Before Committing Capital to an AI Data Centre Upgrade

The customer wants denser AI capacity. The design is feasible and the investment case clears the required return. The committee must now choose how to deliver it.

An advance from the customer may change the price. More debt may constrain the next capacity tranche. Phasing may put the commercial agreement at risk. Each route can make the immediate investment possible while changing its value or limiting another opportunity.

Before signing the variation or releasing the equipment order, the CFO needs to compare the choices across customer economics, financing, delivery and competing capital needs.

Which commitment deserves the capital, what does it displace and which conditions make it acceptable?

Why review the assumptions now?

On 15 September 2026, Khazna and Siemens announced an exploratory MoU covering future AI data centre technologies, including 800 VDC power architectures. Khazna published its first ESG report on 6 October.

These announcements put future technical capability and energy assumptions in focus. They do not establish a mandatory upgrade or a funding shortfall at Khazna. The investment questions and hypothetical scenarios below are CSP’s analysis.

Start with the customer promise

Take the next hall being approved for an identified customer. Compare meeting the contracted specification now, paying for additional capability now, and retaining the option to install that capability later. Each choice needs a cost, a delivery timetable and a commercial justification.

Reserved capacity, actual consumption, an all-in service price and a guaranteed technical capability create different obligations. Reconcile the agreement with the engineering design, acceptance tests, electricity recovery and change provisions.

A customer request becomes an investment obligation when the contract or an approved variation assigns it. Additional capability may earn a premium, secure an enforceable expansion commitment or preserve a valuable option. Show which benefit supports the spending and how the cash flows change if demand arrives later.

Separate energy consumption, price and recovery

The model needs three distinct answers: how much electricity the service requires, what procurement costs and charges apply, and which costs can be billed to the customer.

Khazna reported 27% carbon-free coverage for mechanical cooling electricity consumption in 2025 through onsite solar and certified instruments. That reporting measure does not describe 27% of total facility electricity. It does not establish a tariff, an operating PUE or customer recovery terms.

Illustration 1: who carries a higher electricity cost?

Assume an average IT load of 10 MW over 8,760 hours. Compare annual power usage effectiveness, or PUE, of 1.30 and 1.50. PUE is total facility energy divided by IT equipment energy.

The difference is 17,520 MWh, or AED 6.132 million a year at the illustrative price.

These are comparison assumptions, not Khazna operating data or a UAE tariff estimate. The price is applied uniformly to total energy. Water, demand charges, taxes, maintenance and changes in utilisation are excluded. Customer revenue is held constant before additional recovery.

Full pass-through can still require cash between the supplier payment and customer collection. A fixed all-in price may leave the operator exposed unless repricing is permitted. Capped recovery requires a clear allowance and an identified party bearing costs above it.

Later upgrade spending needs a separate estimate for equipment, installation and downtime. Include tenant-owned GPUs and servers only where the operator is obliged to fund them.

A profitable upgrade can still create an early cash requirement

Illustration 2: compare the choices before covering the gap

This is a separate hypothetical scenario. It is independent of the electricity calculation above and does not describe a client or Khazna project.

Assume the complete investment model clears the required return. An upgrade requires AED 20 million before acceptance. At that payment date, unrestricted cash and committed, drawable financing total AED 14 million, after other obligations and minimum reserves. The customer reimburses AED 8 million only after acceptance.

The immediate shortfall is AED 6 million. The later reimbursement cannot fund the earlier payment. Customer collection and financing conditions still need verification.

Renegotiate the contribution

An advance could close the gap. If the customer asks for a price concession, compare the value surrendered with the funding benefit. Approve the revised economics and usable payment terms together.

Use additional financing

Financing could preserve the delivery date and price. Include its cost, conditions and the headroom it consumes. Test whether it limits a better use of capital elsewhere in the approved pipeline.

Phase the capacity

A smaller first stage could reduce the initial commitment. Test the effect on revenue timing, installation costs and customer acceptance. A lower cash requirement is useful only if the phased service remains commercially and technically viable.

Compare complete cash flows under each route, including later CAPEX and a supported terminal value where relevant. Recommend the option that meets the service promise, remains fundable and justifies the capital relative to the alternatives. If none does, revise or defer.

What belongs in the approval?

The committee needs more than a positive return. The decision should record the option selected, the obligations retained, the funding available at each milestone and the conditions for releasing capital.

Where approval depends on an advance, revised price, financing commitment or scope change, state the dependency explicitly. Assign an owner and a deadline to the unresolved condition.

Your teams establish the technical, legal, commercial and financing facts. CSP challenges the financial logic of the combined decision: does the preferred route earn its place in the capital plan, and which conditions must accompany approval?