YTD US dollar financing for AI infrastructure reaches $338bn, with supply growth expanding from hyperscalers to data centers, chip financing, and securitization

Overview

The report systematically tracks 2026 AI and digital infrastructure issuance across investment-grade bonds, high-yield bonds, chip financing, and the ABS/CMBS markets. Its core conclusion is that financing demand and issuance channels are expanding in tandem: hyperscalers remain the largest source of supply, data center developers and GPU financing are gradually following, and the securitization market is growing rapidly. However, rising issuance volumes could pressure pricing, and financing costs have already become a material constraint for some high-yield issuers.

Core views

As of the report's publication, Barclays tracked approximately $338bn of US dollar financing for AI and digital infrastructure YTD 2026, spanning the investment-grade, high-yield, ABS, and CMBS markets and including data center and chip financing. Approximately $118bn of new issuance has been added since the tracking report was first published on June 9; by comparison, full-year 2025 financing totaled $179bn. This indicates that AI infrastructure capital expenditure has created large-scale financing demand across asset classes and is no longer confined to conventional corporate bonds issued by a handful of technology companies. Hyperscalers remain the primary source of supply. YTD 2026, their cross-currency investment-grade issuance totals approximately $248bn, including $182bn in the US dollar market, nearly twice the 2025 level. Although the report believes issuers' fundamental 2026 financing needs have largely been met, they may still issue opportunistically to lock in funding in advance. Barclays forecasts approximately $285bn of full-year issuance, while its investor survey indicates an expected range of $275bn to $325bn. The YTD maturity profile is relatively balanced: approximately 32% in the 1-6 year segment, 32% in the 7-15 year segment, and 36% in maturities over 15 years. Approximately $88bn has been issued at 30-year and longer maturities, also accounting for roughly 36% of the total. Hyperscalers account for 14% of YTD US dollar investment-grade corporate bond supply, or approximately 17% when data center bonds are included. Their issuance is equivalent to 21% of investment-grade nonfinancial corporate issuance, while technology accounts for 37% of the latter, which would represent a record for a single sector if sustained. Barclays therefore believes that the pace and volume of issuance over the next 12 months will be key drivers of market performance. There are also significant differences among issuers. AMZN has issued $92.401bn across currencies YTD, including $62bn in US dollars and $30.401bn in foreign currencies. GOOGL has issued $80.313bn, including $45bn in US dollars and $35.313bn in foreign currencies. META, ORCL, and SPCX have each issued $25bn, while MSFT has issued nothing. The issuance of each of AMZN and GOOGL exceeds the combined $75bn issued by META, ORCL, and SPCX, and the former two have used foreign-currency transactions more frequently, showing that large issuers are broadening their funding sources through currency and maturity diversification. Data center developers are also shifting from traditional construction loans and project finance toward investment-grade and high-yield bonds. Since the previous update, the market has added two investment-grade transactions totaling $16.5bn and five high-yield transactions totaling $8.9bn. YTD 2026 data center bond issuance totals approximately $76.2bn, comprising $33.6bn of high-yield and $42.6bn of investment-grade issuance, equivalent to 15% and 3%, respectively, of total YTD issuance in their corresponding US dollar markets. Full-year 2025 issuance was only $35.9bn, including $8.6bn of high-yield and $27.3bn of investment-grade issuance. Market structures are becoming more standardized: high-yield transactions typically use an approximately 5NC2 structure with partial amortization, while investment-grade transactions place greater emphasis on construction-risk mitigation and fully amortize. Issuance slowed temporarily during the summer but has rebounded as the development pipeline has expanded. Most developers are becoming repeat issuers, and the report tracks a near-term project pipeline of approximately 4.0GW in the high-yield market. Refinancing of existing bonds and loans, new project entry, and developer expansions collectively support further growth in subsequent supply. Chip and GPU financing represent the primary source of incremental growth in the next phase. Since the previous update, Nebius, CoreWeave, and Lambda have completed three financings totaling $4.3bn, including two broadly syndicated transactions: CoreWeave's $2.6bn S+550 DDTL 5.5 and Lambda's $926mn S+300 DDTL 2.0. CoreWeave previously completed its first broadly syndicated chip financing in May, a $3.1bn S+450 DDTL 5.0. Visible HPC hardware financing tracked by the report totals $68.6bn, of which $55.2bn was issued in 2026. However, because many transactions originate in the private credit market, the actual total is difficult to observe fully. There is a time lag between data center construction and investment in AI cluster hardware, while high-performance computing hardware accounts for approximately 70% of total AI data center capital expenditure. Capacity currently under construction will therefore gradually translate into chip financing demand over the next several years. This view is also supported by NVIDIA's financing platform plan. On August 10, NVIDIA signed non-binding memoranda of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish separate financing platforms intended to mobilize more than $500bn of third-party capital over the long term for NVIDIA-powered AI infrastructure. For projects assessed individually, NVIDIA may provide residual-value support covering no more than 25% of the relevant opportunity. Barclays believes the arrangement is primarily intended to expand the financing market for NVIDIA GPUs and related hardware. However, the memoranda themselves are non-binding, and the support mechanism is also subject to project-by-project review. The viability of conventional unsecured high-yield financing is considerably weaker for neoclouds. CoreWeave is currently the only neocloud to have entered the conventional high-yield market, with cumulative issuance of $10bn, including $6.25bn in 2026. It also has $6.6bn of convertible bonds outstanding and relies primarily on delayed-draw term loans secured by GPUs. With CoreWeave's unsecured bonds now yielding approximately 11%, Barclays believes it is unlikely to issue further high-yield bonds in the near term. Nebius and IREN have likewise not issued conventional high-yield bonds, instead relying on $8.7bn and $9.5bn, respectively, of convertible bonds for holding-company-level unsecured financing. Nebius also launched an additional $4.5bn convertible bond issuance on August 19. The report believes financing structures are shifting more toward convertible bonds and asset-backed loans rather than costly conventional high-yield debt. The data center securitization market has grown rapidly since the first 144A transaction in 2018, reaching $66bn outstanding as of July 31, 2026. Current ABS and CMBS primarily finance stabilized real estate assets and contractual lease cash flows rather than directly assuming AI computing-capacity or hardware risk. Operators provide tenants with space, power, cooling, and connectivity, while tenants are responsible for IT equipment. Barclays expects the market could reach $180bn by the end of 2028 and believes securitization could eventually extend from real estate and rental cash flows to computing hardware such as GPUs. There is already precedent for ABS financing of enterprise technology equipment, so hardware securitization would represent incremental growth beyond the current forecast. The issuance channel is shifting toward ABS. Total new data center securitization issuance was $27bn in 2025, with net issuance of $20bn, more than double the 2024 level. Barclays expects combined ABS and CMBS issuance of approximately $40bn in 2026, potentially reaching $60bn in 2028. However, YTD 2026 issuance is only $18bn, up 4% year over year, as 24% growth in ABS was partially offset by a 20% decline in CMBS. The report believes recent US Securities and Exchange Commission guidance regarding the issuance of data center securitizations in ABS format could reduce ABS execution costs and encourage issuers to shift further from CMBS toward ABS. Securitization financing capacity remains smaller than that of the corporate bond market. Average transaction size is approximately $850mn for ABS and $1.2bn for CMBS. A $3.5bn CMBS transaction completed in November 2025 was the largest data center securitization ever, and the report views it as an approximate upper bound for a single transaction or series. ABS master trusts can dynamically adjust their collateral pools; the largest currently outstanding trust totals $4.2bn across five series. Large financings are therefore still expected to be funded primarily through the investment-grade or high-yield corporate bond markets. The AI exposure of securitized assets is also not fully transparent because disclosures generally do not identify tenants by name. Based on portfolio classifications, nearly 70% of assets in the current market are leased to hyperscalers, while the remainder is supported mainly by multi-tenant colocation facilities. However, some of that capacity may also be leased to hyperscalers or neoclouds. Since 2018, 23 sponsors have entered the ABS or CMBS market. Blackstone-owned QTS is the most active, accounting for 26% of the current market, including most CMBS structures and two separate ABS master trusts. Despite relatively strong industry fundamentals, Barclays believes large-scale supply will create negative technicals, reflected in wider spreads and greater differentiation among issuers or transactions. Investors may also demand additional structural protections such as scheduled amortization. Data center ABS now accounts for 11% of outstanding esoteric ABS and 5% of all ABS, up from 3% in 2020. Data center CMBS represents 7% of the SASB market and 3% of all non-agency CMBS, versus 0% for both in 2020. The rapid increase in market share means investors must absorb significantly greater concentrated supply, and structures and pricing may need to adjust to support the growth projected in the report.

Analysis framework

The report first establishes a cross-market issuance ledger, dividing AI and data center financing into investment-grade, high-yield, chip loans, ABS, and CMBS, and compares YTD 2026 volumes with 2025, 2024, and the initial June 9 tracking results. It then examines the distribution of supply by issuer, currency, maturity, credit quality, and financing structure, before using the data center project pipeline, the lag between construction and hardware procurement, refinancing needs, and HPC's share of capital expenditure to project future issuance. For securitization, the report further compares ABS and CMBS issuance growth, transaction capacity, collateral, tenant composition, sponsor concentration, and regulatory changes, and finally assesses the effects of heavy supply on spreads and transaction structures.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Analysis of financing supply across asset classes and market absorption capacity

  The report aggregates investment-grade, high-yield, chip financing, and securitization supply and compares future project financing needs with investors' absorption capacity to assess how issuance volumes, spreads, and structural terms may change.

• Industry/Sector Analysis FrameworkUpstream-Midstream-Downstream Transmission

  Lagged transmission from data center construction to GPU and HPC hardware financing

  The report believes data centers are constructed first and then equipped with AI cluster hardware, meaning the current project pipeline will translate into chip financing demand over the next several years. HPC hardware's approximately 70% share of total capital expenditure is a key basis for this projection.

• Industry/Sector Analysis FrameworkVolume-price decomposition

  Decomposition of issuance volume, maturity, currency, and financing price

  The report not only measures total issuance but also breaks it down by issuer, currency, maturity, credit quality, and product format, while using yields and spreads to assess the viability of financing channels and supply pressure.

• Fixed Income and Credit AnalysisSpread analysis

  Impact of supply technicals on bond and securitization spreads

  The report uses issue spreads, current spreads, and yields to assess financing costs and concludes that heavy supply could widen spreads, increase credit differentiation, and prompt investors to demand stronger structural protections.

• Event Strategy and Behavioral FinanceEvent-driven analysis

  Analysis of regulatory guidance and financing platform events

  The report evaluates the effect of US Securities and Exchange Commission guidance on the relative execution costs of ABS and CMBS and analyzes the potential role of NVIDIA's financing platform plan in expanding the GPU financing market.

Asset mapping & comparison

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

• Hyperscaler investment-grade bonds

  They constitute the largest source of AI infrastructure financing supply in 2026 and cover different maturities through the US dollar and foreign-currency markets.

  Strengths

  Issuers can raise funds across currencies and maturities, with YTD supply of approximately $248bn and access to deep market channels.

  Weaknesses

  Supply has reached record levels, and the pace of issuance over the next 12 months could dominate market performance.

  Comparison

  The cross-currency issuance of each of AMZN and GOOGL exceeds the combined issuance of META, ORCL, and SPCX.

  Risks

  Opportunistic pre-funding could further increase supply pressure and affect spreads.

• Investment-grade and high-yield data center bonds

  They provide developers with funding sources beyond traditional construction loans and project finance and support new projects, expansions, and refinancing.

  Strengths

  Issuance increased from $35.9bn in 2025 to $76.2bn YTD 2026, and developers are gradually becoming repeat issuers.

  Weaknesses

  Projects carry construction risk; high-yield transactions typically require partial amortization, while investment-grade transactions emphasize construction-risk mitigation and full amortization.

  Comparison

  YTD 2026 high-yield issuance totals $33.6bn and investment-grade issuance totals $42.6bn, accounting for 15% and 3%, respectively, of supply in their corresponding US dollar markets.

  Risks

  The growing project pipeline and refinancing needs could create concentrated supply and pricing pressure.

• GPU and HPC hardware financing

  It funds the procurement of GPUs and related hardware after data centers are completed and is an important financing channel for AI infrastructure capital expenditure.

  Strengths

  HPC hardware accounts for approximately 70% of AI data center capital expenditure, and the growing number of syndicated transactions provides neoclouds with replicable financing templates.

  Weaknesses

  Many transactions occur in the private credit market, limiting transparency into overall market size and terms.

  Comparison

  Visible financing totals $68.6bn, of which $55.2bn was issued in 2026.

  Risks

  The memoranda of understanding for NVIDIA's platforms are non-binding, and residual-value support must be assessed project by project.

• CoreWeave financing

  CoreWeave is currently the only neocloud to have entered the conventional high-yield market and also uses convertible bonds and GPU-secured loans.

  Strengths

  It has established diversified financing channels comprising conventional high-yield bonds, convertible bonds, and delayed-draw loans.

  Weaknesses

  Unsecured bond yields are already approximately 11%, making further conventional high-yield issuance costly.

  Comparison

  Nebius and IREN have not issued conventional high-yield bonds and primarily use convertible bonds for holding-company-level financing.

  Risks

  Barclays believes CoreWeave is unlikely to issue further conventional high-yield bonds in the near term.

• Data center ABS and CMBS

  They primarily finance stabilized data center real estate and contractual lease cash flows and may expand to hardware such as GPUs in the future.

  Strengths

  Outstanding volume has reached $66bn, nearly 70% of portfolios are leased to hyperscalers, and the market is expected to continue expanding.

  Weaknesses

  Individual transaction capacity and market depth are limited, while insufficient tenant disclosure makes precise AI risk exposure difficult to identify.

  Comparison

  YTD 2026 ABS issuance increased 24%, while CMBS declined 20%; average transaction size is $850mn for ABS and $1.2bn for CMBS.

  Risks

  Heavy supply could widen spreads, increase transaction differentiation, and prompt investors to demand additional protections such as scheduled amortization.

Key data

• Total US dollar financing for AI and data centers$338bnYTD 2026; up $118bn since the initial June 9 tracking report, versus $179bn for full-year 2025.
• Cross-currency issuance by hyperscalersApproximately $248bnYTD 2026, including $182bn in the US dollar market; Barclays forecasts approximately $285bn for the full year.
• Investor expectations for full-year hyperscaler supply$275bn-$325bnUS dollar-equivalent, cross-currency basis.
• Maturity distribution of hyperscaler issuanceFront end 32%, intermediate 32%, long end 36%Corresponding to 1-6 years, 7-15 years, and over 15 years, respectively; 30-year and longer maturities total approximately $88bn.
• Data center bond issuance$76.2bnYTD 2026, including $33.6bn of high-yield and $42.6bn of investment-grade issuance; full-year 2025 totaled $35.9bn.
• Near-term high-yield data center project pipelineApproximately 4.0GWNear-term development pipeline tracked by the report.
• Visible HPC hardware financing$68.6bnCumulative visible issuance, of which $55.2bn was issued in 2026.
• HPC hardware share of capital expenditureApproximately 70%Barclays estimates that it accounts for the majority of total AI data center capital expenditure.
• NVIDIA financing platform targetMore than $500bnThird-party capital intended to be mobilized over the long term; NVIDIA may provide residual-value support for up to 25% of an individual opportunity.
• CoreWeave conventional high-yield bonds$10bnCumulative issuance, including $6.25bn in 2026; unsecured yields are currently approximately 11%.
• Outstanding data center securitizations$66bnAs of July 31, 2026; Barclays expects the market could reach $180bn by the end of 2028.
• 2026 data center securitization issuance forecastApproximately $40bnCombined ABS and CMBS; YTD 2026 issuance totals $18bn, up 4% year over year.
• YTD changes in ABS and CMBSABS up 24%, CMBS down 20%After offsetting each other, total data center securitization issuance increased only 4% year over year.
• Average securitization transaction sizeABS $850mn, CMBS $1.2bnBoth are smaller than investment-grade and high-yield corporate bond financings.
• Hyperscaler tenant shareNearly 70%Current market share based on securitization portfolio classifications.
• QTS share of the securitization market26%Blackstone-owned QTS is the most active data center securitization sponsor.

Impact & implications

The report believes AI infrastructure financing is evolving from large technology-company bonds into a multilayered market encompassing developer bonds, GPU-secured loans, convertible bonds, and securitization. The time lag between construction projects and hardware procurement means financing demand could persist for several years, but the capacity of different channels varies. Large projects will continue to rely more heavily on investment-grade or high-yield corporate bonds, high-cost issuers are shifting toward convertible bonds and asset-backed loans, and stabilized data centers are increasingly using ABS or CMBS. As supply expands, fundamental growth may not translate directly into stronger bond performance; spreads, credit differentiation, and structural protections may become more important pricing factors.

Risks

• Expected heavy supply could widen spreads, increase credit differentiation, and force issuers to provide stronger structural protections.
• CoreWeave's unsecured yields are already approximately 11%, and high financing costs could constrain conventional high-yield issuance by it and other neoclouds in the near term.
• Relevant US Securities and Exchange Commission guidance could further reduce CMBS issuance and accelerate the migration of financing toward the ABS format.
• Securitization disclosures generally do not identify tenants, making it difficult for investors to assess precise AI risk exposure within portfolios.
• The depth and individual transaction capacity of the ABS and CMBS markets are limited, preventing them from independently absorbing large AI infrastructure financings.
• The memoranda of understanding for NVIDIA's financing platforms are non-binding, and residual-value support will be determined project by project.

What to watch

• Monitor the pace of hyperscaler issuance over the next 12 months and whether full-year supply falls within investors' expected range of $275bn-$325bn.
• Track the approximately 4.0GW near-term high-yield project pipeline of data center developers, expansion projects, and incremental supply arising from refinancing existing debt.
• Watch the lagged transmission from data center construction to GPU procurement and whether chip financing continues to expand from private credit into broadly syndicated and investment-grade markets.
• Monitor the actual deployment scale and residual-value support terms of NVIDIA's more-than-$500bn financing platform plan.
• Track changes in the issuance shares of ABS and CMBS following US Securities and Exchange Commission guidance and whether total 2026 securitization issuance can reach approximately $40bn.
• Observe whether investors absorb incremental securitization supply through wider spreads, greater differentiation, and terms such as scheduled amortization.