Powered-site + GPU deployment matching

Match a powered site with GPUs.

Start with the asset you can document. See what physically fits, then calculate the return for the party supplying the site or the GPUs.

A powered-site owner does not need to buy the GPUs. A matched GPU counterparty can supply the hardware. Saqrion sizes the deployment and helps qualify the commercial match; ownership and contracts are verified project by project.

Who brings what

Three parties. One deployment.

A GPU deployment needs power, a facility and GPUs. Those are almost never owned by the same party — and with a containerized deployment the facility is delivered, not built. Here is who supplies what, and how each one gets paid.

Party 1

The powered-site owner

Brings
Power, and somewhere to put the containers: an energized or utility-committed supply, a pad that takes the container weight, an electrical interconnect, clear space for heat rejection, fiber and site access.
Gets paid
A contracted hosting fee, per critical IT kW per month.
Does not need
To build a data centre, or to buy a single GPU. The power, IT and heat-rejection modules arrive built.
Party 2

The GPU owner

Brings
The GPUs — a deliverable allocation or delivered hardware, and the capital behind it.
Gets paid
Contracted compute revenue, per GPU-hour, after paying hosting and electricity.
Does not need
To own or build a data centre.
Party 3

Saqrion

Brings
The containerized facility itself — power, IT and heat-rejection modules delivered as built units — plus the engineering that sizes it to your power and thermal envelope, and the qualification and commercial structure connecting the two owners above.
Gets paid
Per engagement, on the deployment we scope and structure.
Does not claim
To own GPUs, control a power allocation, hold a compute buyer, or guarantee utilization or returns. Each is verified project by project.

Two more parties sit outside the deal: the electricity supplier sells power to the deployment, and the compute buyer pays for the output. Neither is included in the returns below unless you enter their terms.

01

What can you document today?

Choose one starting point. You can compare the full project later.

kW after required electrical reserve

Use power that can run continuously after the site’s required reserve—not the number printed on a breaker. The estimate fits complete GPU systems and leaves unused power visible.

Only supplying electricity or discussing a future allocation?This can estimate potential scale, but it does not treat unqualified power as an energized site or calculate an electricity supplier’s margin. Land, interconnection, cooling, fiber and an operator still need qualification.
GPU systemKimi K3 B200 reference selected · change

A Kimi unit is one published 16-GPU logical pod. Two pods can share one physical 48U rack. Other systems use their sourced complete-rack packing.

Prefer to paste what you know?Optional browser-side quick-fill

This private helper recognizes power, price, location, GPU system, PUE and GPU/rack quantity in your browser. It does not verify capacity, qualify a site, find a counterparty or upload your text. Review every value it fills.

ROI calculator

Returns from the configuration above

The deployment above already supplies GPU count, IT load, site draw and rack count. Add verified capital and contract terms below; every return updates immediately.

Linked automatically

Current deployment basis

Power, GPU system, rack or pod quantity, PUE and fabric entered above flow into every return below. Only the missing commercial terms need entry.

Powered-site owner view

Turn documented power into hosting revenue

You provide the powered facility—not the GPUs. Test hosting revenue against the GPU-ready build or retrofit cost and site O&M.

Powered-site ROI

Enter the GPU-ready build or retrofit cost, hosting fee and site O&M.

Annual ROI before financing
Simple payback

Start here · real inputs

Terms connecting the site and GPU owner

Type contracted figures. Grey example values are not included until you enter or load them.

Powered-site owner inputsGPU-ready retrofit cost, site O&M and contract term
GPU-owner inputsHardware cost and documented compute revenue

Implied net compute revenue

Work backwards from the goal

What contract rate would produce your target return?

Choose a goal—not a market forecast. The selected owner view determines whether this solves for a hosting fee or a net GPU-hour payout.

Equivalent simple annual cash yield: 33.3%

Powered-site owner target Add the GPU-ready build or retrofit cost and site O&M to solve the required hosting fee.

The physical deployment above is already linked.

This is the minimum arithmetic threshold under the costs and target entered here—not evidence that the market will offer the rate, a quote, or a guaranteed return.

How do I obtain defensible inputs?Contract, quote and billing evidence
GPU-ready build / retrofit cost
This is not land or GPU hardware. Use a dated site/EPC quote covering electrical equipment, cooling, enclosure/racks, civil work, interconnection, installation and commissioning. For an existing site, use the new retrofit budget or an agreed asset basis—not the historical land purchase price.
Hosting fee
Use a signed agreement or term sheet. If none exists, use the target solver above to calculate the minimum fee to negotiate.
GPU hardware cost
Use the purchase order or delivered-system quote, including servers, networking and required ancillary equipment.
Compute revenue
For dedicated or take-or-pay capacity, enter the contractual installed-GPU-hour value after platform fees. For usage-based sales, enter realized net GPU-hour revenue.
Billable utilization
Use 100% only when all installed hours are paid under the contract. For usage-based sales, use sold hours ÷ available hours. Do not discount a take-or-pay rate a second time.
Electricity and O&M
Use the delivered tariff or PPA including demand charges, taxes and losses, plus documented staffing, maintenance, security, network, insurance, rent and other site costs.

Token or workflow revenue needs an exact-stack throughput benchmark and proven demand. Public API prices are buyer retail references and do not establish GPU-owner payout.

Powered-site owner

Facility capital + hosting contract
Annual unlevered cash yield
Simple payback before financing
Monthly hosting revenue
Monthly site operating cost
Monthly site operating cash

GPU owner

GPU capital + compute contract
Annual unlevered cash yield
Simple payback before financing
Monthly net compute revenue
Monthly hosting + electricity
Monthly operating cash
After-debt detail
Financing debt service
Cash after entered debt service

Operating cash excludes any costs not entered here, including software, staffing, insurance, spares, maintenance, taxes and platform fees outside the net payout.

Facility + GPU project

Electricity supplier remains external
Annual unlevered cash yield
Simple payback before financing
Facility + GPU capital basis
Monthly combined operating cash
Consolidated detail
Cash after entered debt service
Site cash + GPU-owner cash
Measured exact-stack evidence

Where these numbers come from

The 1 MW worked case

The calculator opens with this editable illustration so the live formulas are visible immediately. Clear it to enter your own figures, or reload it here at any time.

Mechanics-only sample:

Editable arithmetic inputs only. They are not one sourced project, a quote, forecast, market benchmark, offer, or guaranteed return.

How the cash flow connects

ElectricityPowered siteGPUsCompute buyer

The hosting fee is revenue to the site and a cost to the GPU owner. Electricity is a deployment cost. Compute payout is revenue to the GPU owner.

Advanced underwriting evidenceOptional throughput and stress cases
User-entered stress cases—not market forecasts

Returns appear only from entered figures. Annual yields are unlevered; payback is simple payback before financing, tax, depreciation and residual value.

Engineering assumptionsOptional · planning defaults are already applied
02

Site assumptions

Inputs are visible because site power is not a GPU specification.

Plain-language glossaryPUE, power, cost and contract terms

These definitions describe how this calculator uses each term. Project contracts and final engineering documents control.

PUE · Power Usage Effectiveness
Total site power ÷ IT-equipment power. A PUE of 1.40 means 100 kW of GPU servers and network equipment needs about 140 kW from the site. The extra 40 kW covers cooling, UPS losses, pumps, fans and controls. Lower is more efficient.
Usable continuous site power
Power the site can deliver continuously after required reserve and derating—not the breaker or utility nameplate maximum.
IT load
Electricity used by the GPU servers plus any entered network-fabric power, before facility overhead is added through PUE.
Facility overhead
Power used by cooling, UPS losses, pumps, fans, controls and other site systems. In this calculator it is total site draw minus IT load.
GPU tranche
A specific group of GPU systems with a known SKU, quantity, ownership or allocation evidence, and delivery window.
GPU-ready site CAPEX
The one-time build or retrofit cost needed to make the site deployable: electrical distribution, cooling, enclosure/racks, civil work, interconnect, installation and commissioning. It is not land and it is not GPU hardware.
O&M / OPEX
Recurring operating costs such as maintenance, staffing, security and service contracts. Electricity is shown separately here.
Hosting fee
The contracted amount paid to the site operator per critical IT kW each month. The model passes electricity through separately.
Provider payout / offtake
Contract revenue payable to the GPU owner or operator per GPU-hour. It is not the public retail token or API price.
Utilization
The share of available GPU hours that is contractually billable or underwritten—not theoretical server uptime.
Balance of plant
The supporting equipment and site work around the main modules: cabling, piping, switchgear, pumps, interconnect, civil works and commissioning.
CDU / GA / SLA
CDU: cooling distribution unit. GA: signed general-arrangement drawing that proves physical fit. SLA: contract defining service performance and remedies.

System architecture

One facility. Three functions.

Power, IT and heat rejection are separate functional blocks because each scales to a different engineering limit. They work as one system; they do not mean three IT containers.

See module roles and engineering flow
01 · POWERmodule(s)

Power module

Receives the site utility service and safely conditions, protects and distributes electricity.

  • Switchgear, UPS and battery support
  • Feeds both the GPU racks and cooling equipment
  • Provisional 400 kW planning block inferred from 2 × 400 kVA; topology and usable capacity are not vendor-confirmed
02 · ITmodule(s)

IT module

Houses the GPU racks, fabric, controls and the equipment-side cooling connections. Fiber enters here.

  • GPU servers turn electrical power into compute—and almost the same amount of heat
  • Kimi pods are logical two-node units; up to four 10U nodes can share one physical 48U rack
  • The source reference allocates three rack positions; this calculator budgets two until signed, platform-specific GA drawings confirm fit
03 · HEAT REJECTIONpackage(s)

Cooling / chiller plant

Collects heat from the racks through a CDU liquid loop or an air-cooling loop, then rejects it outdoors.

  • May use chillers, dry coolers, pumps, tanks and controls
  • Consumes power from the power module
  • May be a container, outdoor skid or site plant—not necessarily a container
Power to compute
Grid / site servicePower moduleIT module + GPU racksCompute output
Fiber / network → IT fabricPower module → IT + cooling equipment
Heat to ambient
Heat from GPUsCDU / air loopChiller / dry coolerAmbient air

Reference functional starting basis: one power block + one IT block + one heat-rejection block, subject to the selected rack density, site conditions and final vendor design. These are functions, not guaranteed shipping-container counts. Counts above scale independently; 10 MW+ requires a campus electrical and thermal design rather than simple container multiplication.

View the full engineering and ROI breakdownFor technical and investment review
Facility budgetQuoted equipment + user-entered deployment
Reference power budget module(s)
Reference IT planning range / budget
Reference partial cooling budget package(s)
Deployment / balance of plant
Partial reference / entered GPU-ready site CAPEX
GPU hardware cost (separate)

Partial source reference only—not delivered facility CAPEX. Power capacity, rack fit, and cooling topology remain provisional pending a signed one-line and platform-specific GA. Chiller, pump, tank, redundancy, and balance of plant remain open.

Thermal engineering checkHeat duty—not a second PUE load
Expected server heat from electrical load
Higher published server thermal rating
Published rating in BTU/h
Published rating in refrigeration tons
Whole-site heat by electrical energy balance
Conservative equipment-selection envelope
Selected cooling topology

Energy balance says almost all electrical input ultimately becomes heat; the whole-site heat line therefore equals total site draw. The server source separately publishes a slightly higher thermal maximum, so early cooling selection uses that higher value and flags the conflict. These are alternate design bases, not additive operating loads.

Transparent $/critical-kW-monthNo hidden token-price assumption
Facility recovery over contract term
Site O&M input
Electricity pass-through
Contracted hosting fee
Proposed all-in monthly rate
Monthly surplus after O&M + facility recovery

Illustrative stress testsUser-entered cases—not market forecasts

Enter downside and upside inputs to show the table.

Scope and evidenceWhat the result includes—and does not
    SAQRIONPowered-site + GPU deployment ·

    Indicative Deployment Scope

    PlatformRegion
    Pods / physical racksNodes
    Customer GPUsCooling
    Critical ITSite PUE
    Total site powerPUE overhead incl. cooling
    ModulesMeasured output

    CAPEX ownership

    GPU hardware in facility budgetFacility budget
    GPU hardware (separate)Hosting fee

    Reconciled economics

    Operator EBITDAOperator ROI / payback ·
    Power pass-throughCustomer invoice
    Provider payoutFinancing service
    Owner operating cashOwner cash after debt
    Owner unlevered yield / payback · Combined project CAPEX
    Project operating cashProject cash after debt
    Project unlevered yield / payback · Stress summary

    Indicative planning scope only. Not a vendor quote, financing offer, offtake commitment, or return guarantee.