ENVIZNThe AI Boom Has a Financing Clock
A data center can have a signed customer and a growing market and still run out of cash before the first invoice is paid. The November 3 election made the distance between those dates harder to predict.
Interest falls due on the dates in the loan agreement, whatever the construction schedule says. In the hypothetical project that is $2.5 million every month.
A data center project answers to three schedules at once. Its lender is paid on the dates in a loan agreement and its contractors on the dates in a construction contract, while its revenue waits on a separate chain of milestones: permits, a grid connection, commissioning, and customer acceptance. When those schedules drift apart, a project with a credible customer and useful equipment can still run short of cash before its first customer payment arrives.
The November 3 election pushed them further apart. Reuters reported that Goldman Sachs expects Democratic victories at the state and local level to bring temporary halts to projects potentially affecting some 45% of planned U.S. data center capacity expansion, and that Goldman thinks those halts will be short-lived. The figure measures delay risk across planned expansion, not a loss of capacity already operating. Even a short halt can matter to an owner whose next interest payment comes due before its next financing closes.
I expect AI spending to stay substantial through all of this. What the election changes is how much it matters that each project’s investment, delivery, and customer payments stay aligned, because projects at different stages and on different financing will feel the same disruption in very different ways.
The schedule was already slipping
Political risk is landing on a delivery problem that predates the election. In May, Goldman estimated that about 60% of the capacity scheduled to come online in the following year would arrive on time, falling to roughly 50% over two years, against a historical rate of about 72% for facilities scheduled within four quarters. It named supply chains and labor shortages as the most common causes, along with developers who file in several regions at once and build only at the most favorable site. Those estimates answer a different question from the post-election 45% and cannot be added to it. Read together, they show political uncertainty arriving in a buildout whose dates were already unreliable.
The pressure does not split neatly by party, and Texas shows it most clearly. Starting in August, Governor Greg Abbott, a Republican, paused data center approvals at the state grid and then at the environmental permitting agency while regulators audited the facilities’ power and water use, a review expected to run into December. On November 3 he lost to Gina Hinojosa, a Democrat who promised a moratorium on all data center approvals until the Legislature passes new protections, and who takes office on January 19 alongside a Democratic Texas House.
On cost, the two candidates sounded much alike. Both wanted Texans spared higher electricity bills from data centers, and both wanted the state’s tax breaks for the facilities ended. An estimate from GLJ Research cited by Reuters puts about a quarter of the U.S. data center capacity due online by 2030 in Texas, and nationally, Data Center Watch found that 43% of the 74 gubernatorial candidates it studied supported a moratorium or a ban. The election is likely to extend pressure that had already been building from both sides.
How optimism pays for expansion
In a rising market, valuations do some of the financing. A company that raises equity at a higher share price funds the same investment with fewer new shares, the money supports more projects, the projects support larger growth forecasts, and the forecasts can lift the share price again. That loop can pay for genuinely productive investment. It also makes access to future capital part of the execution plan, since a company that expects to raise again has less room to absorb a delay than one whose commitments are already funded.
The pressure to keep spending is real. On Alphabet’s second-quarter 2024 earnings call, Sundar Pichai said that for Alphabet the risk of underinvesting was dramatically greater than the risk of overinvesting, and he tied that judgment to the company’s ability to use the same infrastructure across many businesses. That second clause carries much of the argument. A company with diversified revenue can absorb idle capacity that would strain a developer whose only income is the lease on that capacity.
Advance investment is often necessary on its own terms. When customers want capacity sooner than transformers, switchgear, or buildings can be delivered, someone has to reserve equipment and commit to sites before demand is fully contracted. The financing question is who carries those commitments in the meantime, and with what money.
A turn in confidence can expose the difference between those positions before customer demand weakens at all. A lower share price makes the next equity raise more dilutive, lenders may ask for wider spreads or tighter covenants, and the company may have to choose among slowing expansion, selling assets, and accepting worse terms. A falling stock mainly costs the business money once it has to raise capital, refinance a loan, or meet a condition written into its funding.
Where sentiment meets the construction schedule
Consider a hypothetical project carrying $300 million of debt at a fixed 10% annual rate, with interest paid in cash, no principal due during the period, and the rest of the construction budget funded separately. It holds $60 million of unrestricted cash for carrying costs. Site, staffing, and corporate overhead run $3 million a month and cash interest adds $2.5 million, so the project spends $5.5 million a month, with no customer prepayments or other inflows before operations begin. The plan calls for customer billing to start in month six.
| Scenario | Months to cash | Cash used | Left or short |
|---|---|---|---|
| Original schedule | 6 | $33M | $27M left |
| Three-month delay | 9 | $49.5M | $10.5M left |
| Six-month delay | 12 | $66M | $6M short |
All figures are hypothetical and describe no specific company or project. They are not construction-cost benchmarks, and they exclude cost overruns, taxes, principal repayments, and additional working capital.
A six-month delay adds $33 million of carrying costs and exhausts the original reserve, leaving the project at least $6 million short before customer cash arrives. The two figures measure different things: $33 million is the extra cash the delay consumes, and $6 million is the gap left after the reserve is spent. Neither is necessarily the project’s accounting loss.
None of this requires the customer to leave or the equipment to fail, and the finished project could still earn an attractive return. Its owner needs new money anyway, at a moment when investors and lenders may have grown more selective.
At $5.5 million a month, the $60 million reserve covers about 10.9 months. Six of those are already committed to waiting for the first customer payment, which leaves roughly 4.9 months of room for delay. A slip that stays inside that buffer narrows the owner’s options, and one that runs past it forces a financing decision before the first invoice is paid. The same disruption can therefore be manageable for one owner and decisive for another, which is why revenue growth, operating profitability, and financial resilience need to be judged separately.
Contracted demand still has to be financed
Applied Digital’s disclosures show the same gap at real scale. On its October 7 earnings call, chief executive Wes Cummins described about $36 billion of contracted revenue across five campuses in three states, covering roughly 1.41 GW of critical IT load under lease. On November 2 he said the company has financed about 600 MW of that 1.4 GW plan, leaving roughly 800 MW to finance over the next 12 months. He also said the tenants on the unfinanced capacity are rated AA or better, and that construction loans, project finance, and the bond market all remain open to the company, at a higher cost now that Treasury yields and credit spreads have risen.
Nebius shows what the financing side of the clock can look like. On October 22 it announced a $3.5 billion senior secured facility from Citi, which it described as complementing a funding base of customer prepayments and operating cash flow, with asset-backed financing against more than $40 billion of long-term customer contracts still available to it. Its filing calls the agreement a bridge facility. It matures on October 16, 2027, carries a minimum liquidity covenant, and charges Term SOFR plus 2.75%, with the margin rising a quarter point for each 90 days the loan stays outstanding, up to 3.50%. The loan buys roughly a year of building, and its price steps up each quarter the company leaves it in place.
Neither disclosure is a verdict on either company’s prospects. Each shows the mechanism this essay is about: contracted demand and funded capacity arrive on separate schedules, and an owner has to finance the time between them.
The case that a slowdown helps some owners
The strongest counterargument is that slower construction could strengthen operators who already hold approved, powered sites. If demand stays strong while competing supply arrives late, a finished facility with a signed customer and committed financing could gain value against a larger announced pipeline that still needs permits, power, and capital. A more selective buildout could also improve how capital is allocated and bring delivery closer to what customers actually need, while easing competition for capacity that already exists. Reuters reported that many analysts read the election this way, saying a slower U.S. build did not necessarily signal weaker demand for AI or computing power.
Those outcomes would undercut any claim that infrastructure economics are deteriorating across the board, and they remain compatible with disappointment for shareholders whose valuations assumed faster expansion. The likelier result is uneven: fewer projects proceed, some existing assets gain value, and some shareholders absorb dilution or slower growth while the businesses beneath them keep operating.
Geography offers another release. New investment can move toward places with better power availability, faster approvals, or lower costs, preserving demand for globally supplied chips and equipment while changing which developers and regions benefit. HSBC analysts raised that possibility directly after the election, suggesting data centers might simply move from Texas to other states. Moving the next project does not automatically rescue the current one, though. Land, civil works, interconnection agreements, and construction commitments stay tied to the original site, so a company’s freedom to choose where it builds next tells investors little about whether it can recover what it has already committed.
What would change the argument
There is a genuine policy tension in all of this. Avoidable delays to well-funded, technically sound projects can raise costs for customers and slow domestic capacity growth, while predictable rules on grid costs, water, and local impact can make development more durable over time. The IEA’s recent work on energy and AI points to flexible grid connections, proactive management of data center project pipelines, and connection arrangements that shape network charges as ways to speed connections while protecting affordability. A blanket pause and a clear route to approval carry very different economic consequences, and how well states draw that line will help decide how much planned capacity is eventually built.
None of this requires AI demand to disappoint. AI can spread across more products and industries while the return on any single infrastructure investment varies widely, and a business can keep meeting every customer commitment while changes in growth expectations, margins, the timing of cash flows, or its share count reduce what its investors earn.
The useful test for any owner is whether it can finance the distance between promising capacity and getting paid for it. Answering that means tracking delivery dates beside funding dates and reading the timing and conditions of customer payments as closely as the headline contract values. Construction, fit-out, and ongoing services carry different cash profiles, and committed financing has to be read for what it actually requires of the borrower, maturity and covenants included.
My expectation would become less persuasive if projects keep converting contracted demand into cash, if delays stay inside funded buffers, and if financing stays available on workable terms. Evidence pointing the other way would include repeated commissioning slippage, shrinking liquidity cushions, costlier refinancing, or new equity issued to carry commitments that already exist.
AI demand can be right about the long-term forecast and still arrive too late for a particular balance sheet.
An election or an earnings report can move sentiment in a day. A project feels that change later, as it works through its contracts, payment dates, and construction milestones.
The hypothetical project in this essay has 4.9 months of room for delay. Few real projects publish theirs.
All essays