Building a GPU cloud in your data centre? Book a demo →
Book a Demo

Project Coldstart · site qualification

Turn your megawatts into revenue.

Your building becomes a GPU cloud you own, funded by someone else. hosted·ai brings the customers, the financing, the GPUs, the platform and the 24/7 operations. On a 2 MW hall that is around $150M of revenue running through a company you own. Fifteen minutes here tells us whether your site fits.

1.5 MW - multi-GWSingle hall to full campus
hosted·ai sells itYou run the building; hosted·ai brings the customers and runs the cloud.
You get paidFixed facility income whether the GPUs are busy or not, plus your share of what it sells
No CAPEXhosted·ai funds the GPUs, not you
1You tell hosted·ai about your site

Power, cooling, timing, people, and how you would want to work with hosted·ai. One hall or a whole campus. Fifteen to twenty minutes.

2hosted·ai qualifies it

A first reply within one business day, then within five working days: a yes, a no, or the two or three things that would make it a yes.

3You start earning

hosted·ai brings the anchor customer and closes the financing. Nothing is ordered and no money moves until the contract is signed.

No downside

hosted·ai orders nothing and draws no debt until a customer has signed.

00

What you're looking for

This shapes the rest of the form — you will only see the questions that apply to you.

01

Who you are

So we know who to call and what we are looking at.

City and country is enough at this stage.

02

Site control and jurisdiction

The financing depends on this. A lender will not fund GPUs into a building you might lose.

Freehold, or remaining lease term.

Under 5 years is shorter than our operating agreement. We would need a landlord agreement or an extension before financing can close.

A mortgage holder may need to consent to the GPU financing.

This lets the financier keep the site running if the project company fails.

Export controls on advanced accelerators vary by country. Tell us if you already know of restrictions.

03

Power

The one number that decides whether we can start. 1.5 MW is where the economics begin - there is no upper limit, and we build multi-gigawatt campuses in tranches. IT load means power to the racks, not utility headroom.

Firm and energised.

No upper limit - tell us the full campus figure.

A first tranche needs 1.5 MW to work economically. If you can reach that later, or have other sites, tell us about the expansion path in the notes at the end - we scale these deals to gigawatts.
Without a signed connection, the capacity is a plan rather than an asset. We can still talk, but financing waits for firm evidence.

Including transmission and levies.

Curtailment risk changes what we can promise an AI customer.

04

Cooling and density

B300-class nodes run about 14 kW each. Air alone stops working around 40 kW per rack.

Annualised if you have it.

Under 40 kW per rack we would need a cooling upgrade before deployment. Tell us below what the upgrade path looks like.

Facility water loop, CDU space, pipework.

Without a liquid path this site is limited to lower-density GPUs and a shorter useful life. Not a blocker today, but it caps what we can deploy.
05

Space, floor and logistics

A surprising number of sites fail here. GPU racks are unusually heavy and must sit together.

InfiniBand cable lengths mean GPU racks cannot be scattered.

06

Connectivity

07

Resilience and SLA

Whatever the project company promises customers, the facility has to promise it first with more headroom.

Concurrent maintainability is what matters, more than a tier badge.

If maintenance drops customer load, an AI training customer loses days of work. This needs a mitigation plan.

Take-or-pay works both ways.

Honesty here is worth more than a clean sheet.

08

People on site

This decides how much we have to fly in, and how fast a failed node gets fixed.

Unmanned sites need a local partner arrangement before we can deploy. Solvable, but it goes in the plan and the cost.
09

Certifications and compliance

Certifications are inherited by our customers, so they set which buyers the site can serve.

Design, Constructed Facility and Operational Sustainability are different things.

10

Timing

The dates decide whether this is a Q3 project or a next-year project. Estimates are fine.

Delivery, staging, racking.

Nodes energised and networked.

Your site timeline

Add the three dates above and we will show the path to live.

Debt on the building, a grid connection use-it-or-lose-it date, a board commitment.

11

How you want to work with us

Two ways to do this. Neither is wrong - they just change who owns the hardware and who takes the upside.

FACILITY PRICING

Per month, excluding electricity. The number we most need.

Fixed income regardless of GPU utilisation - it is what makes the fee bankable.

FINANCING

The anchor customer prepays part of the contract; that cash replaces equity.

Only the estate we finance and operate - never your whole data centre.

12

Demand, ownership and anything else

Constraints, ambitions, a second site, a deadline, a reason this is urgent.

Readiness

-/100

Answer the sections above

Someone from our team will be in touch within one business day. Within five working days you get a real answer: a yes, a no, or the two or three things we would need to change to make it a yes. If it is a no, we tell you why.

Nothing is sent until you press Send. You can also download your answers as a file and keep a copy.