AI Financing: Global Capital Adequacy, Structuring is Key

Overview

This report analyzes the financing needs and global capital supply capacity during the AI infrastructure construction period, reaching the key conclusion: capital is not a constraint. The report points out that the AI construction cycle is driving an unprecedented wave of infrastructure investment, with cumulative early four-year investments exceeding $1 trillion, which is 10x the early cloud computing cycle. Meanwhile, the global capital pool (including retail and institutional investors, insurance companies, pension funds, sovereign wealth funds, alternative asset managers, and banks) scales to $256 trillion, enough to support $10 trillion in AI investment demands. The report emphasizes that the real challenge is not whether capital is adequate, but how financial markets evolve to intermediated this capital through diversified financing tools (public equity, investment-grade credit, securitization products, private credit, infrastructure capital, etc.). At the same time, power supply and computing capability are the real bottlenecks restricting AI expansion.

Core views

The core viewpoints of the report revolve around three aspects: **Capital Adequacy**: The global $256 trillion capital pool relative to the $10 trillion AI investment demand accounts for only 4% in actual proportion. From the perspective of the global stock market's $159 trillion market value, the US accounts for $76 trillion, providing a deep-level capital foundation. The report believes this investment does not need to trigger large-scale liquidation in capital markets; funds are more likely to be obtained through gradual portfolio adjustments, new issuances, maturity-based bond reinvestments, and private capital deployment methods. In particular, large institutional investors continuously receive principal repayments and interest income from their bond portfolios, forming cyclic reinvestment capacity. Additionally, annuity demands driven by aging populations will continue to support insurance companies' demand for long-duration, yield-oriented credits and structured products. **Market Issuance Capacity is Huge**: Public bonds and loans markets set records in 2025, with global corporate bonds/syndicated loan issuance reaching $13.3 trillion, accounting for approx 43% of US nominal GDP, 7 percentage points higher than the long-term historical average of 40%. If global issuance returns to peak levels, annual financing capacity could reach over $17 trillion, indicating ample space remains for AI financing. The public equity market is similarly underutilized: 2025 global IPO issuance was approx $20 billion, 50% lower than the historical average; global equity capital market (ECM) issuance was approx $1 trillion, 33% lower than the historical average. If restored to historical averages or peak levels, global IPO financing capacity could reach $41-85 billion, global ECM financing capacity could reach $1.6-2.6 trillion, both far exceeding current levels. **Diversified Financing Channels are Forming**: Data center financing cases show that among the $3.2 trillion global data center capex forecast, $1.75 trillion is financed through diverse credit products, including $70 billion private credit and $65 billion investment-grade bonds. Securitization credit (data center ABS/CMBS) currently stands at approx $6.5 billion, expected to increase to approx $20 billion by 2028, nearing a 3x growth. Features of different capital pools determine their roles in AI infrastructure financing: Insurance companies suit long-duration, yield-oriented products; Pension funds and sovereign wealth funds can participate broadly across public equity, private equity, infrastructure, etc.; Alternative asset managers participate directly through data center platforms, power investments, etc.; Banks provide liquidity and structured support through balance sheet capacity and capital market capabilities. **Real Bottlenecks are Power and Computing Capability**: The report clearly states that AI deployment is truly constrained not by capital, but by power, chips, infrastructure, and enterprise readiness. US data center demand is expected to significantly exceed available grid capacity at least until 2028, even considering new energy solutions. Grid access delays, permitting obstacles, labor shortages, key equipment supply chain bottlenecks such as turbines and nuclear reactors all delay deployment. On the chip side, hyperscalers report high-end chip utilization close to full load. As AI shifts from training to inference and agent workloads, CPU, memory, system coordination, and other new constraints are emerging. In addition, enterprise data readiness, talent shortages, security privacy, and legacy system integration remain constraining factors.

Analysis framework

The report adopts an analysis method combining macro capital pool stocks and flows. First, starting from the global capital pool size (retail and institutional assets, insurance assets, pension funds, sovereign wealth funds, etc.), compared with AI investment demands, drawing the conclusion of capital adequacy. Second, by analyzing historical issuance data (long-term average and peak percentages of GDP), deducing market issuance capacity, showing that different financing channels are currently still underutilized. Third, analyzing the matching degree of different capital pools to different levels of the AI financing stack based on investor characteristics and risk-return-maturity preferences, explaining the importance of financial market structuring. Finally, analyzing the hard constraints on AI expansion from the supply side (power, chips, infrastructure), revealing that capital is not the tightest element. This method looks at totals and structures, demand side and supply side, forming a three-dimensional financing analysis framework.

Methodology notes

• Company Fundamentals and Financial FrameworkFree cash flow analysis

  The report evaluates the financing bearing capacity of hyperscalers through free cash flow capability. Currently, internal cash flow of these vendors still supports most early AI capex, but as investment scales expand, external financing demand will inevitably increase.

  Free cash flow represents cash freely disposable after enterprises maintain operations and investment. Although hyperscalers have strong FCF, the growth rate of AI capex has begun to exceed FCF growth rates, determining the critical point when they need to shift to public market financing.

• Industry/Industrial Analysis FrameworkSupply and Demand Framework

  The report clearly distinguishes supply and demand characteristics of AI infrastructure. Demand side looks at model capability improvement, inference demand, enterprise adoption continuing to grow; Supply side looks at computing chips, power, heat dissipation, network infrastructure supply bottlenecks.

  Supply and demand frameworks help identify which elements truly constrain industry expansion. In capital markets, capital liquidity is good, easily mobilized quickly; while physical elements like power and chips are difficult to scale production quickly, forming structural bottlenecks, thus becoming real constraints.

• Industry/Industrial Analysis FrameworkUpstream/Midstream/Downstream Industry Chain Transmission

  The report tracks AI investment transmission along the industry chain: Hyperscaler capex -> Upstream suppliers such as chips, networks, power -> Midstream infrastructure such as data centers -> Downstream applications by enterprise clients.

  Industry chain transmission analysis reveals that capital demand is not only concentrated in hyperscalers but also spreads across the entire ecosystem. Data center financing includes 70 billion private credit and 65 billion investment-grade bonds within the 1.75 trillion, indicating financing demand has spread to midstream infrastructure.

• Macroeconomic frameworkCredit/debt cycle

  The report observes public debt market hitting record highs in 2025 ($13.3 trillion) and compares with historical long-term average, inferring the market is still in a loose cycle, financing cost and issuance difficulty do not constitute constraints.

  Credit cycle determines bond market bearing capacity. When global corporate bonds/loans issuance as % of GDP reaches 43% (higher than 40% long-term mean) without financing difficulties, it indicates the market is in a friendly environment with sufficient capacity to absorb AI-related financing.

• Quantitative/Factor/Portfolio Theory

  The report conducts quantitative statistics on global asset ownership, quantifying specific sizes such as insurance assets $42 trillion, pension funds $42 trillion, sovereign wealth funds $16 trillion, retail assets $153 trillion, etc.

  By quantifying specific sizes and structures of the global capital pool, the report can precisely benchmark against $10 trillion AI investment demand (only 4% of total capital pool), using data-driven rather than qualitative arguments to verify capital adequacy.

• Company Fundamentals and Financial FrameworkWorking capital cycle

  The report emphasizes large institutional investors continuously receive principal and interest cash flows from bond portfolios, forming cyclic reinvestment capacity, able to constantly direct this liquidity towards new AI financing products.

  The concept of working capital cycle is applied here to fund flows of institutional investors. Bond investment portfolios mature, expire, collect payments forming natural reinvestment mechanisms, becoming organic funding sources for AI financing, rather than relying solely on new capital input.

• Valuation Method

  The report calculates future issuance capacity through historical peak issuance volumes. For example, global ECM historical peak relative to GDP level can derive $2.6 trillion annual financing potential.

  This is a normalized valuation method: not just looking at current absolute values, but looking at historical ranges relative to GDP (average, peak) to derive reasonable space. This method applies to financing market analysis, avoiding simple extrapolation.

• Industry/Industrial Analysis FrameworkCost curve analysis

  The report compares costs and applicable objects of different financing tools: Public equity (most liquid, highest risk), Investment-grade credit (medium risk, long duration), Private credit (higher cost, flexible), Infrastructure capital (longest term, lowest liquidity).

  Cost curve analysis helps understand why diverse financing tools coexist: Different investor cost preferences and risk tolerances differ, allowing through gradient financing stacks every type of capital to find most suitable usage, thereby maximizing overall social financing efficiency.

• Event Betting and Behavioral FinanceExpectation Gap / Expectation Management

  The report believes market concerns/expectations regarding AI financing constraints have been overly exaggerated. Global capital is adequate, market capacity huge, but this cognitive gap has not yet been fully priced into public markets.

  When the market believes a variable (such as financing difficulty) will become a bottleneck, but actual constraints are much smaller than expectations, an expectation gap arises. The report eliminates this gap through quantitative data, suggesting investors adjust pricing for AI financing risks.

• Industry/Industrial Analysis Framework

  The report proposes 'Capital Intermediation' as a core concept: The degree of evolution of global financial markets determines whether $256 trillion capital pool can be effectively allocated to $10 trillion AI infrastructure demand.

  Efficiency of capital intermediation depends on richness of financial products, transaction costs, information symmetry, perfection of risk management tools, etc. The report emphasizes that through innovations such as securitization and structured financing, intermediary efficiency can be significantly improved, thereby eliminating financing constraints.

Asset mapping & comparison

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

• Alternative Asset Managers (KKR, BX, APO, BN, BAM, ARES, BLK)

  Directly benefit from private market expansion and increased AI infrastructure financing demand. These institutions possess data center platforms, power investment capabilities, private credit, and infrastructure expertise, core intermediaries in the AI financing ecosystem

  Strengths

  Large scale, ample dry powder, strong platform advantages, multi-asset capabilities, global network breadth. Especially KKR, BX, BLK, etc. have established direct data center, power, and infrastructure platforms through acquisitions and strategic cooperation

  Weaknesses

  Public market liquidity relatively lower than pure stocks, attractiveness depends on return expectations and exit availability

  Comparison

  Compared with traditional banks and insurance companies, alternative managers have stronger flexibility, risk tolerance, and long-term capital investment capacity; Compared with pure technology companies, alternative managers' financing intermediary role gives them cross-cycle stability

  Risks

  Distribution and scaling risks of private funds, exit pressure, impact of macroeconomic downturn on infrastructure returns, geopolitical and regulatory changes

• Large Commercial Banks (BAC, C, GS, JPM, WFC)

  Benefit from AI financing wave through multiple dimensions such as investment banking, trading, lending, capital markets underwriting. Banks are primary intermediaries of AI financing, playing roles of M&A advisor, bond and loan underwriter, counterparty, etc.

  Strengths

  Report emphasizes these banks are well-capitalized ($21.0 billion excess capital, can release $2.5 trillion balance sheet capacity), possessing global networks, strong IB capabilities, strong trading innovation capabilities. Especially Citi has established specialized AI infrastructure banking teams

  Weaknesses

  Regulatory capital requirements high, high dependence on diversification of revenue sources, intermediary businesses easily affected by interest rates and market volatility

  Comparison

  Compared with alternative managers, banks have stronger liquidity and public market financing ability; Compared with pure technology companies, banks have professionalism in risk intermediation and capital allocation

  Risks

  Pressure on lending and fixed-income businesses from rising interest rates, regulatory policy changes, increasing credit risks, intensifying peer competition

• Insurance Companies

  As important source of long-term capital, participate in financing by purchasing AI infrastructure related IG credit, ABS/CMBS and structured products. Pension demand driven by population aging increases insurance companies' demand for long-term investment

  Strengths

  Hold massive long-term capital ($42 trillion), naturally suitable for long-term infrastructure investment, focus on return stability and credit quality makes them stable funding sources

  Weaknesses

  Investment scope strictly regulated, yield seeking may be compressed in low interest rate environments, high sensitivity to credit ratings and risks

  Comparison

  Compared with public market investors, insurers have longer time horizon and stronger risk tolerance; Compared with private funds, insurers have lower liquidity needs but more regulations

  Risks

  Impact of interest rate fluctuations on balance sheets, insurance business cycle risks, increased regulatory capital requirements

Key data

• Early AI Four-Year Investment ScaleExceeds $1 trillionApproximately 10x early cloud computing cycle, indicating scale and speed of AI infrastructure construction are unprecedented
• Hyperscaler Annual Capex Forecast (2027)Exceeds $1 trillionGrowth of 17% and 5% respectively vs 2025 and 2026, planned capex already exceeds market expectations
• Global Capital Pool SizeApproximately $256 trillionIncludes retail assets, institutional investors, insurance companies, pension funds, sovereign wealth funds, endowments, etc., all capital sources
• Global Stock Market Total Market ValueApproximately $159 trillionWhere US accounts for $76 trillion, providing deep-level equity financing foundation
• 2025 Global Corporate Bonds/Syndicated Loan Issuance Scale$13.3 trillion (Record High)Accounts for 43% of US nominal GDP, 7 percentage points higher than historical long-term average of 40%
• Assumed Scenario Global Annual Debt Financing CapacityExceeds $17 trillionIf global issuance recovers to historical peak % of GDP, capacity reaches this level
• 2025 Global IPO Issuance ScaleApproximately $20 billionApprox 50% lower than historical average, indicating market underutilization
• Global IPO Financing Capacity Restored to Historical Average/Peak Level$41 billion / $85 billionBoth far exceed 2025 $20 billion issuance scale, indicating huge future financing space
• 2025 Global Equity Financing (ECM) ScaleApproximately $1 trillionApprox 33% lower than historical average, underutilizing market capacity
• Global ECM Financing Capacity Restored to Historical Average/Peak Level$1.6 trillion / $2.6 trillionBoth significantly higher than current issuance levels, indicating issuance market has sufficient absorption space
• Global Data Center Capex Forecast (2026-2028)$3.2 trillionWhere $1.75 trillion financed through diverse credit products, including $70 billion private credit and $65 billion investment-grade bonds
• Data Center Securitization Credit ScaleCurrent approx $6.5 billion, expected ~$20 billion by 2028Growth approx 3x, reflecting increased importance of securitization financing in AI infrastructure financing
• Private Market Dry Powder ScaleApproximately $4.9 trillion (March 2026)Where infrastructure dry powder $36 billion, account 7%
• Securities Market Fund Flows (2026 YTD)Money market funds net outflow $15 billion, Long-term funds net inflow $48.9 billionIndicates investor risk appetite rising, capital transferring from cash to high-risk assets
• US Large Bank Excess CapitalCurrent $15.4 billion, $21.0 billion under new rulesCan release approx $2.5 trillion balance sheet financing capacity
• US IPO Issuance Scale (2025)Approximately $8 billion50% lower than historical average ($16.5 billion), lower than peak ($46 billion)
• US ECM Issuance Scale (2025)Approximately $50 billion25% lower than historical average ($70 billion), significantly lower than peak ($1.1 trillion)
• Insurance Asset ScaleApproximately $42 trillionNaturally suitable for long-duration, yield-oriented AI infrastructure financing products
• Pension Fund Asset ScaleApproximately $42 trillionCan flexibly allocate across multiple asset classes, including public equity, private equity, infrastructure, etc.
• Sovereign Wealth Fund Asset ScaleApproximately $16 trillionHas long investment terms, suitable for participating in large-scale AI infrastructure projects
• US Money Market Fund Asset ScaleApproximately $7.6 trillionGrowth 10% vs last year, can serve as liquidity buffer for AI financing

Impact & implications

The report believes concerns over AI financing constraints are excessively exaggerated. Global capital is adequate, financing market capacity huge, diversified financing channels forming, meaning capital will not become a factor limiting AI infrastructure deployment. For investors, core implications are: (1) Should not bear short views on AI infrastructure themes due to financing worries; (2) Should focus on intermediaries capable of providing financing solutions, including alternative asset managers, investment banks, private credit managers, etc., whose role in financial structuring is crucial; (3) Should prepare for real bottlenecks — power, chips, infrastructure — which are the key determinants of AI expansion speed. For policy and markets, smooth AI financing relies on financial innovation and depth of market perfection. Securitization products, private credit, structured financing, infrastructure investment tools, and other diverse tool development will be key. At the same time, non-financial factors such as power supply, energy policies, supply chain management must advance simultaneously, otherwise capital adequacy cannot be fully released. For the AI industry ecosystem, financing is no longer the main constraint, this provides clear prospects for the entire industry: Capital will continue to flow, real competition will concentrate on how to efficiently acquire and utilize scarce computing and energy resources.

Risks

• Power and Grid Constraints: US data center demand expected to significantly exceed available grid capacity at least until 2028; grid access delays, permitting obstacles, key equipment supply chain bottlenecks may all delay deployment
• Chip Supply Constraints: Hyperscalers have reported high-end chip utilization approaching full load; with inference and agent workload increases, CPU, memory, and other new constraints are emerging
• Rising Financing Costs: If global interest rates rise, bond financing costs will increase, potentially affecting project economics and financing attractiveness
• Enterprise AI Monetization Below Expectations: If enterprise-side AI monetization and adoption speed falls below expectations, may lead to infrastructure investment ultimately exceeding sustainable demand
• More Efficient Model Architectures: If more energy-efficient model architectures or inference methods found, may reduce computing intensity demand, thus reducing future capex
• Regulation and Geopolitical Risks: Policy changes regarding data privacy, national security, chip restrictions may affect cross-border investment and technology flow
• Infrastructure Cycle Fluctuations: If AI investment ultimately exceeds sustainable demand growth, may lead to decreased asset utilization rates and earnings pressure, especially under long-term locked contract terms

What to watch

• Actual execution and revision of hyperscaler capex plans, especially whether 2026-2027 capex maintains high-speed growth
• IPO and ECM market recovery progress: Focus on whether public equity financing truly returns to historical averages, and proportion of AI-related financing within them
• Actual progress of power and chip supply: Monitor speed of US and global grid capacity expansion, new energy solution deployment, high-end chip capacity release
• Actual scale and growth of securitization products and private credit in data center financing, whether actually reaches $20 billion scale by 2028
• Actual capital deployment amounts and investment terms from insurance companies and alternative managers for AI infrastructure financing
• Enterprise AI adoption and monetization progress, especially commercialization speed of inference workloads and agent applications
• Financing cost changes and market pricing reactions, including data center loan spreads, credit rating changes, financing difficulty indicators