How the AI Boom Is Financed: Data Centers and Debt

Introduction

The artificial intelligence boom is, beneath the model launches and the demos, one of the largest construction projects in history, and someone has to pay for it. Building the data centers, buying the chips, and supplying the power costs staggering sums: the largest cloud companies alone are on track to spend roughly $725 billion of capital expenditure in 2026, and Goldman Sachs expects a cumulative $5.3 trillion of AI-related capital spending between 2025 and 2030. The striking part is not only the scale but where the money comes from. For the first time in a major technology cycle, the build-out is increasingly funded with debt rather than spare cash.

That shift is reshaping credit markets and generating both fees and risk for banks. When companies move from paying for growth out of pocket to borrowing for it at this magnitude, every corner of finance is pulled in: debt capital markets desks underwriting bonds, private credit funds writing huge loans, and structured-finance teams inventing ways to borrow against chips. Follow that money and the AI boom comes into sharper focus than any product launch provides, including why the sky-high valuations of the AI companies themselves rest on the very bet the lenders are now financing, the debt-funded model, the telecom-bust echoes, and the stakes for everyone in finance all included.

To understand the financing, you first have to grasp how much money is involved, because the numbers break the usual frame of reference.

Hyperscaler capex like nothing before

The "hyperscalers", the handful of cloud giants leading the build-out, are each committing more than $100 billion a year to capital spending, for a combined total around $725 billion in 2026. That is a scale of investment normally seen only in capital-intensive industries like utilities and telecoms, not in software. These companies now spend an estimated 45% to 57% of their revenue on capex, a ratio that would have been unthinkable for an asset-light technology business a decade ago.

Hyperscaler

One of the small group of cloud-computing giants (such as Amazon, Microsoft, Google, and Meta) that operate data centers at massive scale. Their combined capital spending on AI infrastructure runs into the hundreds of billions of dollars a year and anchors the entire AI build-out.

Why their cash flow is not enough

Even companies generating enormous profits cannot fund spending at 45%-plus of revenue out of operating cash flow indefinitely. That is the crux of the story: the AI build-out has outgrown the cash the build-out's own sponsors produce. Morgan Stanley and J.P. Morgan project that the technology sector may need to issue as much as $1.5 trillion of new debt over the next few years to fund it. The shift from cash-funded to debt-funded infrastructure is a structural change with no precedent in prior tech investment waves, and it is what drags the credit markets, and the bankers who run them, into the center of the AI story.

The real bottleneck is power

Increasingly the hard constraint is not money or chips but electricity. A large AI data center can draw as much power as a small city, and grid connections, generation capacity, and cooling have become the gating factors on how fast the build-out can proceed. That has pulled financing into a much wider set of assets: power plants, grid upgrades, and even long-term nuclear and small-modular-reactor deals struck to lock in supply. For finance, the AI build-out is therefore not only a technology-debt story but an infrastructure and energy-financing one, drawing in project-finance and power teams alongside the technology bankers. The capital required to power the machines may ultimately rival the capital required to build them.

Master the credit and capital-markets technicals: Practice over 1,000 questions on debt, valuation, and deal structures, download our iOS app to prepare for every angle.

The money reaches the data centers through four main channels, each with its own risk profile and each creating different work for finance.

Corporate balance sheets and bonds

The first and cheapest channel is the hyperscalers' own balance sheets. With strong credit ratings and huge profits, they can borrow in the investment-grade bond market at relatively low rates, and they have begun issuing debt in size to supplement their cash. This is bread-and-butter debt capital markets work, and the flood of AI-driven issuance has become one of the busiest corners of the bond market. Even these blue-chip borrowers are not immune to the sheer scale, though. Rating agencies have begun warning that sustained, debt-funded capex could pressure the pristine credit profiles that make the borrowing cheap in the first place, and a downgrade, or even the threat of one, would raise the cost of the entire channel. The biggest technology balance sheets can absorb a great deal, but they are not bottomless, which is part of why the financing has spilled so quickly into the other three channels.

Private credit steps in

Where bank balance sheets and bond markets will not stretch, private credit has rushed in. Funds run by Blackstone, Blue Owl, Apollo, Pimco, and BlackRock now originate much of the data-center debt, and their loans to AI-related projects have surged from near zero to more than $200 billion in a few years, with Morgan Stanley projecting another $800 billion could come from private credit over the next two years. These lenders can move faster and structure more flexibly than banks, which is exactly why they have become central to financing the build-out, and it is a major driver of the broader growth of private credit and direct lending.

GPU-collateralized debt and the neoclouds

The most novel channel is borrowing against the chips themselves. A new class of "neocloud" companies, which rent out AI computing power, fund their GPU purchases with loans secured by those very GPUs and by their customer contracts. The landmark example is CoreWeave, which closed an $8.5 billion facility, the first investment-grade-rated financing backed by AI infrastructure, according to the company. The catch is subtle and important: the investment-grade rating comes not from the neocloud's own credit, which is speculative-grade, but from the creditworthiness of the blue-chip customer whose contract underpins the loan.

Neocloud

A specialized cloud provider that rents out AI computing power, typically by buying large quantities of GPUs and leasing access to them. Neoclouds such as CoreWeave often finance those chip purchases with debt secured by the GPUs and by customer contracts, a structure new to credit markets.

Off-balance-sheet joint ventures

A growing share of the build-out never touches the hyperscalers' own balance sheets at all. Rather than borrow directly, the cloud giants increasingly form joint ventures and special-purpose vehicles with infrastructure funds, asset managers, and other private-capital partners, who put up much of the money to build and own the data centers while the hyperscaler signs a long-term lease for the capacity. The structure shares the enormous capex, keeps some of the associated debt off the technology company's books, and matches long-lived assets with the patient capital of pension funds and insurers. Goldman Sachs expects private markets to take a growing role in exactly this kind of data-center financing. For analysts, these vehicles are a reminder that a company's true economic exposure to the AI build-out can be larger than its reported balance sheet suggests.

Vendor and circular financing

The fourth channel is the most controversial: the chipmakers and cloud providers themselves help fund the AI developers that are their biggest customers. Nvidia invests in and supplies AI labs; those labs commit hundreds of billions to cloud providers; the cloud providers buy Nvidia chips to deliver the capacity. Cash loops among a handful of interconnected firms, and 2026 analyses put the total of such arrangements at more than $800 billion.

Circular Financing

An arrangement in which a supplier funds, or invests in, the customers that then use the money to buy the supplier's own products. In AI, chipmakers and cloud providers finance the AI labs that become their largest buyers, which can make demand and revenue look more robust and organic than they are.

Mapping the loop makes the concern concrete, and several outlets have charted how the same dollars circulate among a handful of giants.

The financing engine is impressive, but it rhymes uncomfortably with a famous boom that ended badly, and that parallel is the heart of the bubble debate.

The vendor-financing parallel

In the late 1990s, telecom-equipment makers like Lucent and Nortel lent money to the startups buying their gear, which inflated apparent demand until the buyers collapsed and the vendor financing imploded, helping trigger the dot-com bust. Today's circular AI deals look structurally similar: a supplier booking revenue from customers it is itself bankrolling. If end demand for AI fails to match expectations, the same dollars circling the loop can magnify losses rather than spread risk.

The cash-burn problem underneath

Beneath the financing sits a hard fact: many of the AI developers at the center of the loop lose enormous sums. OpenAI alone has been reported as on track to lose around $14 billion in 2026, even as it targets $100 billion of revenue by 2029. Lending and investing heavily into businesses that burn cash on this scale is a bet that the revenue arrives before the financing runs out, which is exactly the kind of company our guide on valuing businesses with no profits is built to assess.

For bankers and investors, the AI build-out is one of the defining business opportunities, and risks, of the decade.

A fee and lending bonanza

The debt-funded model is lucrative for finance. Debt capital markets desks underwrite the hyperscaler bonds, leveraged-finance and private-credit teams arrange the project and GPU-backed loans, and structured-finance specialists design the vehicles that make speculative-grade borrowers investment-grade on paper. The sheer volume of leveraged finance and structured debt tied to AI has become a meaningful share of activity, and the firms that can underwrite and syndicate it are earning accordingly.

The exposure that comes with it

The flip side is concentration. Banks, private-credit funds, and increasingly the insurers and pension funds whose money sits behind private credit are all building large, interconnected exposures to a single theme financed largely with debt. If AI demand or economics disappoint, the losses would not stay contained to equity investors; they would run through the credit system that funded the build-out. That is why regulators and risk officers now watch AI-related lending closely, and why understanding the financing is not optional for anyone in markets.

Go deeper on debt and deal structures: Download our comprehensive 160-page PDF, access the IB Interview Guide covering capital markets, leveraged finance, and credit analysis.

The AI financing story is increasingly a current-events question for capital-markets, credit, and even generalist interviews, because it ties a headline theme to real financial mechanics.

The prompt you are likely to get

An interviewer might ask "how is the AI build-out being paid for?" or "what worries you about AI financing?" A strong answer moves past the equity hype to the debt: the shift from cash-funded to debt-funded capex, the roles of bond markets and private credit, the novelty of GPU-collateralized lending, and the circular-financing risk with its telecom-bust parallel. Naming a concrete figure or deal, the roughly $725 billion of 2026 capex or the CoreWeave facility, signals you actually follow the market.

Holding both sides at once

The deeper test is whether you can hold both sides: that the demand is genuine and the opportunity enormous, and that financing a cash-burning, interconnected ecosystem with debt concentrates risk in dangerous ways. Candidates who can explain why the same structure that accelerates the build-out also amplifies the downside demonstrate exactly the commercial judgment that markets desks hire for, the same instinct behind reading any convertible or structured financing.

How the AI boom is financed may matter as much to markets as the technology itself. A build-out funded by profits is self-limiting; a build-out funded by debt and circular capital can run faster and further, but it also borrows against a future that has to arrive on schedule. Whether this ends as the infrastructure backbone of a new era or as the next great vendor-financing cautionary tale will depend on whether AI demand grows into the trillions of dollars of debt now being raised against it. Either way, the money behind the machines is a story every banker should understand.