The Market Continues to Absorb AI Debt Supply, but the Autumn Financing Wave May Push Spreads Wider

Overview

This report tracks the issuance and relative value of global AI infrastructure financing across investment-grade bonds, high-yield bonds, leveraged loans, non-dollar bonds, ABS, CMBS, and CDS markets. Its core conclusion is that enormous capital expenditure needs have not materially slowed despite wider summer spreads. Deep investment-grade markets and high-quality issuers' balance sheets can continue to support financing, but persistent supply will primarily be absorbed through wider spreads; risk-return differences among financing structures are also expanding.

Core views

AI financing supply continues to grow rapidly. Based on data through the close on August 19, 2026, global credit markets have completed approximately $445 billion of AI-related debt issuance year-to-date in 2026, with full-year issuance expected at $500 billion to $550 billion; issuance totaled approximately $54 billion in July and approximately $41 billion in August through the report period. Since July alone, approximately $92 billion of new AI-related issuance has entered global corporate credit markets, while more than $20 billion of data center-backed debt entered the market in July. Supply has not stalled despite wider spreads, indicating that overall market capacity remains ample, although investors are increasingly price-sensitive and wider spreads have become the primary adjustment mechanism enabling markets to continue funding the AI capital expenditure cycle. Capital expenditure expectations underpin the report's view that supply will not disappear quickly. Including SPCX, aggregate 2027 capital expenditure estimates have approached $1.4 trillion. Even considering only five hyperscalers, their estimated 2027 capital expenditures are nearly double the estimate from November 2025, while expectations for capital expenditures over the next 12 months have nearly tripled from a year ago. The report therefore argues that even if the pace of hyperscaler debt issuance temporarily slows, a brief spread tightening should not be interpreted as the disappearance of supply pressure. As spending shifts from data center shells toward chips, servers, computing equipment, and power, highly rated issuers may also make greater use of AAA- or AA-rated credit to provide credit support for lower-quality, unrated, or privately held companies and attract private capital. The US investment-grade market remains the primary financing channel. The market is approximately $10 trillion in size, is dominated by long-term institutional investors, and has structural demand for high-quality income-producing assets. AI-related debt currently represents approximately 7% of the investment-grade index, while the report views approximately 15% as the maximum sector exposure investors may accept, indicating that substantial absorption capacity remains. Approximately 61% of year-to-date AI issuance has been absorbed by the US public investment-grade market. Public investment-grade AI-related issuance has reached approximately $271 billion and is expected to total $350 billion to $400 billion for the full year. Ten relevant companies account for approximately 7% of the public investment-grade index but around 30% of the S&P 500 Index. Hyperscalers have $431 billion of index-eligible debt, with other large AI companies adding approximately $110 billion. Measured by duration risk, AI transactions have driven a 55% year-over-year increase in overall investment-grade issuance. The balance sheets of high-quality hyperscalers can still accommodate substantial financing volumes. They have issued approximately $223 billion year-to-date, including about $66 billion of non-dollar debt, while bond maturities from the report period through year-end 2027 total only about $22.5 billion. The group's gross leverage is approximately 1.3x, net leverage remains well below 1x, and it has ample liquidity, double-digit EBITDA growth, and strong operating cash flow. The principal constraints arise from rapidly increasing off-balance-sheet long-term leases and long-term purchase commitments for computing capacity, chips, equipment, and power. Rating agencies include these items in adjusted debt, although disclosure remains limited. If such commitments continue to increase, companies' capacity to issue debt while maintaining current ratings will decline. The report believes expected project returns make high-quality issuers relatively insensitive to the current rise in spreads, but lower-quality borrowers will be more materially affected. Credit pricing already reflects concentrated supply and risk exposure. High-quality hyperscaler spreads have widened by approximately 30 basis points year-to-date and are currently about 7 basis points wider than the investment-grade index, while the investment-grade index itself has widened by only around 2 basis points during the year. Some long-dated META bonds are even as much as 50 basis points wider than the BBB curve. NVDA and AVGO bonds with maturities of seven to ten years trade broadly in line with high-quality hyperscalers, while BBB-rated AI companies show a similar direction but wider absolute spreads. The report expects volatility to persist and views buying hyperscaler CDS protection as its preferred way to express the AI theme in credit derivatives markets. Liquidity in the relevant CDS improved materially after their inclusion in CDX in March 2026. Because AI credit risk is concentrated among a small number of companies, financial institutions' demand for non-economic hedging to manage counterparty risk will rise. The AI basket is currently approximately 40 basis points wider than CDX IG, and MSFT five-year CDS is the only constituent still trading tighter than CDX IG. Between unsecured corporate bonds and asset-backed financing, the report prefers collateral-backed risk. Collateral-backed debt in the public and private investment-grade markets has exceeded $100 billion. Investment-grade data center bonds trade approximately 100 basis points wider than the investment-grade index and 50 to 200 basis points wider than tenants' unsecured bonds of comparable maturity, with transaction structure being the principal source of spread differentiation. The report is constructive on the potential for loan-to-value ratios to decline after construction is completed and believes political resistance and power bottlenecks may restrain new project supply in the near term. Its preferred structural features include high-quality tenants, residual value guarantees, full amortization, and provisions allowing rent commencement upon delivery of the building shell with less dependence on power interconnection. Chip financing can reduce reliance on ultra-long-duration asset risk through full amortization and residual value guarantees. Non-dollar markets have become an important source of funding. Of more than $200 billion of hyperscaler bond issuance in 2026, over one-quarter came from non-dollar markets, and non-dollar issuance represented a majority in the second quarter of 2026. Euros accounted for only 40% of year-to-date non-dollar supply, with issuers also using Canadian dollars, pounds sterling, Swiss francs, Australian dollars, and Japanese yen, covering six non-dollar currencies in total. Except for a recent Australian-dollar transaction, these bonds have generally been longer-dated than their respective local investment-grade benchmarks, while Alphabet's yen transaction became the largest yen-denominated bond on record by a foreign issuer. Because only Alphabet and Amazon currently use non-dollar markets, sector concentration outside the dollar market remains relatively low. In performance terms, euro-denominated hyperscaler bonds have underperformed the broader European market since early June; the belly of the curve now trades closer to the A-rated rather than the AA-rated index, and the long end is unusually steep. However, after cross-currency adjustment, euro bonds still trade tighter than corresponding dollar bonds. Leveraged financing is also accelerating but is more sensitive to funding costs. AI-related high-yield bond issuance has totaled approximately $40 billion year-to-date in 2026, while leveraged loan issuance has totaled around $10 billion, approximately $7 billion of which was concentrated in July and August. Around 82% of data center high-yield bonds use secured or project finance-style structures. AI exposure represents only approximately 3% of the high-yield market, up from less than 1% in the prior year. Relative to an assumed 15% concentration limit, the market still has capacity for additional issuance. Meanwhile, excluding AI, year-to-date high-yield bond supply has declined 15% year over year, showing that AI data center transactions are an important source of incremental supply. Higher financing costs may delay or cancel lower-rated transactions. Once projects have stabilized, high-yield construction debt can be refinanced through ABS, CMBS, or even investment-grade markets, while chip and other asset-level financing may create more opportunities for loans and private credit. Differentiation among high-yield data center bonds has widened substantially. BB-rated data center-backed bonds currently trade approximately 180 basis points wider than the BB index, compared with a year-to-date average of only 80 basis points wider. The market now includes 18 project finance-style high-yield bonds covering seven tenants. High-yield data center bonds with investment-grade tenants or investment-grade credit support still trade 100 to 250 basis points wider than tenants' unsecured bonds, while bonds backed by data centers leased by ORCL trade approximately 175 to 260 basis points wider than ORCL unsecured bonds of comparable duration. The report believes the wider spread of high-yield bonds relative to investment-grade data center debt primarily reflects greater construction and refinancing risk. Securitized credit has not yet repriced as visibly as corporate bonds. The data center securitization market added approximately $2.2 billion of transactions in July, bringing year-to-date issuance to around $18 billion, equivalent to approximately 120% of supply during the same period in 2025. ABS spreads have been relatively stable, which the report attributes to the market having already repriced supply factors, lower secondary-market liquidity and less price discovery, lower securitized supply intensity than in corporate bonds, and underlying assets that have already stabilized and are less exposed to construction outcomes and policy changes. Nevertheless, the report believes risks are skewed toward wider spreads from current levels, and securitized spreads may soften when new issuance resumes and secondary trading increases in the autumn. From a regulatory perspective, relevant legal opinions conclude that data center ABS in which assets and leases are transferred into an SPV do not constitute Exchange Act ABS as defined under Section 3(a)(79) of the US Securities Exchange Act of 1934; this conclusion does not apply to CMBS in which mortgage loans are transferred into an SPV. At the single-name level, the report specifically calls for attention to ORCL's and AVGO's September results and to the new financing structures represented by the recent NVDA financing platform. The report expects credit support from highly rated hyperscalers to play a larger role in private capital participation as AI development shifts from building shells toward chips and servers. The construction, residual asset value, tenant, guarantee, and refinancing risks ultimately borne by different creditors will increasingly depend on specific terms rather than merely the issuer's name or headline rating.

Analysis framework

The report first aggregates global AI-related debt issuance through August 19, 2026, and compares year-to-date progress with full-year estimates, the same period in 2025, and capital expenditure expectations. It then assesses financing sustainability through market capacity, issuer leverage and liquidity, off-balance-sheet commitments, and maturity profiles. The relative-value section compares spreads and post-issuance performance across investment-grade bonds, high-yield bonds, loans, CDS, non-dollar bonds, ABS, and CMBS, and further explains pricing differences among structures using tenant credit quality, collateral, loan-to-value ratios, amortization methods, residual value guarantees, and construction status.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Analysis of Credit Supply and Market Absorption Capacity

  The report combines AI financing issuance volumes, index weights, institutional investor demand, and an assumed concentration ceiling to assess whether the market can continue absorbing new debt and how wide spreads must be to clear supply.

• Fixed Income and Credit AnalysisSpread analysis

  Cross-Rating, Cross-Structure, and Cross-Currency Relative-Value Comparison

  The report compares spreads between corporate bonds and ABS, investment-grade and high-yield debt, secured debt and tenants' unsecured debt, and euro and dollar bonds to identify pricing differences caused by supply, liquidity, construction risk, and transaction structure.

• Corporate Fundamentals and Financial FrameworkOperating/Financial Leverage Analysis

  Assessment of Hyperscaler Balance-Sheet Capacity

  The report combines gross leverage, net leverage, liquidity, EBITDA growth, operating cash flow, bond maturities, and off-balance-sheet commitments to assess how much additional debt issuers can bear while maintaining their current ratings.

• Event-Driven Strategy and Behavioral FinanceExpectation Gap/Expectation Management

  Ongoing Revisions to Capital Expenditure and Issuance Expectations

  The report tracks upward revisions to 2027 and next-12-month capital expenditure estimates relative to earlier forecasts and uses them to conclude that the market's repricing of future debt supply may not yet be complete.

• (Out-of-Vocabulary Method)

  Collateralized Financing Transaction Structure Analysis

  Based on loan-to-value ratios, tenant quality, residual value guarantees, full amortization, rent commencement provisions, construction progress, and refinancing pathways, the report distinguishes data center and chip financings that have similar headline ratings but different underlying risks.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• High-Quality Hyperscaler Bonds

  They are the principal public-market channel for financing AI capital expenditures, with ample balance-sheet capacity, but concentrated supply is pushing spreads wider.

  Strengths

  Low leverage, ample liquidity, double-digit EBITDA growth, strong operating cash flow, and high credit ratings.

  Weaknesses

  Rising off-balance-sheet leases and long-term purchase commitments, with market risk concentrated among a small number of large issuers.

  Comparison

  Spreads have widened by approximately 30 basis points year-to-date and are currently about 7 basis points wider than the investment-grade index.

  Risks

  Persistent supply, increasing off-balance-sheet debt, and slower-than-expected realization of returns on AI investment could further weigh on performance.

• Hyperscaler CDS Protection

  The report views this as its preferred way to express AI risk in the credit derivatives market.

  Strengths

  Liquidity has improved following inclusion in CDX, while concentrated risk may create sustained counterparty hedging demand.

  Weaknesses

  CDS spreads could tighten again if issuance falls below expectations or AI commercialization progresses faster than expected.

  Comparison

  The AI basket is approximately 40 basis points wider than CDX IG; MSFT five-year CDS is the only relevant instrument still tighter than CDX IG.

  Risks

  Lower supply or improving fundamentals could adversely affect protection positions.

• Collateral-Backed Data Center and Chip Financing Debt

  The report relatively prefers this risk over unsecured corporate bonds of tenants or guarantors.

  Strengths

  Collateral, full amortization, residual value guarantees, and high-quality tenants can mitigate some credit risk.

  Weaknesses

  Specific risks are highly dependent on construction progress, power interconnection, rent commencement provisions, and transaction documents.

  Comparison

  Investment-grade data center bonds trade approximately 100 basis points wider than the investment-grade index and 50 to 200 basis points wider than tenants' unsecured bonds.

  Risks

  Construction delays, power bottlenecks, policy resistance, tenant risk, and declining residual asset values.

• AI-Related High-Yield Data Center Bonds

  They are an important source of growth in 2026 high-yield supply but are more sensitive to funding costs, construction risk, and refinancing risk.

  Strengths

  Approximately 82% use secured or project finance structures and can transition to ABS, CMBS, or investment-grade refinancing after operations stabilize.

  Weaknesses

  Lower-rated issuers face higher financing costs, while construction-stage and refinancing dependence drive price differentiation.

  Comparison

  BB-rated data center bonds trade approximately 180 basis points wider than the BB index, versus a year-to-date average of only 80 basis points wider.

  Risks

  Rising financing costs may delay or cancel transactions, while construction or refinancing failures could magnify losses.

• Data Center ABS and CMBS

  Underlying assets are generally stabilized, and recent performance has been steadier than corporate credit, but the report expects additional repricing in the autumn.

  Strengths

  They are less directly affected by construction outcomes and policy changes, and current supply intensity is lower than in corporate bonds.

  Weaknesses

  Secondary-market liquidity is lower and price discovery is limited, while actual demand following a recovery in supply remains untested.

  Comparison

  Year-to-date issuance is approximately $18 billion, about 120% of the same period in 2025, but recent spread changes have been materially smaller than in corporate bonds.

  Risks

  Spreads may soften after new issuance resumes and secondary trading increases in the autumn; ABS and CMBS are subject to different regulatory definitions.

• Non-Dollar Hyperscaler Bonds

  They help issuers diversify funding sources and have already gained significant weights in several local investment-grade indices.

  Strengths

  They span six currencies—the euro, Canadian dollar, pound sterling, Swiss franc, Australian dollar, and Japanese yen—and market absorption capacity remains resilient.

  Weaknesses

  Long-dated supply is heavy, and the intermediate and long ends of European curves have performed weakly.

  Comparison

  Euro bonds have underperformed in local markets but still trade tighter than corresponding dollar bonds after cross-currency adjustment.

  Risks

  Large, long-dated issuance may increase local index concentration and spread volatility.

Key data

• Global AI-Related Debt IssuanceApproximately $445 billionYear-to-date in 2026, through the close on August 19, 2026
• 2026 Global Issuance Forecast$500 billion to $550 billionThe report's full-year estimate
• July and August IssuanceApproximately $54 billion; approximately $41 billionAugust figure is month-to-date through the report period
• New Corporate Credit Issuance Since JulyApproximately $92 billionGlobal AI-related corporate credit
• Aggregate 2027 Capital Expenditure EstimateNearly $1.4 trillionIncludes SPCX-related estimates
• Size of the US Investment-Grade MarketApproximately $10 trillionUsed by the report to assess market absorption capacity
• AI Debt Weight in the Investment-Grade IndexApproximately 7%The report believes approximately 15% may be the maximum sector exposure acceptable to investors
• Share of Issuance in the US Public Investment-Grade MarketApproximately 61%Share of year-to-date AI-related issuance
• Hyperscaler Index-Eligible Debt$431 billionOther large AI companies add approximately $110 billion
• Hyperscaler Year-to-Date IssuanceApproximately $223 billionIncluding approximately $66 billion of non-dollar debt
• Hyperscaler LeverageGross leverage of 1.3x, net leverage below 1xUsed to assess balance-sheet financing capacity
• Year-to-Date Change in Hyperscaler SpreadsApproximately 30 basis points widerCurrently approximately 7 basis points wider than the investment-grade index
• AI CDS Basket Spread Versus CDX IGApproximately 40 basis points widerThe report prefers buying CDS protection
• Size of Collateral-Backed DebtMore than $100 billionPublic and private investment-grade markets
• Investment-Grade Data Center Bond Spread Versus the IndexApproximately 100 basis points wider50 to 200 basis points wider than tenants' unsecured debt
• AI-Related High-Yield Bond IssuanceApproximately $40 billionYear-to-date in 2026
• AI-Related Leveraged Loan IssuanceApproximately $10 billionApproximately $7 billion of which occurred in July and August
• AI Debt Weight in the High-Yield IndexApproximately 3%Less than 1% in the prior year
• BB-Rated Data Center Bond Spread Versus the BB IndexApproximately 180 basis points widerYear-to-date average of approximately 80 basis points wider
• Data Center Securitization IssuanceApproximately $18 billion year-to-dateApproximately $2.2 billion added in July; year-to-date issuance is approximately 120% of the same period in 2025

Impact & implications

The report believes the constraint on AI infrastructure financing is not an immediate loss of debt issuance capacity among high-quality issuers, but rather that supply must be absorbed at higher credit spreads. Highly rated hyperscalers can still use their balance sheets and credit support to broaden funding sources, but lower-rated projects, projects under construction, and projects with greater refinancing dependence are more vulnerable to rising funding costs. Collateral, tenant quality, guarantees, and amortization terms will become increasingly important. Securitized credit is currently relatively stable but may catch down after supply resumes in the autumn.

Risks

• Persistent large-scale AI capital expenditures and bond supply may push corporate credit spreads wider.
• Hyperscalers' long-term leases and purchase commitments for computing capacity, chips, equipment, and power may reduce their debt issuance capacity while maintaining current ratings.
• Political resistance, power bottlenecks, and construction delays may affect data center project commissioning, rent commencement, and collateral values.
• Higher financing costs may cause lower-rated borrowers to delay or cancel new transactions and increase refinancing risk for construction-stage projects.
• Securitized credit spreads may soften after new issuance resumes and secondary trading increases in the autumn.
• For the view favoring the purchase of CDS protection, issuance below estimates or faster-than-expected AI commercialization that strengthens fundamentals could cause CDS spreads to tighten.

What to watch

• Track the pace of AI-related debt issuance in the autumn and whether spreads on corporate bonds, ABS, and CMBS continue to widen.
• Monitor whether the nearly $1.4 trillion estimate for 2027 capital expenditures is revised further.
• Observe the impact of hyperscalers' off-balance-sheet leases and long-term purchase commitments on adjusted debt and ratings headroom.
• Monitor new financing structures in which highly rated issuers provide credit support to privately held, unrated, or lower-quality AI companies.
• Track data center projects' power interconnection, final delivery, rent commencement, and progress in transitioning from construction financing to ABS, CMBS, or investment-grade debt.
• Monitor the results reported by ORCL and AVGO in September 2026.
• Observe whether the NVDA financing platform and financing for components such as chips and servers drive increased loan and private-capital issuance.
• Track the maturity structure, cross-currency relative value, and local index concentration of non-dollar hyperscaler bonds.