Can a Data Centre Be Bankable Before It Is Energised?

Data-centre lenders are discovering that a signed tenant lease cannot produce cash flow without deliverable power. Microsoft’s reported cancellations after US facility and power delays show how quickly projected rent can disappear. The projects that survive underwriting turn interconnection evidence into contractual milestones, rather than treating it as an engineering footnote.

Microsoft reportedly cancelled leases on two entire US data centres after facility and power delays.

These were not small tenant disputes or a missed construction target hidden in a footnote. The stakes were the future rent on whole facilities, and the consequence was cancellation. DataCenterDynamics reported the episode in September 2025; SitusAMC reported it again in 2026. Both accounts point to the same exposure: facility and power delays collided with service-level obligations, penalties and termination clauses.

For anyone financing data centres, that should rearrange the usual order of importance.

The familiar story says a hyperscaler commitment is the hard part. Find a tenant with a large balance sheet. Sign a lease. Raise construction money against the future rent. Build the shell, fill it with servers, collect revenue.

But the reported Microsoft cancellations expose the awkward truth beneath that neat sequence. A data centre can be complete enough to photograph, staffed enough to open, and leased enough to impress a credit committee. Yet without electricity arriving in the promised quantity and on the promised date, it has no product to sell.

That is not an engineering inconvenience. It is a financing event.

Microsoft’s reported cancellation of leases on two entire US data centres shows how projected tenant revenue can disappear when facility and power delivery slip. Photo: Atomic Taco / Wikimedia Commons, CC BY-SA 2.0.

The question is simple: if a creditworthy tenant has signed, why is debt not easy?

Because the lender is not truly lending against a building. It is lending against a future stream of rent that begins only when power turns on. If that moment slips, construction costs continue, interest continues, tenant obligations may become impossible to meet, and the carefully modelled first day of revenue never happens.

The comforting story lenders used to tell

For years, the underwriting logic was understandable. Demand for computing capacity was strong. Major tenants had formidable credit. The sponsor brought land, construction plans and equity. The lease appeared to answer the central question: who will pay?

Power sat in another folder, usually under technical diligence. Important, certainly. Decisive, perhaps not.

That mental model is now dangerously old.

Colliers, the commercial real-estate advisory firm that tracks the US data-centre market, reported in March 2026 that power can account for 40% to 50% of total project cost. Smartland Capital, a project-finance analyst focused on data-centre power and capital, put the timing mismatch more plainly in July 2026: a data centre can be built in 12 to 24 months, while a grid connection can take 36 to 84 months.

Imagine you are a lender. Your borrower can construct the asset in two years. The electrical connection may arrive seven years after the loan closes. What, exactly, is servicing the debt during the gap?

A lease does not answer that.

This is where most people stop looking. They hear “interconnection” and assume it means the project has power. It may mean only that the developer has entered a queue. It may not settle the scope of grid upgrades, the substation delivery sequence, the energisation date or whether supply can be curtailed. Those distinctions are the difference between a land option and an operating asset.

ERCOT. Photo: Ercot / Wikimedia Commons, CC BY-SA 4.0.

The queue that looked like capacity

Consider PJM and ERCOT, the US grid systems whose interconnection queues have become a warning for anyone treating demand as a site-selection strategy.

PJM operates across parts of the eastern United States. ERCOT runs most of Texas’s electricity grid. Both matter because they sit in markets drawing intense data-centre interest. Carbon Direct, a carbon-management firm that analysed those two queues, found in May 2026 that 60% of ERCOT queue capacity and 76% of PJM queue capacity may never come online, or may be delayed for years.

A queue is therefore not a promise. It is a place in a process.

DC Hub, a data-centre market intelligence provider behind the Data Center Power Index, put interconnection waits across leading US markets at 24 to 60 months. Its median wait for hyperscale-grade loads was 60 months. Five years.

At first glance, this sounds like a US planning problem. It is more than that. It changes the order in which a project should be financed.

Sponsors that select a site because a tenant wants capacity nearby, then seek a power route later, are effectively asking finance to bridge an unknown. The building timetable may be visible. The tenant’s name may be reassuring. But the one date that converts either into cash, the energisation milestone, is still floating.

That is why a developer can be “leased” and still be unbankable.

The project has not failed because demand disappeared. It failed because demand arrived before supply.

When the strongest tenant becomes the sharpest risk

The Microsoft case adds a nastier twist.

A big tenant is supposed to lower risk. Usually it does. Lenders like reliable counterparties because reliable counterparties pay rent. But a sophisticated tenant also has service-level expectations and termination protections. If a facility cannot be energised and made available as agreed, those clauses can turn the tenant from the project’s great comfort into its most immediate source of pain.

DataCenterDynamics’ September 2025 report and SitusAMC’s 2026 reporting identified Microsoft’s cancellation of leases on two entire US data centres after facility and power delays. The research brief identifies the missing protection clearly: power-milestone security had not been aligned against service-level triggers and termination rights.

The important part is not the identity of the tenant. It is the sequence.

First, projected rent supports the financing case. Then power is delayed. Then the tenant’s contractual rights bite. Then projected rent is no longer a future certainty but a vanished assumption. Debt does not receive comfort from an impressive logo on an unsigned continuation of a lease.

This is the same uncomfortable logic explored in why lenders can love your customers and still reject your business. A customer contract is valuable only if the borrower can actually perform it.

GI Network’s view: In data-centre financing, the power date should be treated as a revenue date. If it is not evidenced, synchronised and protected, the lease is not security. It is exposure.

Dubai’s quieter counterexample

The story is not that grid-dependent projects can never be financed. Dubai Silicon Oasis, the United Arab Emirates development zone, offers a more useful counterexample.

Matchpoint Partners, an adviser covering GCC data-centre bankability, reported in August 2026 that a development there had 29 MW readily available and another 100 MW of committed power. MW means megawatts, the measure of electrical capacity.

Dubai Silicon Oasis shows how available capacity and separately committed later capacity can be matched to staged capital deployment. Photo: wilhelmtittes / Wikimedia Commons, CC BY 3.0.

The later 100 MW was not the same thing as power already available. There was still delivery uncertainty across phases. Yet that is precisely why the case matters. Explicit separation between the 29 MW available now and committed later capacity allowed financing to structure capital deployment in phases.

This is where the opposite can be true: grid dependence need not kill financing when uncertainty is contained, evidenced and matched to the amount of capital at risk. A lender can assess the first phase against usable capacity, then make later capital contingent on proof that the next block is progressing.

What surprised us was that the Gulf lesson is often described as simply “more power”. The stronger lesson is contractual and organisational. Coordinated, phased delivery gives finance something it can measure against construction and tenant dates.

Put the US queues, the Microsoft cancellations and Dubai Silicon Oasis side by side and a pattern appears that none of the reports states quite so plainly: the real divide is not between markets with demand and markets without it. It is between projects where electricity is a forecast and projects where electricity is a sequence of enforceable evidence.

Amazon changed the route, not just the energy source

Amazon’s Pennsylvania project beside a nuclear plant points to another route around the bottleneck.

AP News reported in 2025 that Amazon was building a data centre with a behind-the-meter connection, a direct connection on the generation side of the wider grid, beside the nuclear facility. The attraction was not merely access to low-carbon power. It was speed and reliability relative to a congested grid-interconnection process.

That makes behind-the-meter generation a finance strategy, not only an energy-purchasing choice.

If the conventional route leaves a project waiting years for grid delivery, a direct generation link can alter the central underwriting question. Instead of asking whether a distant grid process will finish in time, the financier asks whether the generation connection, its regulatory position and its delivery arrangements are credible. It is still complex. The research brief is clear that this path requires higher capital and regulatory complexity. It is not a universal escape hatch.

But it demonstrates something important: power risk can sometimes be redesigned, not merely endured.

The railway financiers of the nineteenth century understood the underlying principle. Traffic revenue meant little until rights of way and track had been secured. Early wind and solar developers learned a modern version when projects entered grid queues and discovered that generating electricity and delivering it are separate achievements.

Data centres are now learning it at extraordinary speed.

Regulation may shorten the queue. It cannot erase today’s risk.

In July 2026, the Federal Energy Regulatory Commission, the US energy regulator known as FERC, directed grid operators to remove barriers and accelerate delivery for AI-scale data centres. The action also requires developers to fund grid upgrades.

That is a recognition that interconnection delays are no longer a private inconvenience. They are an infrastructure-finance problem with wider economic consequences.

Still, policy intent is not an energised substation.

FERC’s direction may improve future timelines. In the meantime, a developer must understand who funds upgrades, what happens if they cost more or take longer, and whether the financing can survive that outcome. Experienced investors do not give regulatory reform a date certain merely because it has a headline.

What founders should do before the capital process

Do not raise against an artist’s impression and a tenant letter while power remains a separate workstream.

Start by putting the lease, construction timetable, power documents and financing model on one calendar. Mark the date at which rent starts, the date at which the tenant can terminate, the expected completion date and the evidenced energisation date. If those dates do not line up, the project does not yet have one coherent financing case.

Next, split capacity by phase. Dubai Silicon Oasis shows why. Evidence for available capacity should support the first deployment. Committed later capacity should govern later deployment. This reduces the temptation to finance a full build-out against a single broad claim about future supply.

Then model the ugly but ordinary scenario: energisation slips by 12 to 18 months. The research does not provide a representative numerical sensitivity, so no honest article should invent one. But the direction is clear. Rent commencement is postponed, construction interest continues, debt service has less support and more equity may be required. If the model only works when the utility date is perfect, it does not work.

Finally, put power risk where it can be managed. If a contractor controls a deliverable, construction milestones should reflect it. If a tenant has termination rights, the sponsor should know exactly what power delay triggers them. If upgrades sit with the developer, the budget and funding obligations need to be visible before debt terms are negotiated.

A nearly complete data room will not compensate for the missing document that proves electricity. As GI Network has written about lenders rejecting an almost-complete data room, completeness and decision usefulness are different things.

What investors should see before they see the headline rent

Investors often fixate on the tenant name because it feels like the cleanest answer to uncertainty. That instinct is human. It is also incomplete.

The smarter question is: what has to happen before this tenant can start paying, and who is on the hook if it does not?

Check whether capacity is available, committed or merely sought. Check the timing of any substation and grid upgrades. Check whether the proposed debt assumes revenue before power is contractually secure. Check the lease for service requirements, termination rights and the dates that matter. Check whether the sponsor has sufficient equity or completion support to absorb a delay.

Do not assume that more leverage is evidence of a stronger project. It can simply mean that the financing model has ignored the period in which a finished building earns nothing. The same discipline applies to any apparently impressive funding figure, as The Most Dangerous Number in a Funding Announcement argues in a different context.

The five-question test

GI Network would begin a data-centre capital mandate by testing the power proof stack before investor outreach: reconcile the interconnection position with the construction programme and tenant milestones, expose gaps in the cash-flow model, identify the documents and remedies needed for underwriting, then map capital providers whose risk appetite matches the actual delivery path. That is not administrative preparation. It is the work of deciding whether the proposed rent can be underwritten at all.

Use the Five Dates and Five Proofs test before calling a project financeable:

1. 1.Capacity date: When is each MW available or committed?
2. 2.Build date: When can the facility physically receive and use it?
3. 3.Tenant date: When does rent start, and when can the tenant walk away?
4. 4.Debt date: When do interest and repayment rely on operating cash flow?
5. 5.Remedy date: At what point does a delay trigger payment, step-in or further equity?

Then demand five proofs: entitlement, delivery path, energisation timing, operational firmness and remedies.

If any date does not match, or any proof is missing, the project may still be attractive. It is simply not yet the low-risk leased infrastructure asset it claims to be.

That is the myth worth discarding. The lease is not the asset.

The delivered power is.