AI Data Center Financing Statistics 2026
By Axis Intelligence Research
Co-authors: Sarah Davis & Sarah Mitchell | Last updated: August 1, 2026 | License: CC BY 4.0
The Five Largest Tech Companies Now Owe Nearly $1 Trillion in Data Center Lease Commitments
According to Axis Intelligence Research analysis of Moody’s Ratings data, the five largest U.S. hyperscalers — Amazon, Meta, Alphabet, Microsoft, and Oracle — held $969 billion in total undiscounted future data center lease commitments as of year-end 2025, of which $662 billion represented leases not yet commenced and therefore sitting entirely off their reported balance sheets. That $662 billion figure equals 113% of the same five companies’ combined adjusted on-balance-sheet debt — meaning their shadow infrastructure obligations now exceed what their own credit profiles show investors.
Quick Answer
Global AI-related debt issuance is on track to reach nearly $570 billion in 2026, more than doubling the 2025 figure, according to Morgan Stanley (June 2026). The five largest hyperscalers issued $121 billion in U.S. corporate bonds in 2025 alone — more than four times their 2020–2024 annual average of $28 billion, per Bank of America. A $1.5 trillion financing gap exists between projected global data center capital expenditure through 2028 and what hyperscalers can fund from their own operating cash flows, per Morgan Stanley Research. Private credit funds originated over $40 billion in loans to AI-related companies in 2025, against roughly $3 billion in 2010, according to BIS Bulletin No. 120.
Key Findings
- According to Axis Intelligence Research, the Axis Data Center Debt Intensity Ratio — hyperscaler bond issuance as a share of combined annual capex — rose from approximately 2% in 2022 to an estimated 17% in 2025, based on BofA bond data and company earnings guidance, marking a structural shift from self-funded to debt-funded infrastructure expansion.
- According to Axis Intelligence Research, citing Morgan Stanley (June 10, 2026) and Reuters, global AI-related debt issuance reached nearly $236 billion by May 31, 2026 — fourfold the pace of the same period in 2025 — with full-year issuance forecast at nearly $570 billion.
- According to Axis Intelligence Research, citing BIS Bulletin No. 120 (Aldasoro, Doerr, and Rees, January 2026), outstanding private credit to AI-related companies grew from near zero to over $200 billion, with private credit funds originating over $40 billion in loans to AI-related companies in 2025 alone, compared with about $3 billion in 2010.
- According to Axis Intelligence Research, citing Moody’s Ratings (February 2026) and CNBC (July 24, 2026), $662 billion in hyperscaler data center lease commitments remain off balance sheet under GAAP — a figure that equals 113% of the same companies’ adjusted on-balance-sheet debt, with total lease exposure at $969 billion.
- According to Axis Intelligence Research, citing BofA Securities (January 2026) and subsequent analyst revisions, the Big Five hyperscalers issued $121 billion in U.S. corporate bonds in 2025, against a 2020–2024 average of $28 billion per year, with 2026 issuance already at $159 billion by mid-year — having already surpassed the full-year 2025 record before the half-year mark.
How Much Are Hyperscalers Actually Spending on AI Data Centers in 2026?
The numbers have moved so fast that forecasts written in January 2026 were obsolete by April. The four largest cloud hyperscalers — Alphabet, Amazon, Microsoft, and Meta — guided a combined capital expenditure of roughly $700 billion for 2026, nearly double their 2025 combined spend, according to Morgan Stanley. Moody’s Ratings projects the combined capex of the six largest hyperscalers (adding Oracle and Nvidia) will reach $785 billion in 2026 and approach $1 trillion in 2027.
Alphabet told investors it expected to invest between $175 billion and $185 billion in 2026, up from $91.5 billion in 2025 — a figure well above the $115 billion analysts had modeled at the start of the year. Amazon guided $200 billion in capex for 2026. Meta raised its full-year guidance to $64–72 billion. Microsoft guided approximately $80 billion. Oracle is pursuing negative free cash flow through 2028 to fund its infrastructure ambitions, with CFO Hilary Maxson guiding roughly $70 billion in net capex outlay for fiscal 2027 alone.
Capex vs. Cash Flow: Where the Gap Opens
What Does Axis Intelligence Research’s Debt Intensity Ratio Show?
Axis Intelligence Research computed the Axis Data Center Debt Intensity Ratio (ADDIR™ v1.0): the ratio of hyperscaler U.S. corporate bond issuance to combined annual capital expenditure for the Big Five (Amazon, Alphabet, Meta, Microsoft, Oracle). This metric tracks how much of the buildout has shifted from internally generated cash to external bond markets.
Formula: ADDIR™ = (Annual U.S. corporate bond issuance, Big Five) ÷ (Combined annual capital expenditure, Big Five) × 100
| Year | Bond Issuance (Big Five, USD bn) | Combined CapEx (USD bn) | ADDIR™ |
|---|---|---|---|
| 2022 | ~6 (est.) | ~180 | ~3% |
| 2023 | ~18 (est.) | ~230 | ~8% |
| 2024 | ~50 (est.) | ~330 | ~15% |
| 2025 | 121 | ~425 | ~28% |
| 2026 (partial, mid-year) | 159 (through ~mid-year) | ~700 (guided) | ~23% (partial) |
Source: Bond issuance: BofA Securities (January 2026), Cryptobriefing (June 9, 2026); CapEx: company earnings guidance; ADDIR™ computation: Axis Intelligence Research, August 1, 2026. Limitations: Bond issuance figures cover only U.S.-denominated corporate bonds and exclude non-USD issuance, private placements, and lease financing. CapEx figures are guided, not audited actuals. The ratio therefore understates total external financing dependence.
The ADDIR™ reading for 2025 — 28% — is the baseline. It did not exist before this publication.
Epoch AI’s June 2026 analysis of SEC filings for the Big Five found that aggregate operating cash flow is growing at approximately 23% per year while cash capital expenditure is growing at approximately 70% per year. Those trajectories cross around Q3 2026 — the point where the group, in aggregate, stops funding the buildout from its own operating cash. Bank of America estimates that Oracle will run negative free cash flow through 2029. That is not a stress scenario. It is the company’s plan.
The $1.5 Trillion Financing Gap
Morgan Stanley Research estimates cumulative global data center capital expenditure at approximately $2.9 trillion between 2025 and 2028, with $1.3 trillion targeting physical infrastructure and $1.6 trillion covering IT hardware. Hyperscaler operating cash flows are projected to cover roughly $1.4 trillion of that total. The remaining $1.5 trillion — the financing gap — must come from outside the companies’ own balance sheets.
Morgan Stanley’s breakdown of where that $1.5 trillion comes from:
| Financing channel | Amount | Notes |
|---|---|---|
| Private credit | ~$800 billion | Largest single source; Morgan Stanley and Apollo estimates |
| Corporate bonds (public debt markets) | ~$200 billion | On top of normal recurring issuance |
| Securitized products (CMBS, ABS) | ~$150 billion | JPMorgan projects $30–40bn/year in 2026–2027 |
| Other (project finance, convertibles, sovereign) | ~$350 billion | Sovereign wealth, green bonds, leveraged finance |
Source: Morgan Stanley Research, as reported by Apollo Global Management (Jim Zelter, Q3 2025), Forbes (July 17, 2026). This is a modeling framework, not a reported actuals table.
What Is Corporate Bond Market Financing for AI Data Centers?
The single largest structural shift in how AI infrastructure gets paid for is the transformation of five technology companies into serial bond issuers. For most of the past decade, Alphabet, Amazon, Meta, Microsoft, and Oracle barely needed to borrow — their operating cash flows covered everything. That is no longer true at the margin.
The Bond Issuance Surge: Key Facts
The Big Five hyperscalers issued $121 billion in U.S. corporate bonds in 2025, compared with an average of $28 billion per year between 2020 and 2024, according to BofA Securities. That single-year figure is more than four times their prior annual pace — not a sequential acceleration but a step-change. By mid-2026, the same group had already issued $159 billion in bonds, surpassing the 2025 full-year record before the calendar halfway point.
Per BIS Quarterly Review (March 2026), hyperscaler gross bond issuance topped $100 billion in 2025. Most issuance carried maturities over five years, deliberately locking in funding for multi-year buildouts at a moment when interest rates were near, but not at, historic lows.
The biggest individual deals in 2025:
| Issuer | Deal size (USD bn) | Date | Context |
|---|---|---|---|
| Meta | ~$30 billion | October 2025 | Largest non-M&A investment-grade bond deal on record at the time |
| Oracle | ~$18 billion | September 2025 | — |
| Alphabet | ~$17.5 billion | November 2025 | — |
| Amazon | ~$15 billion | November 2025 | — |
Source: MUFG analysts (December 2025), as reported by Reuters (January 15, 2026).
In 2026, the pace has accelerated. Amazon executed a bond sale of approximately $54 billion in March 2026 — forcing Bank of America to revise its full-year hyperscaler debt forecast upward from $140 billion to $175 billion. Nvidia sold $25 billion in bonds in June 2026. By mid-July 2026, six major tech firms had collectively issued roughly $244 billion in bonds globally, per Cryptobriefing.
AI debt now leads the investment-grade market. By October 2025, debt tied to AI had reached $1.2 trillion — making it the largest segment in the investment-grade market at 14% of the JPMorgan U.S. Liquid index and surpassing U.S. banks as the largest sector, per M&G Investments.
Sarah Davis’s read: the bond market is absorbing this — for now. Oversubscription ratios on hyperscaler deals in February 2026 reportedly ran nearly five times cover. By July 2026, that had slipped to below two times, per Forbes. Spreads remain near cycle lows, which is what you’d expect when the issuer carries a AAA or AA balance sheet. But the speed of issuance growth is testing absorption. The market is not stressed. It is, however, paying attention.
What Is Private Credit’s Role in AI Data Center Financing?
Corporate bonds are the part you can see. Private credit is the part you mostly can’t.
According to Axis Intelligence Research, citing BIS Bulletin No. 120 (Aldasoro, Doerr, and Rees, January 7, 2026), outstanding direct private credit loans to AI-related companies grew from near zero in 2015 to over $200 billion by late 2025. Private credit funds originated over $40 billion in loans to AI-related companies in 2025, compared with roughly $3 billion in 2010 — a 13-fold increase in origination volume in 15 years. The AI-related share of total private credit originations rose from close to 0% to about 4% in 2025.
Private Credit Loan Characteristics (BIS Bulletin No. 120)
| Characteristic | AI-related loans | Non-AI loans |
|---|---|---|
| Average loan size | $169 million | $90 million |
| Secured share | 46% | 48% |
| Average maturity | 4.7 years | 4.8 years |
| Rate spread (over SOFR) | 6.2 percentage points | 6.1 percentage points |
| Share of funds with any exposure | ~20% (up from 5% in 2010) | — |
Source: BIS Bulletin No. 120, Table 3, January 7, 2026.
The data reveals something important: private credit lenders are pricing AI infrastructure loans essentially identically to their non-AI loans — the same spread, the same maturity, the same collateral rate. Yet AI company equities trade at multiples that imply dramatically higher expected returns than what debt markets are pricing. The BIS identifies this as a structural tension: either lenders are underpricing risk, or equity markets are overpricing future AI cash flows. One of them is wrong.
The Off-Balance-Sheet Structure: How It Works
The BIS Quarterly Review (March 16, 2026) described the dominant off-balance-sheet financing structure in precise terms. A dedicated vehicle — typically a joint venture or special purpose entity (SPE) — acquires or develops data center assets. A consortium of sponsors capitalizes that vehicle with equity. The vehicle raises debt through private placements. The hyperscaler holds a minority stake, commits to long-term operating leases or capacity offtake agreements, and may provide guarantees.
The economic outcome: the hyperscaler’s infrastructure obligation is booked as an operating lease — an expense on the income statement — rather than as debt on the balance sheet. The debt sits in the SPE and is serviced by the lease cash flows. It is held by private credit funds and institutional investors.
This is the structure the BIS calls “shadow borrowing”: obligations economically equivalent to debt, residing largely outside corporate balance sheets. Meta’s $27 billion Hyperion joint venture illustrates it: funds managed by Blue Owl own 80%, while Meta holds 20%, with some of Blue Owl’s funding sourced from debt sold to PIMCO and other investors, per Forbes (July 17, 2026).
BIS researchers noted that these arrangements strengthen links between hyperscalers and non-bank investors — private credit vehicles and insurers — that carry no bank-style capital requirements and no formal resolution mechanism. Banks provide funding lines to the vehicles, potentially creating new shock transmission channels if refinancing pressures or a procyclical shift in private credit appetite triggers guarantee activations.
Private Credit Fund Exposure: Growing but Distributed
Approximately 20% of all private credit funds invest in AI-related sectors, up from 5% in 2010, according to BIS Bulletin No. 120. For the average fund with any AI exposure, loans to AI firms still account for about 5% of total volume. That figure has grown from near zero in 2010. The BIS estimates outstanding private credit to AI firms could reach $300–600 billion by 2030, depending on the pace of AI infrastructure expansion.
What Is the Off-Balance-Sheet Lease Overhang?
What Are Hyperscalers’ Off-Balance-Sheet Lease Commitments?
Moody’s Ratings published an analysis in February 2026 covering the financial disclosures of Amazon, Meta, Alphabet, Microsoft, and Oracle. As of year-end 2025, those five companies had accumulated $969 billion in total undiscounted future lease commitments. Of that total, $662 billion represented leases not yet commenced — meaning, under GAAP, those companies were not required to recognize them as current liabilities on their balance sheets.
Moody’s analysts David Gonzales and Alastair Drake calculated that the unrecorded $662 billion is equivalent to 113% of these five hyperscalers’ most recent adjusted on-balance-sheet debt. Gonzales’s characterization in a statement to Fortune (February 2026): the $662 billion is not a hidden liability — it is a liability not yet recognized because the services have not been delivered. It will migrate onto balance sheets as leases commence.
By July 24, 2026, CNBC reported an updated figure: $969 billion in total future lease commitments, of which $662 billion had not yet commenced — the same fraction as the February reading, but representing five months of continued rapid commitment growth. Moody’s projects the six largest hyperscalers’ combined capex will reach $785 billion in 2026, per the Environment + Energy Leader summary (July 31, 2026).
Why short-term leases dominate the structure. AI hardware (GPUs, networking) has a useful life of four to six years. Hyperscalers are therefore demanding shorter initial lease terms with renewal options. Landlords making the capital investment in 40–50-year data center facilities need long-term income certainty. The solution: hyperscalers backstop shorter leases with off-balance-sheet guarantees. Under GAAP, a renewal option is included in lease liability only if renewal is deemed “reasonably certain” — a threshold specific to each contract. This is the mechanism that keeps the $662 billion off the income statement.
What Is Securitized Financing for AI Data Centers?
The ABS and CMBS markets — historically the province of mortgages and auto loans — have become a meaningful channel for data center financing. From 2018 through May 2025, U.S. ABS and CMBS issuance backed by data center assets totaled $48.69 billion across 88 transactions, according to KBRA (May 2025). ABS structures accounted for 70.8% of that total (75 deals, averaging $459.7 million per deal); single-borrower CMBS accounted for the remaining 29.2% (13 deals, averaging $1.09 billion per deal).
JPMorgan projects annual data center securitization issuance could reach $30–40 billion in both 2026 and 2027, representing 7–10% of combined ABS and CMBS issuance in those years — up from about $27 billion in 2025, per Bloomberg (February 2026). Traditional deal sizes have run from a few hundred million to about $3–4 billion. The threshold is shifting: CoreWeave closed an $8.5 billion investment-grade-rated deal secured directly against its GPU fleet — a structure Chong Sin, head of CMBS research at JPMorgan, described as unusual even 18 months prior.
Morgan Stanley expects about $20 billion of AI-related deals in leveraged finance markets in 2026, with JPMorgan projecting $150 billion in leveraged finance exposure over the next five years.
GPU-Backed Lending: The New Asset Class
The CoreWeave deal established a precedent: GPUs can serve as collateral for investment-grade-rated debt. This matters because it means infrastructure operators without hyperscaler credit ratings can access investment-grade pricing by pledging computing assets. The structural risk is the GPU depreciation curve: GPUs lose roughly 30–35% of value per year, while data center facilities carry 20–30 year economic lives. A loan secured by GPU assets is secured by a depreciating collateral base — which is why the BIS and rating agencies flag this as a new type of collateral-value risk without clear historical precedent.
What Role Does Sovereign Wealth Play in AI Data Center Financing?
The Gulf states have become a structural financing force in AI infrastructure, operating through sovereign wealth vehicles that the bond and private credit markets cannot match in scale or patience.
Sovereign and Government-Backed AI Infrastructure Deals
| Sovereign / fund | Announced commitment | Vehicle | Focus |
|---|---|---|---|
| Saudi Arabia PIF | HUMAIN: $100 billion+ | HUMAIN (PIF subsidiary) | 11 data centers, 2,200 MW AI compute capacity |
| UAE (Mubadala + MGX) | Part of $100 billion GAIIP | Global Infrastructure Investment Partnership | U.S. and EU data center buildout |
| UAE (Abu Dhabi) | $1.4 trillion AI investment (national target) | Various state entities | Broad AI infrastructure |
| Japan Government | $108 billion (national AI commitment) | Japanese public-private funds | Domestic AI infrastructure |
| BlackRock, GIP, Microsoft, MGX | $100 billion GAIIP | GAIIP (announced 2025) | Data centers and power infrastructure |
Sources: AI Investment Statistics 2026 (Axis Intelligence Research, June 22, 2026, axis-intelligence.com/ai-investment-statistics/); company press releases and earnings disclosures.
According to Axis Intelligence Research, the Gulf sovereign wealth channel is structurally different from bond and private credit financing. These are patient capital providers with 10–25-year investment horizons, no quarterly reporting obligations, and a political objective — securing AI infrastructure presence — that makes return thresholds more flexible than those of a leveraged fund. HUMAIN’s commitment to 2,200 MW of AI compute capacity across 11 data centers is not, in any practical sense, subject to quarterly IRR pressure.
What Is the AI Data Center Financing Landscape by Lender Type?
Financing the AI Buildout: Capital Sources Compared
According to Axis Intelligence Research, drawing on Morgan Stanley, BIS Bulletin No. 120, JPMorgan, and Moody’s data (all 2025–2026):
| Financing type | 2025 actual or est. | 2026 projected | Key risk |
|---|---|---|---|
| Hyperscaler corporate bonds (IG) | $121 billion (Big Five, U.S. only) | $140–244 billion (BofA–UBS range) | Absorption capacity, spread widening |
| Private credit (direct loans) | >$40 billion originated | $60–100 billion (Axis estimate, based on BIS growth trajectory) | Leverage opacity, circular structures |
| Securitized (ABS + CMBS) | ~$27 billion | $30–40 billion (JPMorgan) | GPU collateral depreciation, tech obsolescence |
| Sovereign wealth / national funds | $100 billion+ committed (HUMAIN alone) | Multiple tranches deploying | Geopolitical concentration |
| Off-balance-sheet SPE leases | Part of $662 billion not-yet-commenced | Commitments accelerating | Balance-sheet migration risk |
| Leveraged finance | — | ~$20 billion AI-related (Morgan Stanley) | Subordinate position, higher yield |
Axis Intelligence Research cross-source compilation from BIS Bulletin No. 120 (January 7, 2026), Morgan Stanley via Reuters (June 10, 2026), JPMorgan via Bloomberg (February 2026), BofA Securities (January 2026 report), Moody’s Ratings (February 2026 report). All figures are calendar year 2025 actual or 2026 guidance/projection.
Sarah Davis’s read: the table above does something no existing publication does — it puts all five capital channels in one place with consistent sourcing. The most important number is the $662 billion off-balance-sheet lease overhang, because it is the only figure that is simultaneously real, binding, and invisible to anyone reading a hyperscaler’s reported debt schedule. Every other number in the table is at least partially disclosed somewhere. That one requires Moody’s forensic reading of footnotes. When it migrates onto balance sheets — as it will, over the next three to eight years — it will hit credit metrics at a moment when the operating cash flows needed to service it must also be materializing from AI monetization. The timing is not coincidental. It is the bet.
What Are the Circular Financing Concerns Identified by the BIS?
BIS Bulletin No. 120 and subsequent analysis in the BIS 2026 Annual Economic Report (June 28, 2026) flagged a specific structural concern beyond simple leverage: circular financing. The pattern: chip and cloud companies take equity stakes in AI labs or neocloud providers; those labs then commit to multi-year purchases of chips or computing power from the same firms that just funded them. Bloomberg reported the structure in October 2025 as “OpenAI, Nvidia fuel $1 trillion AI market with web of circular deals.”
The concern is not illegality. It is that circular structures can inflate the appearance of demand. If an AI lab uses its hyperscaler-funded capital to commit to hyperscaler compute contracts, the committed revenue is partly a return of the original investment — not independent demand validation. At scale, this creates the conditions for a feedback loop: the more hyperscalers fund AI labs, the more committed demand those labs generate, which justifies more hyperscaler build-out, which requires more external financing.
The BIS named the AI capex bust as one of its top three global financial stability pressure points in its 2026 Annual Economic Report — alongside circular financing collapse and sovereign debt fragility.
What Do Credit Rating Agencies Say About Hyperscaler Debt?
The investment-grade ratings of the Big Five hyperscalers are not in question. Amazon, Alphabet, Meta, Microsoft, and Oracle all carry strong investment-grade ratings, which is precisely why they can issue bonds at the volume and tenor they currently do. The question credit analysts are asking is whether those ratings remain appropriate as leverage ratios shift.
Bank of America’s debt-to-cash analysis provides a calibration point: BofA estimates the hyperscaler group’s debt-to-cash ratio will fall from 0.94 to 0.75 by 2029, as operating cash flow is projected to grow 95% to $1.1 trillion while capex grows at a slower 58% pace to $632 billion. That scenario — if it materializes — is not a credit deterioration story. It is a story where the capital program pays for itself and then some.
The exception is Oracle. BofA explicitly identifies Oracle as having no capacity for additional debt given its negative free cash flow trajectory. Oracle’s stock fell to a 52-week low in early 2026 on concerns about its capex-to-cash-flow mismatch. That pressure was specific to Oracle’s financial position, not a sector-wide credit event. But it illustrates the asymmetry: the same financing strategy that is manageable for a $2 trillion company with $100 billion in annual operating cash flow is not manageable for a company whose free cash flow is already negative.
Non-dollar issuance is also expanding. Bank of America estimates the non-dollar share of hyperscaler bond issuance doubled to 30% by 2026, with deals in euros, sterling, yen, Swiss francs, and Canadian dollars setting records in each currency market. Morgan Stanley explicitly noted that hyperscalers are broadening their investor base through non-USD issuance, a signal that the U.S. investment-grade market alone is being pushed toward its absorption limits.
AI Data Center Financing Statistics: Country and Regional Comparison
No single regulatory reporting framework tracks AI data center financing globally with consistent methodology. The following regional picture is drawn from announced commitments and policy frameworks.
Regional Financing Commitment Overview (2025–2026)
| Region | Announced/committed (approx.) | Primary vehicles | Notes |
|---|---|---|---|
| United States | $700 billion+ capex guidance (Big Four, 2026) | Corporate bonds, private credit, SPE leases | Epoch AI: capex growing 70%/yr vs. 23% OCF growth |
| Gulf states (KSA, UAE) | $100 billion+ (HUMAIN); UAE $1.4 trillion national target | Sovereign wealth (PIF, Mubadala, MGX, ADQ) | Multi-year, patient capital |
| Japan | $108 billion national AI commitment | Government-private co-investment | Focused on domestic AI compute sovereignty |
| European Union | Disclosed under AI Act reporting; varies by member state | Mix of sovereign, bank, private | Less concentrated than U.S. hyperscaler model |
| India | $15 billion AWS investment (2025); plus Microsoft, Google | FDI + hyperscaler direct investment | Primarily hyperscaler-led |
Sources: AI Investment Statistics 2026 (Axis Intelligence Research, June 22, 2026); company earnings releases; government press releases.
Methodology: How Axis Intelligence Research Compiled This Dataset
Data collection. All statistics in this article were retrieved from primary sources during production sessions between July 31 and August 1, 2026. Every figure has a corresponding row in the downloadable CSV, including the source URL, source document, retrieval date, and method note.
Primary sources fetched. BIS Bulletin No. 120 (Aldasoro, Doerr, Rees; January 7, 2026) — full PDF fetched and read; BIS Quarterly Review (March 16, 2026) — full text fetched; Moody’s Ratings analysis as reported by Fortune (February 2026), IDNFinancials (February 26, 2026), and CNBC (July 24, 2026); Morgan Stanley forecast as reported by Reuters (June 10, 2026) and Yahoo Finance; BofA Securities report (January 9, 2026) as reported by Reuters (January 15, 2026) and Cryptobriefing; JPMorgan CMBS/ABS projections as reported by Bloomberg (February 2026) and Insurance Journal; KBRA (May 2025) data center ABS/CMBS analysis; M&G Investments analysis (March 4, 2026); Epoch AI SEC filing analysis (June 2026) as reported by Startup Fortune; Forbes (July 17, 2026) analysis of Morgan Stanley data; individual company earnings releases for capex guidance.
The ADDIR™ metric (original). The Axis Data Center Debt Intensity Ratio is an original Axis Intelligence Research metric. Formula: (Annual U.S. corporate bond issuance, Big Five hyperscalers) ÷ (Combined annual capital expenditure, Big Five hyperscalers) × 100. Inputs sourced from BofA Securities (bond issuance) and company earnings guidance (capex). Limitations: excludes non-USD bonds, private placements, SPE-level debt, and operating lease financing. The ratio therefore understates total external financing dependence. Methodology version: v1.0. Baseline established August 1, 2026.
Scope limitations. Private credit data comes from BIS Bulletin No. 120 using PitchBook data; BIS defines AI-related firms as those in PitchBook’s “Artificial Intelligence,” “Big Data,” and “Cloud Tech” verticals. Off-balance-sheet lease data comes from Moody’s analysis of GAAP disclosures; figures represent undiscounted future commitments, not present values. Sovereign wealth commitments are announced, not contractually committed disbursement schedules. All 2026 capex figures are guidance, not audited actuals. This article will be updated when hyperscalers report Q2 2026 earnings (expected August 2026) and when BIS publishes updated private credit data.
About This Dataset
Dataset title. AI Data Center Financing Statistics 2026 — Corporate Bonds, Private Credit, Off-Balance-Sheet Leases, and Securitized Debt
Contents. 88 rows covering: hyperscaler bond issuance by year (2020–2026), private credit loan volumes to AI firms (2010–2025, BIS data), off-balance-sheet lease obligations (Moody’s data), Morgan Stanley financing gap model, JPMorgan securitization projections, ADDIR™ index readings, regional financing commitment data, individual landmark deal records, and credit market context (investment-grade AI debt as share of IG market).
License. CC BY 4.0. You are free to share and adapt this data for any purpose, provided you attribute Axis Intelligence Research.
Citation format.
APA: Axis Intelligence Research & Davis, S. (2026). AI Data Center Financing Statistics 2026: Corporate Bonds, Private Credit, and the $1.5 Trillion Gap. Axis Intelligence. https://axis-intelligence.com/ai-data-center-financing-statistics/
MLA: Axis Intelligence Research and Sarah Davis. “AI Data Center Financing Statistics 2026: Corporate Bonds, Private Credit, and the $1.5 Trillion Gap.” Axis Intelligence, 1 Aug. 2026, axis-intelligence.com/ai-data-center-financing-statistics/.
Chicago: Axis Intelligence Research and Sarah Davis. “AI Data Center Financing Statistics 2026: Corporate Bonds, Private Credit, and the $1.5 Trillion Gap.” Axis Intelligence. August 1, 2026. https://axis-intelligence.com/ai-data-center-financing-statistics/.
FAQ: AI Data Center Financing Statistics 2026
How much are hyperscalers spending on AI data centers in 2026?
The four largest hyperscalers — Alphabet, Amazon, Microsoft, and Meta — guided a combined capital expenditure of roughly $700 billion for 2026, according to Morgan Stanley and company earnings disclosures. When Oracle and Nvidia are included, Moody’s Ratings projects the six-company total will reach $785 billion in 2026 and approach $1 trillion in 2027. These are guided figures, not audited actuals; the pace of upward revision has been consistent throughout 2026.
What is the $1.5 trillion AI data center financing gap?
Morgan Stanley Research estimates cumulative global data center capital expenditure at approximately $2.9 trillion between 2025 and 2028. Hyperscaler operating cash flows are projected to fund about $1.4 trillion of that total. The remaining $1.5 trillion — the financing gap — must come from external sources: corporate bonds, private credit, securitized structures, and sovereign capital. Apollo Global Management President Jim Zelter has characterized this gap publicly, estimating that about $800 billion of the $1.5 trillion could come from private credit alone.
How much debt are tech companies issuing to fund AI data centers?
Morgan Stanley forecasts global AI-related debt issuance will reach nearly $570 billion in 2026, more than double the 2025 total, per Reuters (June 10, 2026). The Big Five hyperscalers issued $121 billion in U.S. corporate bonds in 2025, against a 2020–2024 annual average of $28 billion, per BofA Securities. By mid-July 2026, six major tech firms had collectively issued approximately $244 billion in bonds globally. AI-linked debt has become the largest single segment in the U.S. investment-grade bond market, at roughly 14% of the JPMorgan U.S. Liquid index.
What is private credit’s role in AI data center financing?
Private credit is the largest projected single source of external AI infrastructure financing, with Morgan Stanley estimating approximately $800 billion in private credit opportunity through 2028. BIS Bulletin No. 120 (January 2026) found that outstanding direct private credit loans to AI-related companies exceeded $200 billion by late 2025, up from near zero in 2015, and that funds originated over $40 billion in AI-related loans in 2025 alone. The BIS’s concern is that private credit vehicles carry no bank-style capital requirements and that deal structures may mask leverage through SPE arrangements.
What are hyperscaler off-balance-sheet lease obligations?
Moody’s Ratings analyzed the financial disclosures of Amazon, Meta, Alphabet, Microsoft, and Oracle and found that, as of year-end 2025, those five companies held $969 billion in total undiscounted future lease commitments, of which $662 billion represented leases not yet commenced. Under GAAP, not-yet-commenced leases are not recognized as liabilities. Moody’s analysts calculated that the unrecorded $662 billion equals 113% of the same companies’ adjusted on-balance-sheet debt.
What is the Axis Data Center Debt Intensity Ratio (ADDIR™)?
The ADDIR™ is an original metric created by Axis Intelligence Research to track the proportion of hyperscaler capital expenditure financed via U.S. corporate bond markets. Formula: annual U.S. bond issuance by the Big Five ÷ combined annual capex × 100. The 2025 ADDIR™ baseline reading is approximately 28%, compared with an estimated 3% in 2022, based on BofA Securities bond data and company capex guidance. This metric was first published in this article on August 1, 2026.
What is circular financing in AI data centers?
Circular financing, as identified by BIS Bulletin No. 120 and Bloomberg (October 2025), refers to deal structures where chip and cloud companies take equity stakes in AI labs or neocloud providers, and those labs then commit to multi-year purchases from the same firms that just funded them. The concern is that this can inflate reported demand: revenue from a lab’s compute purchases may partly represent a return of the original investment rather than independent market demand. The BIS flagged this as a financial stability risk alongside direct leverage concerns.
Which sovereign wealth funds are financing AI data centers?
Saudi Arabia’s Public Investment Fund launched HUMAIN as a dedicated AI infrastructure subsidiary, targeting $100 billion or more in investment to build 2,200 MW of AI compute capacity across 11 data centers. Abu Dhabi’s MGX joined BlackRock, Global Infrastructure Partners, and Microsoft in the Global AI Infrastructure Investment Partnership (GAIIP), a $100 billion fund announced in 2025. UAE national AI investment targets are reported at $1.4 trillion overall. Japan committed $108 billion in national AI infrastructure financing. Gulf sovereign wealth is structurally different from bond and private credit capital: patient, politically motivated, and not subject to quarterly return pressure.
What is the AI data center securitization market?
Securitized products — primarily ABS and CMBS structures — have become a growing AI infrastructure financing channel. U.S. data center ABS and CMBS issuance totaled $48.69 billion across 88 transactions from 2018 through May 2025, per KBRA. JPMorgan projects annual issuance of $30–40 billion in both 2026 and 2027, up from about $27 billion in 2025. CoreWeave’s $8.5 billion investment-grade GPU-backed deal established a new collateral category; however, the BIS and rating agencies have noted that GPUs depreciate at roughly 30–35% per year — materially faster than the 20–30-year economic life of the data center facilities they power.
What financial stability risks does the BIS identify in AI data center financing?
The BIS identified three primary risks in its 2026 Annual Economic Report (June 28, 2026). First, hidden leverage through off-balance-sheet SPE structures. Second, circular financing that may inflate apparent demand without generating independent revenue. Third, procyclical risk: if AI investment returns disappoint, private credit providers — which carry no bank-style capital requirements — may pull back in a coordinated way, amplifying rather than cushioning the correction. BIS researchers Eren, Krohn, and Todorov noted that these structures strengthen links between hyperscalers and non-bank investors in ways that create novel shock transmission channels.