## The Macro Relevance of the AI Boom ## Source details **Canonical URL:** [The Macro Relevance of the AI Boom](https://www.imf.org/-/media/files/publications/tnm/2026/english/tnmea2026013.pdf) ## Other formats - [Markdown version](/-/media/files/publications/tnm/2026/english/tnmea2026013.pdf.md) - [Structured JSON version](/-/media/files/publications/tnm/2026/english/tnmea2026013.pdf.json) --- ### Macroeconomic and market impacts - Since the public release of ChatGPT in late 2022, AI-related investment has accelerated rapidly across cloud infrastructure, semiconductors, data centers, power systems, and software deployment. - Technology investment as a share of GDP has surpassed levels last observed during the dot-com period. - Hyperscalers alone accounted for roughly $500 billion in capital expenditure (capex) in 2025. - Total AI-related capex is projected to exceed $4 trillion by 2029, with close to 75 percent originating from hyperscalers. - The largest AI firms account for about 35 percent of total market capitalization and are driving equity concentration measures toward historical highs. - Aggregate operating conditions: - Strong earnings growth among technology firms has supported broader improvement in earnings growth beyond the technology sector. - The AI investment boom has bolstered US growth amid tighter monetary policy, tariff uncertainty, and geopolitical shocks. - Global spillovers: - Since late 2022, holdings of US equities as a share of portfolio investment assets have risen markedly across countries and jurisdictions, particularly in the euro area, the United Kingdom, and advanced Asian economies. - East Asian economies have benefited through semiconductor and hardware supply chains; commodity exporters, such as in Latin America, have benefited from rising demand for copper, energy, and power infrastructure. ### Mapping the AI system to balance sheet vulnerabilities - The AI Value Stack is a stylized framework composed of five functional layers capturing 74 firms across the AI supply chain; it is tilted toward incumbents but intended to provide a system-level view. - Table 1 — Functional Layers in the AI Value Stack (as presented): - Broad AI Builders — Construction, engineering, infrastructure, industrial equipment — 103 — (Number of Countries column: 125) - Energizers — Utilities, power generation, grid, cooling, energy infrastructure — 125 — (Number of Companies column: 124) - Core AI — Chip developers — Semiconductors, GPUs, memory, networking equipment — 124 — (Number of Countries column: 93) - Hardware developers — Device manufacturers, lithography machine manufacturers, assemblers and designers of AI server systems — 93 - Hyperscalers — Cloud platforms and large-scale compute providers — 93 - Operator/applications — Software, data managers, AI-enabled applications, services — 226 - Note: Table entries are reported as presented and sourced to IMF staff. - Balance sheet vulnerability assessment uses five fundamental dimensions with quartile-based thresholds (Table 2): - Leverage: Debt to equity (multiples) — Weakest (red) Over 1.5×; Strongest (green) Below 0.5× - Liquidity: Current assets to current liabilities (multiples) — Weakest (red) Below 1×; Strongest (green) Over 1.5× - Profitability: Operating margin (percent) — Weakest (red) Below 10%; Strongest (green) Over 20% - Capex intensity: Capex to sales (multiple) — Weakest (red) Over 35×; Strongest (green) Below 5× - Valuation: Forward P/E (multiples) — Weakest (red) Over 30×; Strongest (green) Below 17.6× - Key findings as of 2026:Q2 and August 2026 (past): - Core AI firms like hyperscalers and chip developers exhibited the strongest fundamentals, with an aggregate operating margin approaching 30 percent. - Downstream and infrastructure-linked layers (operators and data centers) ranked weaker across leverage, profitability, and capital intensity. - The GPU/Neocloud segment showed extreme vulnerability: leverage above 600 percent, still negative profitability, and capex intensity exceeding 300 percent. - Firms involved in “circular financing” tended to combine strong profitability with high capital expenditure. - Profitability had shown the broadest improvement since November 2022. - Chip stack recorded over 300 percent revenue growth, supporting a nearly 700 percent gain in market capitalization. - Among core AI firms, rapid investment growth had absorbed internal cash flow, leading to worsening liquidity and greater debt financing, though leverage remained contained so far. - Infrastructure-related stacks had increasingly relied on external financing; the weakest stack—GPU/Neocloud—saw debt increase more than 10× alongside a 20× increase in capex. - Valuation expansion had been broad-based despite divergence in balance sheet fundamentals. - Private foundational model labs: - The three largest private foundational model labs are collectively valued at over $2 trillion based on their latest funding rounds, despite generating less than $75 billion of combined revenue as of August 2026. - Anticipation of future public listings has attracted retail interest through publicly traded vehicles and other structures that provide indirect pre-IPO exposure to private AI firms. ### Financing the AI boom - Financing shift: - The scale of the AI investment boom has shifted financing from internal cash flows toward external sources. - Historically, large technology firms funded most investment through operating cash flows, supported by high margins, asset-light business models, and strong balance sheets. - While profitability remains exceptionally strong, capital intensity has risen sharply as firms increase spending on data centers, compute infrastructure, power systems, and associated facilities. - Capital allocation trade-offs: - Many of the largest firms continue to return substantial capital to shareholders through dividends and buybacks. - The projected scale of AI-related investment increasingly exceeds what can be comfortably financed internally. - External financing is expected to play a growing role as investment requirements continue to expand. - In the absence of additional financing, firms would increasingly face the trade-off between sustaining investment plans and shareholder distributions. - Systemic implications: - External financing and more interconnected financing structures raise the importance of assessing sustainability of earnings and valuations, balance sheet vulnerabilities, and amplification channels embedded within the broader AI ecosystem. ### Financing sources, bond issuance, and private credit - Traditional corporate bonds remain an important channel, with over $530 billion raised by AI-related firms included in the stack since 2023, dominated by core AI and GPU/Neocloud firms. - Of the $530 billion in bonds issued by the firms in the AI stack since 2023, $405 billion has been issued since January 2025. - Estimates suggest that approximately $2.9 trillion of datacenter capital expenditure through 2028 will be financed through a combination of hyperscaler cash flow, bilateral credit, private capital, public debt, and securitization markets (April 2026 Global Financial Stability Report). - Firms included in the AI Value Stack represent 53 percent of all artificial intelligence–related bond issuances in Canada, the United States, Ireland, The Netherlands, Switzerland, the United Kingdom, Australia, China, Hong Kong SAR, Japan, South Korea, and Taiwan Province of China since 2023, and 65 percent since January 2025. - Private credit has emerged as one of the fastest-growing financing channels; system-wide exposures remain relatively modest, but expansion increasingly links commercial banks, nonbank lenders, and AI-related borrowers. ### Circular financing: mechanics, scale, and quantitative findings - Circular financing links investment, financing, and commercial relationships across the AI ecosystem: large firms finance suppliers, developers, and infrastructure providers while simultaneously acting as customers of those same firms. - Investor equity investments often accompany revenue commitments and other financing commitments (long-term leases, project financing guarantees), allowing modest equity to underpin substantially larger financing structures. - Deal-level and network-level circularity create overlapping financial dependences among counterparties. - IMF staff investigated 14 of the largest circular financing deals in the United States with public disclosures and media reports: - Publicly listed firms involved in these deals are captured by the AI stack and account for close to 50 and 40 percent of revenue and market capitalization, respectively. - When combined with private AI-related firms involved in the deals, their market-implied valuation is about $20 trillion. - Publicly disclosed circular financing arrangements already involve economically significant firms and substantial financial commitments; current estimates likely represent a lower bound on the eventual scale of a circular financing arrangement. - Quantitative findings on circular financing and leverage: - Total direct equity investment across the 14 deals amounts to approximately $300 billion, equivalent to about 1.4 percent of the investors’ aggregate market capitalization. - These transactions are associated with future revenue commitments totaling around $1.4 trillion over an average horizon of five years, implying an annual gross return on equity investment of approximately 65 percent for this period. - Out of 14 deals, 9 were followed by additional debt financing totaling over $650 billion—implying aggregate deal-level leverage of about 130 percent. - Equity financings appear to have been raised at a substantial discount to prevailing market valuations: the implied deal-level price-to-sales multiple averages around 1.5×, less than one-tenth of the multiple implied by the investee’s own public market valuations. - A key source of uncertainty is opacity over ultimate holders of AI-related debt and the role of off-balance-sheet commitments in concealing exposures. ### Obsolescence risk and market fragility - Hyperscalers’ reported depreciation schedules indicate an estimated average useful lifespan of approximately seven years for property, plant, and equipment. - GPUs and advanced chips may become obsolete substantially faster than broader PP&E given rapid technological progress. - Scenario analysis in the April 2026 Global Financial Stability Report suggests that reducing useful life assumptions from seven years to three years lowers aggregate earnings before interest and taxes margins by more than 9 percentage points. - Under a high capital-intensity scenario, operating profitability can be fully absorbed by replacement investment. - Under a stylized debt-financing assumption, aggregate hyperscaler debt rises from approximately $800 billion to more than $1 trillion, while modeled credit spreads widen by roughly 60 basis points, compared with current spreads that range between 20 and 160 basis points. - AI-related stocks have been the dominant driver of US equity market performance: out of the approximately $36 trillion increase in S&P 500 market capitalization since late 2022, about $21 trillion has come from firms in the AI stacks. - Conditional return correlations (DCC-GARCH) for AI-related stock returns have risen markedly and now exceed those observed among the broader technology sector and the “Magnificent 7.” - Counterfactual exercises using M-GARCH imply cumulative returns could decline by approximately 8–10 percent under stressed correlation assumptions relative to baseline conditions. - Cumulative return simulations cover the period September 1, 2025, to March 31, 2026, using return data starting January 1, 2020, and average outcomes across 50,000 model simulations. ### Financial stability outlook and policy considerations - The current AI boom differs from prior technology cycles: it is led by incumbent firms with substantial earnings, cash flows, and financing capacity, allowing investment expansion without a commensurate rise in leverage; valuations are elevated but remain below dotcom extremes; much of the market-cap increase has accompanied strong earnings growth. - Concentration risks are materially larger: AI-related firms account for a historically large share of equity market capitalization and index performance, and the macrofinancial footprint extends well beyond the technology sector. - Downside scenarios include sudden interruptions in investment (recession, funding disruption, geopolitical escalation, tighter regulation) or gradual deterioration in AI economics (weaker monetization than expected, slower productivity gains). - Financing structures could amplify adverse dynamics via obsolescence risks and circular financing arrangements transmitting stress across firms, investors, and funding channels. - The note highlights measures aimed at improving transparency, monitoring interconnected exposures, and identifying leverage-related vulnerabilities as increasingly important as AI-related financing expands. - The note emphasizes the need for forward-looking, adaptable financial stability frameworks, enhanced data collection on AI-related exposures, scenario-based risk assessments, and monitoring of cross-border spillovers. *Source: IMF staff.* ### References 16 ### References 16 ### The Macro Relevance of the AI Boom - Since the public release of ChatGPT in late 2022, AI-related investment has accelerated rapidly across cloud infrastructure, semiconductors, data centers, power systems, and software deployment. - Technology investment as a share of GDP has surpassed levels last observed during the dot-com period. - Hyperscalers alone accounted for roughly $500 billion in capital expenditure (capex) in 2025. - Total AI-related capex is projected to exceed $4 trillion by 2029, with close to 75 percent originating from hyperscalers. - The largest AI firms account for about 35 percent of total market capitalization and are driving equity concentration measures toward historical highs. - Aggregate operating conditions: - Strong earnings growth among technology firms has supported broader improvement in earnings growth beyond the technology sector. - The AI investment boom has bolstered US growth amid tighter monetary policy, tariff uncertainty, and geopolitical shocks. - Global spillovers: - Since late 2022, holdings of US equities as a share of portfolio investment assets have risen markedly across countries and jurisdictions, particularly in the euro area, the United Kingdom, and advanced Asian economies (Figure 1, panel 3). - East Asian economies have benefited through semiconductor and hardware supply chains; commodity exporters, such as in Latin America, have benefited from rising demand for copper, energy, and power infrastructure. ### Mapping the AI System to Balance Sheet Vulnerabilities - The AI Value Stack is a stylized framework composed of five functional layers capturing 74 firms across the AI supply chain; it is tilted toward incumbents but intended to provide a system-level view. - Table 1 — Functional Layers in the AI Value Stack (as presented) - Broad AI Builders — Construction, engineering, infrastructure, industrial equipment — 103 — (Number of Countries column: 125) - Energizers — Utilities, power generation, grid, cooling, energy infrastructure — 125 — (Number of Companies column: 124) - Core AI — Chip developers — Semiconductors, GPUs, memory, networking equipment — 124 — (Number of Countries column: 93) - Hardware developers — Device manufacturers, lithography machine manufacturers, assemblers and designers of AI server systems — 93 - Hyperscalers — Cloud platforms and large-scale compute providers — 93 - Operator/applications — Software, data managers, AI-enabled applications, services — 226 - Note: Table entries are reported as presented and sourced to IMF staff. - Balance sheet vulnerability assessment uses five fundamental dimensions with quartile-based thresholds (Table 2): - Leverage: Debt to equity (multiples) — Weakest (red) Over 1.5×; Strongest (green) Below 0.5× - Liquidity: Current assets to current liabilities (multiples) — Weakest (red) Below 1×; Strongest (green) Over 1.5× - Profitability: Operating margin (percent) — Weakest (red) Below 10%; Strongest (green) Over 20% - Capex intensity: Capex to sales (multiple) — Weakest (red) Over 35×; Strongest (green) Below 5× - Valuation: Forward P/E (multiples) — Weakest (red) Over 30×; Strongest (green) Below 17.6× - Key findings as of 2026:Q2 and August 2026: - Core AI firms like hyperscalers and chip developers exhibit the strongest fundamentals, with an aggregate operating margin approaching 30 percent. - Downstream and infrastructure-linked layers (operators and data centers) rank weaker across leverage, profitability, and capital intensity. - The GPU/Neocloud segment shows extreme vulnerability: leverage above 600 percent, still negative profitability, and capex intensity exceeding 300 percent. - Firms involved in “circular financing” tend to combine strong profitability with high capital expenditure. - Profitability has shown the broadest improvement since November 2022. - Chip stack recorded over 300 percent revenue growth, supporting a nearly 700 percent gain in market capitalization. - Among core AI firms, rapid investment growth has absorbed internal cash flow, leading to worsening liquidity and greater debt financing, though leverage has remained contained so far. - Infrastructure-related stacks have increasingly relied on external financing; the weakest stack—GPU/Neocloud—saw debt increase more than 10× alongside a 20× increase in capex. - Valuation expansion has been broad-based despite divergence in balance sheet fundamentals. - Private foundational model labs: - The three largest private foundational model labs are collectively valued at over $2 trillion based on their latest funding rounds, despite generating less than $75 billion of combined revenue as of August 2026. - Anticipation of future public listings has attracted retail interest through publicly traded vehicles and other structures that provide indirect pre-IPO exposure to private AI firms. ### Financing the AI Boom - Financing shift: - The scale of the AI investment boom has shifted financing from internal cash flows toward external sources. - Historically, large technology firms funded most investment through operating cash flows, supported by high margins, asset-light business models, and strong balance sheets. - While profitability remains exceptionally strong, capital intensity has risen sharply as firms increase spending on data centers, compute infrastructure, power systems, and associated facilities. - Capital allocation trade-offs: - Many of the largest firms continue to return substantial capital to shareholders through dividends and buybacks. - The projected scale of AI-related investment increasingly exceeds what can be comfortably financed internally. - External financing is expected to play a growing role as investment requirements continue to expand. - In the absence of additional financing, firms would increasingly face the trade-off between sustaining investment plans and shareholder distributions. - Systemic implications: - External financing and more interconnected financing structures raise the importance of assessing sustainability of earnings and valuations, balance sheet vulnerabilities, and amplification channels embedded within the broader AI ecosystem. *Source: IMF staff.* ### 1. S&P 500 Operating Income and Capex2. Bond Issuance by AI Value Stack ### 1. S&P 500 Operating Income and Capex2. Bond Issuance by AI Value Stack ### Financing sources and bond issuance - Traditional corporate bonds remain an important channel, with over $530 billion raised by AI-related firms included in the stack since 2023, dominated by core AI and GPU/Neocloud firms. - Of the $530 billion in bonds issued by the firms in the AI stack since 2023, $405 billion has been issued since January 2025. - Estimates suggest that approximately $2.9 trillion of datacenter capital expenditure through 2028 will be financed through a combination of hyperscaler cash flow, bilateral credit, private capital, public debt, and securitization markets (April 2026 Global Financial Stability Report). - Firms included in the AI Value Stack represent 53 percent of all artificial intelligence–related bond issuances in Canada, the United States, Ireland, The Netherlands, Switzerland, the United Kingdom, Australia, China, Hong Kong SAR, Japan, South Korea, and Taiwan Province of China since 2023, and 65 percent since January 2025. - Private credit has emerged as one of the fastest-growing financing channels; system-wide exposures remain relatively modest (Aldasoro and others 2026), but expansion increasingly links commercial banks, nonbank lenders, and AI-related borrowers. ### Circular financing structures: mechanics and risks - Circular financing links investment, financing, and commercial relationships across the AI ecosystem: large firms finance suppliers, developers, and infrastructure providers while simultaneously acting as customers of those same firms. - Investor equity investments often accompany revenue commitments and other financing commitments (long-term leases, project financing guarantees), allowing modest equity to underpin substantially larger financing structures. - These arrangements create deal-level circularity (multiple channels of financial dependence among counterparties) and network-level circularity (replication across firms, creating overlapping relationships among investors, investees, and other counterparties). - IMF staff investigated 14 of the largest circular financing deals in the United States with public disclosures and media reports; publicly listed firms involved in these deals are captured by the AI stack and account for close to 50 and 40 percent of revenue and market capitalization, respectively. When combined with private AI-related firms involved in the deals, their market-implied valuation is about $20 trillion. - Publicly disclosed circular financing arrangements already involve economically significant firms and substantial financial commitments; current estimates likely represent a lower bound on the eventual scale of a circular financing arrangement. ### Quantitative findings on circular financing and leverage - Total direct equity investment across the 14 deals amounts to approximately $300 billion, equivalent to about 1.4 percent of the investors’ aggregate market capitalization. - These transactions are associated with future revenue commitments totaling around $1.4 trillion over an average horizon of five years, implying an annual gross return on equity investment of approximately 65 percent for this period. - Out of 14 deals, 9 were followed by additional debt financing totaling over $650 billion—implying aggregate deal-level leverage of about 130 percent. - Equity financings appear to have been raised at a substantial discount to prevailing market valuations: the implied deal-level price-to-sales multiple averages around 1.5×, less than one-tenth of the multiple implied by the investee’s own public market valuations. - A key source of uncertainty is opacity over ultimate holders of AI-related debt and the role of off-balance-sheet commitments in concealing exposures. ### Obsolescence risk and market fragility - Hyperscalers’ reported depreciation schedules indicate an estimated average useful lifespan of approximately seven years for property, plant, and equipment. - GPUs and advanced chips may become obsolete substantially faster than broader PP&E given rapid technological progress. - Scenario analysis in the April 2026 Global Financial Stability Report suggests that reducing useful life assumptions from seven years to three years lowers aggregate earnings before interest and taxes margins by more than 9 percentage points. - Under a high capital-intensity scenario, operating profitability can be fully absorbed by replacement investment. - Under a stylized debt-financing assumption, aggregate hyperscaler debt rises from approximately $800 billion to more than $1 trillion, while modeled credit spreads widen by roughly 60 basis points, compared with current spreads that range between 20 and 160 basis points. - AI-related stocks have been the dominant driver of US equity market performance: out of the approximately $36 trillion increase in S&P 500 market capitalization since late 2022, about $21 trillion has come from firms in the AI stacks. - Conditional return correlations (DCC-GARCH) for AI-related stock returns have risen markedly and now exceed those observed among the broader technology sector and the “Magnificent 7.” - Counterfactual exercises using M-GARCH imply cumulative returns could decline by approximately 8–10 percent under stressed correlation assumptions relative to baseline conditions. - Cumulative return simulations cover the period September 1, 2025, to March 31, 2026, using return data starting January 1, 2020, and average outcomes across 50,000 model simulations. ### Financial stability outlook and policy considerations - The current AI boom differs from prior technology cycles: it is led by incumbent firms with substantial earnings, cash flows, and financing capacity, allowing investment expansion without a commensurate rise in leverage; valuations are elevated but remain below dotcom extremes; much of the market-cap increase has accompanied strong earnings growth. - Concentration risks are materially larger: AI-related firms account for a historically large share of equity market capitalization and index performance, and the macrofinancial footprint extends well beyond the technology sector. - Downside scenarios include sudden interruptions in investment (recession, funding disruption, geopolitical escalation, tighter regulation) or gradual deterioration in AI economics (weaker monetization than expected, slower productivity gains). - Financing structures could amplify adverse dynamics via obsolescence risks and circular financing arrangements transmitting stress across firms, investors, and funding channels. - While detailed policy recommendations are outside the scope of the note, measures aimed at improving transparency, monitoring interconnected exposures, and identifying leverage-related vulnerabilities are highlighted as increasingly important as AI-related financing expands. - The note emphasizes the need for forward-looking, adaptable financial stability frameworks, enhanced data collection on AI-related exposures, scenario-based risk assessments, and monitoring of cross-border spillovers (Barhoumi and others 2026). --- _Source: https://www.imf.org/-/media/files/publications/tnm/2026/english/tnmea2026013.pdf_