How much debt is funding the AI boom?
In our AI infrastructure market deck, you will find everything you need to understand the market
SUMMARY
How much debt is funding the AI boom? Our best current estimate is roughly half a trillion dollars, with a reasonable funded-debt range of $400 billion to $600 billion.
That estimate is different from the roughly $489 billion of AI-related debt issued this year. Issuance measures financing activity, while the $400 billion to $600 billion range estimates debt still supporting chips, cloud capacity, data centers and related infrastructure after adjusting for refinancing and overlap.
The larger financial exposure sits in leases. The five largest US hyperscalers have around $969 billion of future lease commitments, including about $662 billion for facilities whose leases have not yet begun, but those figures are future payment streams rather than cash borrowed upfront.
Debt entered the AI story because infrastructure is being built years before its full revenue arrives. Companies must secure land, power, networking and accelerators now, then hope utilization and customer payments catch up.
The strongest hyperscalers are not borrowing because they have run out of money. For Microsoft, Alphabet and, to a lesser degree, Meta, debt still provides flexibility; for Oracle, CoreWeave and independent data-center developers, external financing is increasingly essential.
CoreWeave shows how leveraged the outer edge of the boom has become. Its quarterly infrastructure spending exceeded operating cash generation by billions of dollars, while interest expense already consumed more than one quarter of quarterly revenue.
The debt total cannot be calculated by adding every announced bond, loan and project facility. The same campus may appear in construction debt, private-credit databases and a later bond refinancing, while general corporate borrowing can support AI alongside unrelated spending.
Private credit has become a central source of AI financing. More than $200 billion of outstanding private-credit loans are tied to companies classified as AI, big data or cloud technology, yet those unusually large loans appear to carry almost no additional pricing premium.
The main danger is a mismatch between the useful life of the collateral and the life of the obligation. A data-center shell may operate for decades, but GPUs can lose commercial value within a few product generations, long before the related loan or lease has matured.
AI can succeed technologically while its lenders lose money. Falling compute prices, stronger chips and more efficient models could accelerate adoption while weakening the economics of older clusters, overbuilt campuses and aggressively priced capacity contracts.
The wider financial system is not yet in crisis territory. Exposure remains concentrated, the largest customers have substantial cash flow, and most private-credit funds have limited portfolio exposure, although the links between banks, insurers, funds, project vehicles and hyperscaler guarantees are becoming harder to trace.
The boom therefore still runs on cash flow, equity and customer prepayments as well as debt. But its fastest-growing infrastructure layer is now clearly leveraged, and the weakest projects will eventually have to prove that demand, utilization and compute prices can cover the financing structures built around them.
How much debt is really funding the AI boom right now?
Roughly $500 billion of AI-related debt is being raised this year, and we estimate that about $400 billion to $600 billion of funded debt is already supporting the boom. Those figures describe different things. The first measures bonds and loans reaching the market during the year. The second is our estimate of the debt currently backing chips, data centers, cloud capacity and related infrastructure after allowing for refinancing and overlap.
The wider exposure is much larger. Moody’s says the five largest US hyperscalers have about $969 billion of future lease commitments, including roughly $662 billion for leases that have yet to begin. Those contracts create real financial pressure, although they stretch across many years and do not represent cash borrowed upfront.
Why did AI suddenly become a debt story?
AI became a debt story when infrastructure spending grew faster than even the richest technology companies wanted to fund from normal cash flow.
The spending numbers explain the shift. Goldman Sachs currently models about $765 billion of annual AI infrastructure spending across compute, data centers and power, rising to roughly $1.6 trillion by 2031. Its baseline adds up to $7.6 trillion between 2026 and 2031. Even if that forecast proves too high, AI is already far beyond the budget of a normal software expansion.
The assets also have to be bought before customers produce the revenue that pays for them. A company must secure land, power and networking equipment, install thousands of accelerators and often commit to a campus years before demand is fully visible. Waiting for each project to repay itself before starting the next one would slow expansion dramatically.
Debt lets companies build several waves of capacity at once. It also protects cash that management may prefer to keep for acquisitions, dividends, buybacks or a downturn. These days, the marginal AI campus is increasingly financed through bonds, equipment loans, project vehicles and leases even when the company behind it remains profitable.
This market map, featured in our AI infrastructure market deck, highlights top companies and startups in the AI infrastructure market
What should we actually count as AI debt?
We should count borrowed cash tied to chips, data centers, cloud capacity and supporting infrastructure, while keeping leases and customer commitments in a separate debt-like category.
Corporate bonds count when their proceeds help finance AI expansion, even if the legal documents say “general corporate purposes.” Bank loans and private-credit facilities count when they fund servers, GPUs, buildings or power connections. Project debt also counts when a separate vehicle borrows to construct a campus for a hyperscaler.
Leases need different treatment. A company may promise $20 billion of rent over 15 years, but the landlord does not hand it $20 billion on day one. The total includes financing costs, the owner’s return and payments for operating the site. We track that obligation because it can become painful, yet we keep it outside the funded-debt estimate.
Customer prepayments are different again. Oracle recently disclosed $75 billion of prepaid or customer-supplied hardware within large AI contracts. That money reduces Oracle’s financing need, but it does not become Oracle debt simply because it helps buy GPUs.
| Financing item | How we count it | Why |
|---|---|---|
| Corporate bonds and term loans | Funded debt | Cash is raised now and must be repaid |
| Equipment and GPU financing | Funded debt | The hardware directly supports AI capacity |
| Data-center project debt | Funded debt | A project vehicle borrows against the campus and tenant contract |
| Future lease payments | Debt-like exposure | Payments are fixed, but spread across many years |
| Customer prepayments | Financing, not debt | Customers provide cash or hardware without creating a repayment claim |
| Announced capital spending | Planned investment | A budget does not show how the spending will be financed |
Does $489 billion of AI debt mean the same amount of new data centers?
The AI debt headline captures a huge borrowing wave, but it overstates how much brand-new infrastructure has been financed.
A recent Goldman Sachs estimate counted $489 billion of AI-related issuance during the year, already above the $322 billion it counted for the whole previous year. Hyperscalers represented about 40% of the total. Goldman separately reported that four large technology companies had issued more than $170 billion of corporate debt, over four times their pre-AI annual average.
Part of that money replaces older debt. Part can sit in cash. Some bonds support cloud computing, acquisitions or ordinary corporate spending alongside AI. A data-center landlord may also host storage, cybersecurity and conventional enterprise workloads in the same buildings.
There is another trap. One project can appear several times as it moves through the financial system. A private lender may fund construction, a bank may lend to that private-credit vehicle, and the finished campus may later be refinanced with bonds. Adding every transaction would count the same asset more than once.
We use the figure as a measure of financing activity. It shows that AI has become one of the largest forces in global credit markets, while telling us little about the exact value of new servers and buildings.
If you want more recent data on this point, please see our latest AI infrastructure market report.
As this chart shows, and as featured in our AI infrastructure market deck, search interest in AI infrastructure has risen sharply
How much actual AI debt can we trace?
We can trace enough borrowing to support a current estimate of $400 billion to $600 billion, although no public database can produce an exact AI-debt total.
Our range starts with more than $200 billion of outstanding private-credit loans to companies classified by the Bank for International Settlements as AI, big data or cloud technology. On top of that sit large corporate bond programs, bank loans, equipment facilities and project financings that do not all appear in the BIS private-credit data.
Oracle offers one unusually clear example. Its latest annual results say it raised $43 billion of debt financing during its fiscal year to expand cloud infrastructure. CoreWeave ended the previous year with $21.6 billion of debt, then arranged a $4 billion convertible-note offering, a high-yield senior-note deal and another $3.1 billion delayed-draw facility during the following months. Meta’s Hyperion campus has about $27 billion of development costs, with part of Blue Owl’s capital funded through debt sold to institutional investors.
We cannot simply add those numbers. CoreWeave appears in several lending datasets, project debt can overlap with private credit, and some corporate issuance refinances earlier borrowing. After allowing for those overlaps, a total below $400 billion looks too small to fit the observable market. Moving beyond $600 billion requires us to count a broad share of general corporate borrowing or future leases.
| Observable financing layer | Current evidence | Our treatment |
|---|---|---|
| AI-related private credit | Already a major direct-lending category | Core anchor, adjusted for broad classifications |
| Large-tech corporate issuance | Running far above its pre-AI pace | Adjusted for refinancing and non-AI uses |
| CoreWeave debt and new facilities | More than $20B before its latest financings | Strong direct link, with overlap across instruments |
| Large project vehicles | Individual projects around $20B–$30B | Included where debt is funded or firmly committed |
| Our funded AI-debt estimate | $400B–$600B | Best current range after overlap adjustments |
Are Big Tech companies borrowing because they are running out of cash?
Microsoft, Alphabet and Meta still produce enough cash to fund a large share of AI spending, but the current buildout is too large for cash alone to remain the comfortable default.
Meta shows both sides of the story. Its latest first-quarter results reported $32.2 billion of operating cash flow and $19.8 billion of capital expenditure. It also raised its annual capex guidance to $125 billion to $145 billion. The advertising business can finance a great deal, though spending at that level quickly consumes the cushion that once funded buybacks and other projects.
Oracle is already in a tighter position. It generated $32 billion of operating cash flow in its latest fiscal year, spent enough to produce negative free cash flow of $23.7 billion, and raised $43 billion through debt. Customer prepayments now cover part of the hardware bill, but the company still expects another large mix of debt and equity financing in the next fiscal year.
Strong hyperscalers borrow because debt gives them speed and flexibility. More stretched infrastructure providers borrow because they cannot execute the plan without outside capital. Both groups appear in the same AI-debt totals, even though their credit risk is very different.
This chart, included in our AI infrastructure market deck, shows annual VC investment in AI infrastructure startups
Which AI companies actually depend on debt to keep expanding?
CoreWeave and Oracle currently rely much more heavily on external financing than Microsoft, Alphabet or Meta.
CoreWeave’s latest first-quarter filing showed $11.8 billion outstanding under delayed-draw term loans and $6.4 billion of notes before several later financing moves. During that quarter, it spent $7.7 billion on property and equipment, generated $3 billion of operating cash flow and received $3.9 billion from financing activities. Its infrastructure spending therefore exceeded operating cash generation by about $4.7 billion in only three months.
Since that filing, CoreWeave has completed a $4 billion convertible-note sale, issued senior notes carrying a 9.75% coupon and arranged a $3.1 billion delayed-draw loan mainly for GPUs and related infrastructure. Its first-quarter interest expense reached $536 million, more than one quarter of its $2.08 billion quarterly revenue. The company also reported a $99.4 billion revenue backlog, which explains why lenders remain willing to fund it.
Oracle has a mature software business, but its AI cloud plan is now larger than the cash that business can comfortably provide. The latest accounts show tens of billions of debt financing, $5 billion of equity financing and a $23.7 billion free-cash-flow deficit. Microsoft and Alphabet have much stronger internal funding positions, while Meta sits between the two groups because it combines huge cash generation with increasingly large project partnerships.
| Company or model | Current debt dependence | What the evidence says |
|---|---|---|
| Microsoft and Alphabet | Low | Cash flow still covers a large share of expansion |
| Meta | Moderate | Strong cash generation, plus large leases and project vehicles |
| Oracle | High for incremental AI capacity | Debt and equity are filling a clear funding gap |
| CoreWeave and similar neoclouds | Very high | Borrowing is built into the GPU acquisition model |
| Independent data-center developers | Very high | Projects depend on loans, tenant contracts and refinancing |
If you want more recent data on this point, please see our latest AI infrastructure market report.
Are giant AI data-center leases basically debt?
Giant AI data-center leases behave a lot like debt when the customer has little practical freedom to walk away.
Moody’s currently estimates about $969 billion of future lease commitments across the five largest US hyperscalers. Roughly $662 billion relates to facilities whose leases have yet to start. That second number has grown so large because companies are reserving campuses that are still being built.
The accounting label hides much of the economic relationship. A project vehicle borrows to construct a data center, while the technology company signs a long contract to use it. Lenders care about the tenant’s payments far more than the thinly capitalized vehicle that formally owes the debt.
Meta’s Hyperion arrangement makes the connection easy to see. Blue Owl funds own 80% of the $27 billion venture, Meta owns 20%, and Meta will lease all the facilities. Meta also supplied a residual-value guarantee that may require a capped payment if leases end and the campus is worth less than expected. A portion of the outside capital came from debt placed with PIMCO and other bond investors.
We keep those lease commitments outside our funded-debt estimate because they are an undiscounted stream of future rent. Still, ignoring them would understate how much of the AI buildout already depends on fixed payments from a small number of technology companies.
This chart, included in our AI infrastructure market deck, shows why CoreWeave is winning in AI infrastructure
Who is lending all this money besides bond investors?
Private-credit funds, insurers and structured-finance investors now provide a major share of AI infrastructure financing.
The BIS calculates that outstanding private-credit loans to AI-related companies have climbed from almost nothing to more than $200 billion, close to 8% of all private-credit lending. Funds originated over $40 billion of new loans to these companies in 2025, compared with about $3 billion in 2010.
The average AI-related private-credit loan is unusually large at about $169 million, versus $90 million for other borrowers. Yet average maturity and pricing look almost identical: 4.7 years and a 6.2 percentage-point spread for AI loans, compared with 4.8 years and 6.1 points elsewhere. Lenders are accepting nearly twice the average loan size without charging a visibly larger risk premium.
Data-center owners are also turning their leases into bonds. KBRA counted $48.69 billion across 88 US data-center ABS and CMBS transactions from 2018 through the first part of 2025. Newer deals are increasingly tied to hyperscale and AI campuses, with lease payments packaged into securities that pension funds, insurers and credit funds can buy.
Banks remain involved even when they disappear from the headline. They arrange bond sales, provide construction loans and lend to private-credit vehicles. Risk that appears to have left the banking system can therefore return through funding lines, guarantees or a difficult refinancing.
Is AI debt starting to look like the telecom bubble?
AI financing now resembles the late-1990s telecom buildout more than a normal technology investment cycle.
The pattern is familiar: demand looks enormous, companies rush to build capacity before rivals, lenders finance assets with long lives, and technological progress keeps changing what the best equipment looks like. Goldman Sachs estimates that hyperscaler spending would reach the telecom boom’s peak intensity at around $700 billion in one year. Current AI infrastructure estimates are already close to that scale.
AI demand could keep rising even if the financing disappoints, just as internet traffic kept growing after telecom investors lost money. Too many networks were financed at prices and utilization assumptions that could not survive competition.
AI could follow the same path. Model usage may grow quickly while the rental price of older GPUs falls. New chips may handle more work per dollar, leaving some financed clusters below their expected utilization. The technology can transform the economy while specific lenders, landlords and cloud providers earn poor returns.
Today’s boom has stronger borrowers than the telecom sector had in many cases. Microsoft, Alphabet, Amazon and Meta bring real cash flow and investment-grade credit. The fragile edge sits with projects that depend on one customer, one generation of hardware or repeated refinancing.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, shows annual funding in AI infrastructure startups
What would make AI debt go bad?
AI debt turns dangerous when compute prices or utilization fall before the loans, leases and guarantees have run off.
The hardware can age much faster than the debt. A data-center shell may last 20 years, while a GPU can lose commercial value within a few product cycles. CoreWeave’s delayed-draw facilities are secured partly by GPU servers and customer cash flows. If newer systems deliver much more output per dollar, the resale value of older collateral can drop faster than lenders expected.
A single customer can also carry most of a project’s economics. Several investors may own the debt even though one AI laboratory or hyperscaler supplies most of the revenue. If that customer delays deployment, renegotiates capacity or runs into its own financing problem, the shock travels quickly through the structure.
Refinancing arrives long before many campuses reach the end of their useful lives. The BIS finds an average maturity of 4.7 years for private-credit loans to AI-related companies. Many data-center campuses and leases run far longer. Borrowers therefore need fresh capital well before the physical asset reaches the end of its life.
Overbuilding could do the most damage. A project can have real demand and still earn too little if several competitors add capacity in the same region. Lower compute prices help AI users but hurt anyone who borrowed on the assumption that today’s rental rates would last.
Could AI keep booming while lenders lose money?
AI usage can keep growing rapidly while lenders and data-center investors suffer large losses.
We saw this with the internet. Traffic, online commerce and digital services kept expanding after the dot-com crash, even though many telecom networks, hosting companies and investors failed. Society gained the infrastructure at a lower cost after creditors absorbed part of the original bill.
The same split can happen with AI. Better chips and more efficient models may increase the amount of useful work produced while reducing the price customers pay for each unit of compute. Enterprises could adopt AI faster precisely because infrastructure owners are forced to cut prices.
Debt makes that gap harsher. An equity-funded company can wait for demand to catch up. A leveraged data center must meet interest and principal schedules. CoreWeave’s first-quarter interest bill of $536 million shows how quickly financing costs become a hard constraint even when revenue and backlog are growing.
We should judge the technology and the financing separately. Strong AI adoption would support the long-term need for compute, but it would not rescue every campus built in the wrong place, every old GPU fleet or every contract priced too aggressively.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, included in our AI infrastructure market deck, compares the main business model options for AI cloud infrastructure providers
Could AI debt threaten the wider financial system?
AI debt is creating real pockets of financial stress today, yet the wider financial system remains well short of crisis.
So far, the risk is concentrated rather than spread evenly across finance. The BIS says around 20% of private-credit funds have made at least one AI-related loan, but the average fund has only about 5% of its lending in the category. The largest hyperscalers also retain substantial cash flow, liquidity and access to equity markets.
The links are getting harder to see, though. Private-credit funds lend to project vehicles, banks finance those funds, insurers buy the bonds, and hyperscalers support the projects through leases or guarantees. A failure at one campus may stay contained. Several customer defaults or refinancing failures at the same time could activate guarantees and force lenders to sell similar assets together.
Bond and loan prices still look surprisingly relaxed. Goldman Sachs recently noted that the four largest technology issuers had sold more than $170 billion of debt while spreads remained near historic lows. The BIS found almost no pricing premium on private AI loans despite their much larger average size.
The calm looks too generous around the riskiest borrowers. Investors are still being paid almost the same spread for much larger private AI loans, and most of these structures have yet to face falling compute prices. We see growing fragility around neoclouds and single-tenant projects, while the strongest hyperscalers still keep the wider system away from crisis territory.
So how much debt is funding the AI boom?
The best current answer is roughly half a trillion dollars of funded AI debt, with a reasonable range of $400 billion to $600 billion and a much larger layer of future lease commitments.
As seen above, about $489 billion of broadly defined AI-related debt has reached the market this year. That is the freshest measure of the borrowing wave, although refinancing, general corporate use and repeated financing of the same assets prevent us from treating it as net new infrastructure debt.
Using the figures already examined, our funded-debt estimate rests on more than $200 billion in AI-related private credit, unprecedented corporate issuance, Oracle’s $43 billion debt raise, CoreWeave’s expanding loan and bond stack, and multibillion-dollar project vehicles. We discount those figures for overlap rather than adding them mechanically.
As noted earlier, the five largest US hyperscalers carry about $969 billion of future lease commitments, including around $662 billion for facilities that have yet to open. Those contracts push the wider financial exposure close to $1 trillion, even though they are not loans received today.
Debt now finances a large minority of the incremental AI infrastructure buildout. It remains optional financial flexibility for the strongest hyperscalers. For Oracle, CoreWeave, independent data-center developers and many project vehicles, borrowed money has become essential. The AI boom still runs on cash, customer prepayments and equity too, but its fastest-growing edge is now unmistakably leveraged.
If you want more recent data on this point, please see our latest AI infrastructure market report.
This chart, featured in our AI infrastructure market deck, shows the share of revenue generated by each customer segment in the AI infrastructure market
OUR METHODOLOGY
This analysis estimates how much funded debt currently supports the AI infrastructure boom. We separate debt raised during the year from debt still outstanding, and we keep future lease commitments outside the funded-debt total because they represent payments spread across many years rather than cash borrowed upfront.
We broke the question into several financing layers: corporate bonds and term loans, private-credit facilities, equipment and GPU financing, data-center project debt, securitized lease structures and long-term contractual obligations. Announced capital expenditure was not counted unless the available evidence showed how it would be financed.
We prioritized capital that had already been raised, remained outstanding or had been firmly committed. Company filings, completed financing announcements, loan data, cash-flow statements and project disclosures received more weight than broad estimates of future infrastructure spending.
No single market figure was treated as the total. Annual issuance measures how active the borrowing market has become, private-credit data captures one large but incomplete pool of outstanding loans, and company disclosures show how the financing reaches specific GPUs, cloud deployments and data-center campuses.
We did not add every disclosed transaction mechanically. A single project may first use construction financing, later appear in private-credit data and eventually be refinanced through bonds or securitization. We adjusted for that overlap, along with corporate borrowing that supports AI expansion alongside acquisitions or ordinary cloud investment.
Our $400 billion to $600 billion range was formed by comparing observable financing layers and testing whether market-level evidence matched company-level disclosures. A figure below that range does not fit the scale of private credit, corporate issuance and known project financing, while a figure above it would require counting a broad share of general corporate debt or future leases.
We evaluated debt dependence separately from debt size. Microsoft, Alphabet and Meta can still finance substantial expansion internally, so borrowing often provides speed and flexibility. Oracle, CoreWeave, neoclouds and independent data-center developers rely more directly on debt, equity or customer financing to continue building.
Lease obligations were analyzed as debt-like exposure rather than funded debt. They can create fixed, long-lasting payment obligations and support borrowing by project vehicles, but their undiscounted totals also include financing costs, property-owner returns and operating payments.
Key sources include the Bank for International Settlements on debt and private credit financing the AI boom, Goldman Sachs on projected AI infrastructure spending, and Goldman Sachs on large-technology-company debt issuance and credit-market pricing.
Company-level evidence came primarily from Oracle’s fiscal 2026 annual report, Oracle’s infrastructure funding plan, CoreWeave’s first-quarter 2026 filing, CoreWeave’s 2025 annual report, CoreWeave’s GPU-backed loan disclosure, and Meta’s first-quarter 2026 results.
For project financing, leases and securitization, we used Meta’s disclosure of the Hyperion partnership with Blue Owl Capital, KBRA’s research on data-center ABS and CMBS structures, and Fortune’s report on Moody’s hyperscaler lease-commitment analysis.
This chart, included in our AI infrastructure market deck, shows how GPU cloud infrastructure technology has evolved over time
Related blog posts
- Is the AI infrastructure boom a bubble?
