Colocation from 2 MW, or fully managed GPU-as-a-Service. Tell us your power, density and date — we return a priced, contract-ready proposal. You commit nothing until you have read it.
Market conditions
Capital is not the constraint on AI deployment. Access is. The operators with power, density and a delivery date are not the ones answering inbound email.
JLL · Midyear 2026
CBRE
CBRE
CBRE
Take the hall and fill it yourself, or take the compute fully managed. Same sourcing, same oversight, same point of contact.
You own the compute. We deliver everything it plugs into.
Rack-ready white space with power and cooling landed at the rack position. You ship servers; the hall is conditioned, cabled and staffed before they arrive.
We deliver the compute too. You bring workloads.
Dedicated GPU capacity, sourced, installed, commissioned and operated. No hardware procurement, no capex, no data center team — validated clusters against an SLA.
Scope
Line by line, for both tracks. Everything on our side of the split is ours to source, contract, manage and answer for.
| Component | Colocation | GPU-as-a-Service |
|---|---|---|
| Site, shell and physical security | Kalastra | Kalastra |
| Utility power and distribution to rack | Kalastra | Kalastra |
| Cooling plant and delivery to rack | Kalastra | Kalastra |
| Racks, PDUs, busway tap-off | Kalastra | Kalastra |
| Fire, life safety, BMS / DCIM | Kalastra | Kalastra |
| Carrier access and cross-connects | Kalastra | Kalastra |
| GPU servers and network fabric | You | Kalastra |
| Hardware commissioning and burn-in | You | Kalastra |
| Facility O&M and remote hands | Kalastra | Kalastra |
| OS, orchestration, model stack | You | You |
| Your workloads and your data | You | You |
Scope shown is the customer-facing split. The contracting structure and the operating partners behind each line are confirmed per deal and set out in the proposal before you commit.
The delivery path
Everything before the letter of intent is ours to carry. You see a price and a full scope before anything is binding.
Power, rack density, region, redundancy tier, cooling type and your ready-for-service date. Colocation or GPU-as-a-Service.
Output
Scoped requirement
We shortlist facilities against your spec from our operator network — power availability, density headroom, cooling capability, delivery date and commercial terms verified before anything reaches you.
Output
Qualified shortlist
We issue scoped RFQs across O&M, cooling, power, GPU supply, connectivity and EPC, and run the bids. You are not asked to manage vendors.
Output
Competitive bids
One document, one price, full scope and demarcation set out line by line. No obligation attaches to it — you can read it and walk away.
Output
Contract-ready proposal
You issue a non-binding letter of intent so we can lock terms with the facility and the service stack. Definitive agreements follow once the structure is papered.
Output
Executed agreement
Fit-out, power and cooling commissioning, cluster validation where compute is in scope, then handover against the SLA — with delivery oversight and reporting throughout.
Output
Live capacity
Infrastructure
Our team has direct experience commissioning liquid-cooled AI compute environments. From power architecture and coolant distribution to InfiniBand fabric and GPU cluster validation, we manage the technical build-out alongside our facility partners.
Partner network
Four parties touch a GB300-class deployment, and two of them are written into NVIDIA's warranty terms. We record where each one stands — per platform, because an authorisation for GB200 does not carry to GB300.
Indirect warranty condition
Builds the system. Already an NVIDIA manufacturing partner by definition — we confirm the SKU is on the qualified catalogue.
Direct warranty condition
Racks and commissions the hardware. Install outside NVIDIA's documented procedure is an explicit warranty exclusion.
Indirect warranty condition
Houses the deployment. Power and cooling outside spec can trigger the same exclusion, even where the site is otherwise sound.
Direct warranty condition
Services it for the life of the contract. Repair by anyone not NVIDIA-authorised voids cover.
Questions
If a requirement is not a fit for us, the fastest thing we can do for you is say so.
No. We source capacity from an operator network we have qualified, then assemble, price and manage the deal around it. Owning the asset is not what makes a deployment land on time — sourcing it, contracting it correctly and holding the vendors to schedule is.
No. A broker hands you an introduction and invoices for it. We shortlist against your spec, run competitive bids across the full service stack, price it as one number, and remain the party you hold accountable through ready-for-service. You deal with us, not with six vendors.
No. This is dedicated, reserved wholesale capacity from 2 MW upward — not per-hour instances on shared hardware. If you need a handful of GPUs for an afternoon, a hyperscaler or a GPU cloud is the right answer and we will tell you so.
Nothing, and it carries no obligation. Stages D1 through D4 — brief, sourcing, bids, priced proposal — are ours to carry. You commit at the letter of intent, once the price and the full scope are already in front of you.
The letter of intent is with Kalastra. The definitive agreements depend on how the deal is structured, which varies with the facility and the jurisdiction. That structure is set out in the proposal before you sign anything — we do not ask anyone to commit to a counterparty they have not seen.
Yes, and it is a common path. The demarcation shifts but the facility scope does not, so adding managed compute to an existing colocation footprint is a scope change rather than a new deployment.
We tell you, early, and we tell you why — whether it is power, density, cooling, timing or price. A requirement we cannot fill is worth more to you as a fast no than as a slow maybe.
Yes. If you are representing an end client you stay in the deal and in the loop — the requirement form has a field for it and nothing routes around you.
Global Operator
Headquartered in the United States. Every colo node and GPU cluster is qualified before it reaches you.
US-based colo and inference infrastructure across Tier-1 markets.
Browse capacityGet started
From a single GPU node to a multi-megawatt colocation deployment — share your requirement and our team will respond within one business day with a structured, priced proposal.
Contact
We work with hyperscalers, sovereign AI programs, GPU cloud resellers, and enterprises needing 10MW or more. If you have a requirement, tell us and we will respond within one business day.
Hyperscalers
Dedicated GPU-aaS blocks at 10MW+
Sovereign AI
Full facility lease with national data residency
GPU Cloud Resellers
Wholesale colocation and GPU-aaS for resale
