoints, training jobs and containers themselves, without an engineer setting each one up.

Illustrative · a 32-node cluster
How it works

From your spec to capacity you can sell.

  1. Step 1: Send the spec

    Platform, size, region, fabric and the acceptance tests you use today.

    You getA written proposal
  2. Step 2: Site and design

    We find a site if you don't have one, then design the fabric, power and cooling around it.

    You getA design and a dated plan
  3. Step 3: Source and build

    Hardware through our relationships with NVIDIA and server manufacturers, racked, cabled and burned in wave by wave.

    You getAccepted waves
  4. Step 4: Operate

    Your team runs it, or we do, on our software, to an SLA in the contract.

    You getYour choice of operator
The usual way

Most neoclouds start with a deployment.

Some later hand us day-to-day operations, and the cluster becomes a managed cluster.

For each way to work with us: who owns the cluster and who does each stage
Way to work with usOwnerDesignSourceBuildOperateSoftware
Cluster deploymentWe design, source and build it, then hand it over burned in. Your team runs it.YouLightyardLightyardLightyardYouOptional
Managed clusterYou own it, we run it. We can build it first, or take over a cluster that's already live.YouOptionalOptionalOptionalLightyardLightyard
SoftwareYour team runs every stage, on our stack.YouYouYouYouYouLightyard
Reserved clusterWe own and run it. You reserve it for 60 months, with rollover into the next generation.LightyardLightyardLightyardLightyardLightyardLightyard
Sites

Where your next cluster can go.

Colocation in the region your customers need is the usual choice. The design stage confirms it against power and timing.

  • 01 Colocation

    A Tier III facility in the region you choose.

    Colocation
  • 02 On-prem

    Your data center, with the cluster designed around it.

    On-prem
  • 03 Retrofit

    A building or mining site upgraded for GPU density.

    Retrofit
  • 04 Modular

    A prefabricated site on land with power.

    Modular
Hardware

The platforms we build on.

HGX B300 nodes and GB300 NVL72 racks today. VR200 NVL72 is on the waitlist for first allocation.

All hardware

Questions

Questions we get on the first call.

Something missing? Ask us directly. We usually reply within a few hours.

Keep reading

Solutions, sites and who we build for.

Solutions
Sites
  • ColocationA Tier III facility in the region you choose.
  • On-premYour data center, with the cluster designed around it.
  • RetrofitA building or mining site upgraded for GPU density.
  • ModularA prefabricated site on land with power.
Industries
  • NeocloudsCapacity built to your standard, ready to sell.
  • Venture capitalOne cluster, shared by your portfolio companies.
  • Enterprises & sovereignYour own cluster in your region.
  • Private equityGPU capacity on portfolio sites, or a neocloud as a platform.
  • Asset ownersHardware, a hall or a power site, turned into a running cluster.
  • InvestorsAn operated GPU asset, without building an operator.
Get started

Tell us where you're starting.

Capital, hardware, a site or a team: tell us what you have and when you need it running. We usually reply within a few hours and come back with a written plan.

sales@lightyard.ai