Two of Wall Street’s most influential banks have issued stark warnings about the artificial intelligence investment boom, drawing uncomfortable parallels to the dot-com crash that devastated markets 25 years ago. In a November 9, 2025 note to clients, Goldman Sachs strategists Dominic Wilson and Vickie Chang warned that the AI trade now resembles the tech sector in 1997 roughly three years before the bubble burst. Meanwhile, Bank of America reported that tech companies issued a staggering $75 billion in AI-focused debt during just September and October 2025, more than double the annual average from 2015 to 2024. These warnings come as Amazon, Microsoft, Alphabet, Apple, and Meta collectively race toward $349 billion in capital expenditures for 2025, with projections climbing to $571 billion by 2026.
The financial firepower fueling artificial intelligence infrastructure has reached unprecedented levels, but the question keeping institutional investors awake isn’t whether AI will transform the economy, it’s whether the debt-fueled frenzy will collapse before returns materialize. Goldman Sachs emphasized that while stocks don’t yet appear to be in their “1999 moment,” five critical warning signs from the late 1990s bubble are beginning to flash amber. This article examines those warning signals, breaks down the debt data behind the AI boom, and provides investors with actionable frameworks for monitoring risk in their portfolios.Table of Contents
Understanding the AI Debt Explosion
What’s Driving Tech Giants to Debt Markets
For years, technology giants funded their infrastructure investments from robust operating cash flows generated by cloud services, digital advertising, and software subscriptions. Meta produces over $20 billion in quarterly operating cash flow, while Microsoft and Alphabet generate similar amounts. Yet the scale of AI data center buildouts has overwhelmed even these massive cash engines, forcing companies to tap external debt markets in record volumes.
Paul Kedrosky, a venture capitalist and MIT research fellow, notes that Big Tech firms are increasingly taking on off-balance-sheet debt loans that don’t appear in standard financial statements to finance AI ambitions. Ruth Yang, global head of private markets analytics at S&P Global Ratings, confirms that “sometimes they put it on their balance sheets, and sometimes they don’t,” creating opacity around actual debt loads. This structural shift represents a fundamental change in how the industry finances growth, moving from predictable internal funding to complex financial instruments that introduce leverage and refinancing risks.
UBS analysts project that global AI capital expenditure will reach $423 billion in 2025 and surge to $571 billion by 2026, implying a 25% compound annual growth rate through 2030. The bank lifted its 2030 forecast to $1.3 trillion, driven by “robust fundamentals” and accelerating demand for computing power to support increasingly complex AI applications. This spending trajectory necessitates extensive debt financing across every segment of capital markets.
The Numbers Behind the Boom
Bank of America’s research reveals that $75 billion in investment-grade debt from AI-focused companies hit markets in just September and October 2025, a two-month period that exceeded the typical annual issuance from 2015 to 2024. This acceleration signals a phase shift in financing patterns, with companies moving beyond measured, opportunistic borrowing to aggressive, necessity-driven debt raises.
Meta’s November 2025 bond offering stands as the year’s largest corporate debt sale, totaling $30 billion with $27 billion in debt and $3 billion in equity components. The offering drew $125 billion in orders more than four times oversubscribed demonstrating robust investor appetite despite mounting bubble concerns. The proceeds fund a massive Louisiana data center through a special purpose vehicle structure, keeping some debt off Meta’s primary balance sheet.
Oracle secured $18 billion in bond issuance and an additional $38 billion loan in late 2025 to finance data center construction for the Stargate AI infrastructure initiative in New Mexico. The company’s total debt now approaches $96 billion, with quarterly net income around $3 billion. If interest rates rise or AI revenue disappoints, Oracle’s interest payments could consume a disproportionate share of profits, constraining financial flexibility.
JPMorgan Chase strategists led by Tarek Hamid project that AI hyperscalers will need approximately $1.5 trillion in investment-grade bonds over the next five years. Even with leveraged finance contributing around $150 billion and data center securitizations providing up to $40 billion annually, a $1.4 trillion funding gap remains. Private credit markets and government support will likely fill this void, but at what cost to corporate balance sheets and financial system stability ?
Goldman Sachs’ Five Dot-Com Warning Signals
Warning Signal #1: Peak Investment Spending
Goldman Sachs strategists identified peaks in technology investment spending as the first critical warning sign that preceded the dot-com collapse. In the late 1990s, capital expenditure by technology and telecommunications companies surged to unsustainable levels as firms raced to build internet infrastructure, often with little regard for return on investment. The 2025 AI boom exhibits similar dynamics: Amazon, Microsoft, Alphabet, Apple, and Meta are projected to spend $349 billion on capital expenditures in 2025, with the broader industry reaching $423 billion.
UBS data shows that capex intensity among major U.S. tech companies has nearly doubled to 20.8% over the past five years and is forecast to reach 27% by 2030. This represents capital spending as a percentage of revenue, a metric that climbed to dangerous levels before the 2000 crash when companies prioritized market share over profitability. Jensen Huang, Nvidia’s CEO, revealed in November 2025 that his company holds $500 billion in chip orders from AI infrastructure buyers, confirming the extraordinary scale of committed spending.
Jacob Sonnenberg, portfolio manager at Irving Investors, notes that “people expected big numbers and they got big numbers,” referring to third-quarter 2025 capex announcements. The predictability of ever-increasing investment suggests companies are locked into an arms race where pulling back risks ceding competitive position, even if financial prudence dictates restraint.
Warning Signal #2: Declining Corporate Profitability
The second warning signal involves deteriorating corporate profitability despite rising investment. Goldman Sachs strategists emphasized that “the combination of rising investment and falling profitability pushed the corporate sector’s financial balance and the difference between savings and investment into deficit” during the late 1990s. When companies invest more than they save, they must fund the gap through debt or equity issuance, creating financial vulnerability.
While current AI leaders like Meta, Microsoft, and Alphabet maintain strong profitability and massive cash flows, the trajectory shows concerning patterns. Meta increased its 2025 capital expenditure guidance to $70-72 billion from a previous range of $66-72 billion, with CFO Susan Li signaling continued aggressive spending into 2026. Despite generating over $20 billion in quarterly operating cash flow, Meta’s free cash flow which remains after capital expenditures has compressed significantly.
Smaller AI-focused players face more acute pressure. Advanced Micro Devices (AMD), while currently carrying under $3 billion in debt, may need to borrow billions more for capex, potentially at higher rates due to lower credit ratings. For these companies, interest costs could erode already-slim margins, triggering the profitability squeeze that Goldman warns preceded the dot-com collapse.
Warning Signal #3: Rapid Corporate Debt Growth
The third warning signal focuses on accelerating corporate debt accumulation, particularly when debt grows faster than profits. Bank of America’s research documents this phenomenon clearly: the $75 billion in AI-related debt issued during September and October 2025 alone represents explosive growth compared to historical averages. Meta’s $37 billion in total debt, Oracle’s $96 billion, and the sector’s aggregate borrowing trajectory all point to leverage building across the industry.
Paul Kedrosky highlights a particularly concerning development: the use of special purpose vehicles and off-balance-sheet debt structures that obscure true leverage levels. “Since much of this debt sits in special purpose vehicles, it can be hard to know who ultimately holds the risk: Big Tech, the lenders, or the investors in funds that own the loans,” he explains. This opacity echoes the structured finance complexity that amplified the 2008 financial crisis.
Goldman Sachs notes that while current debt-to-profit ratios appear “significantly lower than at the peak of the internet bubble,” the velocity of debt accumulation matters as much as absolute levels. Rapid increases signal that companies are stretching to maintain growth momentum, potentially overextending financial capacity if AI monetization lags expectations.
Private credit has emerged as a major funding source, running at approximately $50 billion per quarter for AI infrastructure. Meta reportedly explored a $29 billion private credit deal with up to $26 billion in debt to fund data center buildouts. These private market transactions add another layer of risk concentration, as losses could cascade through interconnected lender networks outside regulatory oversight.
Warning Signal #4: Federal Reserve Rate Cuts
The fourth warning signal identified by Goldman Sachs involves Federal Reserve rate cuts that provide monetary fuel for asset bubbles. In the late 1990s, the Fed cut rates during a mid-cycle adjustment, and “lower rates and capital inflows added fuel to the equity market,” the strategists wrote. By reducing borrowing costs and encouraging risk-taking, rate cuts can inflate valuations beyond fundamental support.
The Federal Reserve cut interest rates by 25 basis points at its October 2025 policy meeting, with investors expecting another 25 basis-point reduction in December 2025 according to CME FedWatch Tool data. This easing cycle arrives as AI enthusiasm reaches fever pitch, creating conditions where cheap capital chases speculative opportunities. Ray Dalio, founder of Bridgewater Associates, has specifically warned that the Fed’s easing cycle could help inflate a bubble in markets.
The timing raises particular concern because rate cuts typically occur either during economic weakness or when the Fed believes inflation is under control. If the current cutting cycle stems from confidence about inflation rather than economic necessity, it could encourage excessive leverage and risk-taking among AI investors and companies. Lower borrowing costs make debt-financed investments appear more attractive, potentially driving capital allocation toward projects with marginal returns.
Warning Signal #5: Widening Credit Spreads
The fifth and perhaps most immediate warning signal involves widening credit spreads the additional yield investors demand over Treasury bonds to compensate for corporate credit risk. Goldman Sachs noted that credit spreads widened meaningfully before the dot-com crash as investors began pricing in higher default probabilities. Bank of America’s research confirms this pattern is emerging: credit spreads on hyperscaler bonds have widened from 50 to 80 basis points over U.S. Treasuries.
Widening spreads signal deteriorating investor confidence in borrowers’ ability to service debt, especially if AI investments fail to generate anticipated returns. Bank of America chief strategist Michael Hartnett recommended clients open short positions on hyperscaler bonds, anticipating further debt issuance that could pressure prices and widen spreads further. When seasoned strategists at major banks advise shorting corporate bonds, it reflects serious concerns about credit quality and market pricing.
The spread widening from 50 to 80 basis points represents a 60% increase in the risk premium investors demand. While 80 basis points remains relatively modest compared to distressed levels (300+ basis points), the trajectory matters. If spreads continue widening toward 100-150 basis points, refinancing costs for companies with maturing debt could spike dramatically, potentially triggering a credit crunch.
Inside the Data Center Debt Frenzy
Meta’s Record-Breaking Bond Offering
Meta’s $30 billion financing in early November 2025 stands as the year’s largest corporate debt sale and among the largest in history. The transaction included $27 billion in debt and $3 billion in equity, structured through a special purpose vehicle to finance a Louisiana data center. The offering drew $125 billion in orders, a 4.2x oversubscription ratio demonstrating that institutional investors remain willing to fund AI infrastructure despite mounting bubble concerns.
Meta’s existing debt load totals approximately $37 billion, offset by over $60 billion in cash reserves. The company generates more than $20 billion in quarterly operating cash flow with interest expenses under $200 million per quarter, maintaining a comfortable debt service coverage ratio. However, the new $27 billion in debt will increase interest obligations substantially, particularly if issued at yields reflecting the wider credit spreads Bank of America flagged.
The special purpose vehicle structure merits scrutiny because it allows Meta to pursue off-balance-sheet accounting treatment for some obligations. While this can optimize capital structure and tax efficiency, it also reduces transparency for equity and bond investors trying to assess true leverage and risk exposure. Paul Kedrosky warns this approach creates ambiguity about “who ultimately holds the risk” if AI investments underperform.
Oracle’s Aggressive Leverage Strategy
Oracle has pursued an even more aggressive debt strategy, with total debt reaching approximately $96 billion following its $18 billion bond issuance and a separate $38 billion loan secured in late 2025. The company is building data centers for the Stargate AI infrastructure initiative in New Mexico, betting on massive demand for cloud-based AI computing capacity. With quarterly net income around $3 billion, Oracle’s debt-to-annual-earnings ratio exceeds 8x a level that leaves little margin for error.
If interest rates rise or AI revenue growth disappoints, Oracle’s interest payments could climb to consume a substantial portion of profits. The company essentially “is spending money it doesn’t have on facilities that haven’t been built for customers it doesn’t have,” as one analyst described the speculative nature of the buildout. This bet-the-company approach resembles the telecommunications infrastructure overbuilding that preceded the dot-com crash, when companies like WorldCom and Global Crossing bankrupted themselves constructing networks with insufficient demand.
Oracle’s credit profile depends entirely on AI adoption accelerating fast enough to fill its data centers with paying customers before debt service obligations overwhelm cash flow. Any delay in monetization or pricing pressure from competitors could trigger a liquidity crisis, making Oracle a bellwether for AI debt sustainability.
The Hyperscaler Spending Arms Race
Amazon, Microsoft, Alphabet, and Apple join Meta in a collective spending spree projected to reach $349 billion in 2025 for the five companies alone. Google’s Gemini AI product reported a 130-fold increase in AI token consumption over the past 18 months, while Meta’s compute needs “continued to expand meaningfully and exceeded its expectations”. These usage metrics justify continued investment from a demand perspective, but the scale has outstripped internal funding capacity.
Microsoft announced it will purchase $9.7 billion worth of computing capacity from Australian data center operator IREN, while OpenAI and Amazon signed a seven-year, $38 billion agreement for AI infrastructure. These commitments lock companies into massive future capital outlays regardless of near-term economic conditions or AI monetization progress.
JPMorgan strategists frame the financing challenge bluntly: “The question is not ‘which market will finance the AI boom?’ Rather, the question is ‘how will financings be structured to access every capital market?'”. The bank projects that even with $1.5 trillion in investment-grade bonds and $150 billion in leveraged finance over five years, a $1.4 trillion gap remains. Private credit and government support must fill this void, raising systemic risk questions about concentration and interconnectedness.
McKinsey projects $7 trillion in total data center investment needed by 2030, with Morgan Stanley estimating $800 billion in private credit required within the next two years alone. The New York Times reported that debt financing for data centers could exceed $1 trillion by 2028, representing over a third of all spending. These figures dwarf the capital intensity of previous technology buildouts, including the dot-com era internet infrastructure and 3G/4G wireless networks.
Bank of America’s Cash Crunch Warning
The Self-Funding Model Breaks Down
Bank of America’s research highlights a critical inflection point: the self-funded model that drove AI’s initial expansion is breaking down under the weight of infrastructure demands. For years, tech giants financed R&D and cloud expansion from strong cash flows generated by advertising, software subscriptions, and existing cloud services. This internal funding approach minimized financial risk and allowed companies to scale without leverage.
The data center requirements for training and deploying advanced AI models have shattered this equilibrium. Bank of America analysts note that capex shares for Amazon, Alphabet, Microsoft, and others “rising sharply” suggests “the self-funded model that drove rapid AI progress may be causing strain, forcing companies to tap debt markets to maintain momentum”. The $75 billion in bonds and loans issued during just September and October 2025 provides stark evidence of this shift.
Ruth Yang from S&P Global Ratings observes that “the Metas, the Apples of the world are all funding” AI, but critically, “sometimes they put it on their balance sheets, and sometimes they don’t”. This selective disclosure raises transparency concerns for investors trying to assess true debt exposure and financial risk. The pivot to external financing introduces refinancing risk, interest rate sensitivity, and covenant constraints that didn’t exist under the self-funded model.
Chief Strategist Michael Hartnett’s Short Position Call
Michael Hartnett, Bank of America’s chief investment strategist, issued a striking recommendation in November 2025: open short positions on hyperscaler bonds in anticipation of further debt issuance. When a major bank strategist advises clients to bet against corporate bonds, it signals serious concern about credit quality, supply-demand dynamics, or both.
Hartnett’s call reflects expectations that ongoing AI infrastructure needs will force companies to issue additional debt, potentially flooding the market and depressing bond prices. As supply increases, investors can demand higher yields (wider spreads), making existing bonds worth less. The credit spread widening from 50 to 80 basis points validates this thesis.
The recommendation also suggests Bank of America analysts believe current bond prices don’t adequately reflect the risk that AI investments may fail to generate returns sufficient to service growing debt loads. If monetization disappoints or competition compresses margins, highly leveraged companies could face credit rating downgrades or default risk, hammering bond values.
Credit Market Warning Signs Investors Can’t Ignore
Widening Credit Spreads Explained
Credit spreads measure the difference between yields on corporate bonds and risk-free government bonds (Treasuries). When spreads widen, it means investors demand higher compensation for lending to corporations, reflecting increased perceived risk. The movement from 50 to 80 basis points on hyperscaler bonds represents a 60% increase in the risk premium.
To put this in practical terms: if 10-year Treasury bonds yield 4.0%, a corporate bond with a 50 basis point spread would yield 4.5%, while an 80 basis point spread pushes the yield to 4.8%. For a $10 billion bond issuance, the difference amounts to $30 million in additional annual interest expense multiplied across dozens of transactions, this increases corporate debt burdens significantly.
Goldman Sachs noted that credit spreads widened before the dot-com crash as investors began pricing in higher default risk. The current widening pattern suggests bond market participants are less confident about AI companies’ ability to generate returns that justify their debt accumulation. If spreads continue toward 100-150 basis points, it would signal acute stress and potentially trigger a broader repricing of AI-exposed securities.
Private Credit’s Opacity Problem
The rise of private credit as a funding source for AI infrastructure introduces systemic transparency challenges. Private credit involves loans from non-bank lenders like Apollo Global Management, Blackstone, and Ares Management, often structured through complex vehicles that avoid public disclosure requirements.
Paul Kedrosky describes the problem: “Since much of this debt sits in special purpose vehicles, it can be hard to know who ultimately holds the risk: Big Tech, the lenders, or the investors in funds that own the loans”. This opacity matters because hidden leverage and interconnected exposures amplified both the 2000 tech crash and the 2008 financial crisis.
Meta’s reported exploration of a $29 billion private credit deal, potentially including up to $26 billion in debt, exemplifies the scale. Blue Owl committed $15 billion to a data center joint venture with OpenAI, while the broader $500 billion AI data center project involving SoftBank and Oracle relies heavily on private financing. If these investments underperform, losses could cascade through pension funds, insurance companies, and wealth management platforms that own private credit funds.
Ruth Yang from S&P Global warns that investors are likely to see “more private debt moving forward,” further complicating efforts to assess systemic risk. Unlike public bonds traded on exchanges with transparent pricing, private credit transactions occur in dark markets where valuations can diverge significantly from economic reality until defaults force markdowns.
Comparison to Late 1990s Credit Conditions
Goldman Sachs strategists explicitly compare current AI sector dynamics to the technology credit environment of 1997-2000. During the late 1990s, telecommunications and internet companies borrowed aggressively to build infrastructure fiber optic networks, data centers, internet exchanges assuming demand would justify the investment. When the dot-com bubble burst, many of these companies declared bankruptcy, triggering tens of billions in bondholder losses.
Key similarities include: rapid debt accumulation to fund infrastructure with uncertain demand; widening credit spreads as investors grow nervous; Federal Reserve rate cuts that fuel risk-taking; and competitive pressure forcing companies to spend regardless of financial prudence. Goldman notes that several warning signs appeared “at least two years before the dot-com bubble burst,” suggesting the current 2025 signals could precede a 2027 crisis if the pattern holds.
Important differences also exist. Today’s AI leaders like Meta, Microsoft, and Alphabet generate massive, real revenues and profits, unlike the late-1990s dot-coms with negligible cash flow. Current debt-to-profit ratios remain “significantly lower than at the peak of the internet bubble,” according to Goldman. Most firms today finance a substantial portion of capex with free cash flow, reducing immediate leverage risk.
However, the Bank of England’s Financial Policy Committee warned in October 2025 that “the risk of a sharp market correction has increased,” marking its most serious caution yet regarding AI-induced market risks. The BoE noted that the risk of repercussions for the UK’s financial system from an AI market correction is “material,” suggesting international banking regulators view the debt accumulation with alarm.
The Dot-Com Parallel: How Worried Should Investors Be?
Where We Are in the Cycle
Goldman Sachs strategists carefully calibrated their warning, stating that the AI trade now looks like tech stocks did in 1997 “several years before the bubble burst” rather than the peak 1999 mania. This timing matters because it suggests investors have a window to adjust positions and monitor risk indicators before a potential crisis.
The bank emphasized that “stocks don’t look like they’re in their 1999 moment yet,” but “there is a growing risk that the imbalances that built up in the 1990s will become more visible as the AI investment boom extends”. This graduated warning acknowledges AI’s transformative potential while flagging financial excesses that could derail the trend.
Historical patterns show that warning signs appeared “at least two years before the dot-com bubble burst”. The NASDAQ peaked in March 2000, but telecommunications companies began showing stress in 1998-1999 as debt burdens mounted and returns disappointed. If the current AI cycle follows a similar trajectory, the 2025 warning signals could precede a 2027-2028 correction.
Ray Dalio, founder of Bridgewater Associates, has warned that the Fed’s easing cycle could inflate market bubbles, adding credibility to Goldman’s concerns. Bank of England officials escalated their warnings in October 2025, noting “material” risk to financial system stability from AI-related exposures. BofA’s October 2025 survey found that 54% of investors believe AI stocks are in a bubble, though 38% disagree.
Key Differences From 2000
Several critical differences distinguish today’s AI boom from the 2000 dot-com bubble. First, current AI leaders generate massive, real revenues and profits. Meta produces over $20 billion in quarterly operating cash flow; Microsoft and Alphabet post similar figures. In contrast, many dot-com darlings of 1999-2000 recorded negligible revenue and burned cash at unsustainable rates.
Second, debt-to-profit ratios remain significantly lower than at the internet bubble’s peak, according to Goldman Sachs. While Oracle’s 8x debt-to-earnings ratio causes concern, the industry average looks healthier. Nvidia, the AI chip leader, carries less than $10 billion in debt against massive free cash flows from semiconductor sales, exemplifying strong balance sheet management.
Third, most leading AI companies fund a substantial portion of capital expenditures with free cash flow rather than pure debt financing. Even as external borrowing increases, internal cash generation provides a buffer that didn’t exist for many dot-com companies. Meta’s debt service coverage ratio operating cash flow divided by interest expense remains comfortably above 100x.
Fourth, AI technology has demonstrated clear commercial applications and revenue potential, whereas many dot-com business models lacked viable monetization paths. Google’s 130-fold increase in AI token consumption and Meta’s exceeding compute demand projections show real usage driving infrastructure needs.
Concerning Similarities
Despite these differences, troubling parallels emerge. The competitive pressure forcing companies to spend regardless of immediate returns echoes the late 1990s. Paul Kedrosky captures this dynamic: “The big tech companies don’t care whether the investment has any return, because they’re in a race”. This “arms race” mentality led to catastrophic overinvestment in the dot-com era.
Goldman Sachs CEO David Solomon expressed similar concerns at the Italian Tech Week conference, warning he expects a “drawdown” in the stock market within the next couple of years because “there will be a lot of capital that’s deployed that will turn out to not deliver returns”. Solomon’s observation that some investors are “out on the risk curve because they’re excited” describes classic bubble psychology.
Credit market strain is building exactly as it did before 2000. Widening spreads, surging debt issuance, and increasing reliance on complex off-balance-sheet structures all replicate warning signs from the late 1990s. The Bank of America strategist’s call to short hyperscaler bonds echoes the credit market skepticism that emerged in 1999 before the crash.
Jeff Bezos acknowledged at the Italian Tech Week conference that “there is a bubble in the AI industry,” though he remained optimistic about long-term technology benefits. When founder-CEOs of major AI investors concede bubble conditions exist, it validates concerns that valuations and spending have detached from near-term fundamentals.
What This Means for Different Types of Investors
For Tech Stock Holders
Equity investors holding tech stocks must assess their portfolio’s exposure to AI debt accumulation and spending risks. Companies carrying manageable debt with strong cash flow generation like Nvidia, with minimal leverage and robust chip sales present lower risk than highly leveraged players like Oracle. Investors should calculate debt-to-equity ratios, interest coverage ratios (EBITDA divided by interest expense), and capex as a percentage of revenue for all AI-exposed holdings.
Valuation multiples deserve particular scrutiny. Goldman CEO David Solomon warned that investors are “out on the risk curve because they’re excited,” leading to valuations that assume flawless execution. Companies trading at 40+ times earnings with accelerating debt issuance face significant downside if AI monetization disappoints. Historical precedent shows that high-multiple stocks suffer disproportionate corrections when growth narratives stumble.
Diversification becomes critical when concentrated risks emerge in a sector. Investors heavily weighted toward AI-exposed tech stocks should consider rebalancing toward quality companies with strong balance sheets, defensive sectors less dependent on AI hype, and international markets with different risk profiles. Position sizing should reflect the uncertainty Goldman Sachs and Bank of America have highlighted, limiting exposure to any single company or theme to levels that won’t devastate portfolio returns if corrections occur.
For Bond Investors
Fixed income investors face direct exposure to the credit risks Bank of America and Goldman flagged. Credit spread widening from 50 to 80 basis points signals deteriorating risk-adjusted returns for hyperscaler bonds. Investors holding existing bonds have experienced mark-to-market losses as spreads widened; those considering new purchases must evaluate whether 80 basis points adequately compensates for potential further deterioration.
Michael Hartnett’s recommendation to short hyperscaler bonds suggests Bank of America analysts expect additional supply to pressure prices. Bond investors should monitor issuance calendars and assess whether their holdings face refinancing risks in coming years. Companies with large debt maturities in 2026-2028 could face substantially higher borrowing costs if spreads continue widening.
Duration considerations matter for bond portfolio management. Longer-dated bonds (10-30 years) experience larger price fluctuations when spreads change than shorter-dated securities. Investors concerned about credit quality deterioration should consider shortening duration by rotating toward 3-5 year maturities, reducing exposure to long-term default and refinancing risks.
Investment-grade versus high-yield AI exposure requires differentiation. JPMorgan projects $1.5 trillion in investment-grade bond issuance over five years, with an additional $150 billion in leveraged (high-yield) finance. Investment-grade issuers like Meta and Microsoft present lower default risk than smaller, unprofitable AI companies issuing junk bonds. Credit ratings, interest coverage ratios, and debt-to-EBITDA metrics help investors distinguish quality levels within AI-exposed debt.
For Institutional Portfolio Managers
Institutional investors managing pension funds, endowments, or wealth management portfolios must assess systemic exposure to AI debt across asset classes. Direct holdings of tech stocks and bonds represent obvious exposures, but indirect risks lurk in private credit funds, collateralized loan obligations, and infrastructure investments.
Stress testing portfolio performance under AI correction scenarios provides essential risk management. The Bank of England warned that repercussions from an AI market correction pose “material” risk to financial system stability, suggesting contagion could spread beyond direct AI holdings. Modeling a 30-40% drawdown in AI-exposed equities, combined with credit spread widening to 150-200 basis points, reveals concentration risks and guides position adjustments.
Correlation risk deserves particular attention. If AI debt concerns trigger broader tech sector selling or risk-off sentiment, correlations across assets may spike toward 1.0, undermining diversification benefits. Institutional portfolios that appear diversified under normal conditions can experience synchronized losses during systemic events. Adding truly non-correlated assets, certain commodities, international bonds, alternative strategies enhances resilience.
Hedging strategies merit consideration for portfolios with significant AI exposure. Options strategies, tail risk hedging, or strategic shorts (as Bank of America’s Hartnett suggested) can offset downside while maintaining upside participation. Position sizing across the risk spectrum from conservative investment-grade bonds to aggressive growth equity should reflect Goldman’s warning that imbalances are growing.
Expert Perspectives and Institutional Views
The Bank of England’s Financial Policy Committee issued its most serious warning yet in October 2025, stating that “the risk of a sharp market correction has increased” due to AI-related market dynamics. The BoE assessment that repercussions for the UK’s financial system are “material” signals international regulatory concern about contagion risks from AI debt accumulation. Central banks rarely issue specific warnings about technology sector risks, making this statement particularly notable.
David Solomon, CEO of Goldman Sachs, told the Italian Tech Week conference audience that he expects a “drawdown” in stock markets within the next couple of years because of enormous capital deployment into AI projects. “I think that there will be a lot of capital that’s deployed that will turn out to not deliver returns, and when that happens, people won’t feel good,” Solomon explained. His observation that some investors are “out on the risk curve because they’re excited” captures the speculative sentiment driving valuations.
Jeff Bezos, Amazon founder and major AI investor, acknowledged at the same conference that “there is a bubble in the AI industry”. However, Bezos maintained long-term optimism about AI’s transformative potential, drawing parallels to past technology cycles where bubbles accompanied genuine innovation. His candid acknowledgment validates concerns while arguing that eventual value creation justifies near-term excess.
Paul Kedrosky, venture capitalist and MIT research fellow, emphasizes the monetization uncertainty underlying AI debt: “It remains unclear how that debt will be paid back: AI isn’t making money (yet), and with each new chip cycle comes costly upgrades”. Kedrosky warns this “financing binge could pop the AI bubble” if returns fail to materialize before debt service obligations mount.
Ruth Yang, global head of private markets analytics at S&P Global Ratings, highlights transparency problems: “Sometimes they put it on their balance sheets, and sometimes they don’t”. Yang warns that “investors are likely to see more private debt moving forward,” potentially concentrating risk in less-regulated markets.
BofA Global Research conducted a survey in October 2025 finding that 54% of investors believe AI stocks are in a bubble, while 38% disagree. This divided sentiment suggests the market hasn’t reached the euphoric consensus that typically characterizes bubble peaks, where skepticism vanishes entirely. However, the fact that a majority sees bubble conditions while prices continue rising indicates that fear of missing out may be overriding fundamental concerns.
Actionable Risk Monitoring Framework
Key Metrics to Watch Quarterly
Investors should systematically monitor five critical metrics each quarter to track whether AI debt risks are escalating:
1. Aggregate Tech Capex as Percentage of Sector Revenue: UBS projects this ratio will reach 27% by 2030, up from 20.8% currently. If quarterly reports show capex intensity accelerating beyond 30%, it signals companies are prioritizing growth over profitability in potentially unsustainable ways. Download earnings transcripts and financial supplements to calculate this ratio for your holdings.
2. Corporate Debt Growth Rates: Track total debt and net debt (debt minus cash) growth rates for AI-exposed companies. Bank of America flagged the $75 billion in just two months as abnormal. If quarterly debt issuance continues at this pace, it confirms the self-funding model breakdown and escalating leverage risks.
3. Credit Spread Movements: Monitor spreads on investment-grade tech bonds weekly using financial data terminals or public bond market data. The movement from 50 to 80 basis points represents a meaningful risk repricing. If spreads widen beyond 100 basis points, it signals acute credit stress and potential rating downgrades.
4. Free Cash Flow vs. Capex Ratios: Calculate free cash flow (operating cash flow minus capital expenditures) for AI companies each quarter. Goldman Sachs noted that most firms currently fund substantial capex from free cash flow. If this ratio turns negative capex exceeding operating cash flow it forces companies into pure debt financing, dramatically increasing financial risk.
5. Fed Policy Trajectory and Rate Expectations: Track Federal Reserve statements, meeting minutes, and CME FedWatch probabilities for rate changes. Goldman identified Fed rate cuts as a warning signal that fuels bubbles. If the Fed pivots back to rate hikes or pauses cuts due to inflation concerns, borrowing costs for AI infrastructure would spike, potentially triggering refinancing stress.
Red Flags That Signal Heightened Risk
Several specific events should trigger portfolio rebalancing or hedging actions:
Capex Announcements Exceeding Analyst Estimates by More Than 15%: When companies like Meta raise guidance from $66-72 billion to $70-72 billion, it shows spending accelerating beyond expectations. If future revisions exceed analyst consensus by 15%+ repeatedly, it suggests companies are locked in an arms race losing financial discipline.
Multiple Bond Offerings Within Six-Month Windows: Meta’s $30 billion offering followed by additional issuance within months would signal desperate capital needs. Frequent debt market visits indicate companies can’t fund operations from cash flow and face rollover risk.
Credit Rating Downgrades: Any major AI company suffering a credit rating downgrade from agencies like S&P, Moody’s, or Fitch signals deteriorating financial health. Downgrades trigger higher borrowing costs and can force institutional investors to sell holdings due to mandate restrictions.
Widening Spreads Beyond 100 Basis Points: If credit spreads on hyperscaler bonds exceed 100 basis points over Treasuries (vs. current 80), it indicates severe credit market stress. At 150+ basis points, distressed conditions emerge, presaging potential defaults.
Insider Selling Patterns at Tech Giants: Monitor SEC Form 4 filings showing insider transactions. If executives at multiple AI leaders simultaneously increase stock sales, it may signal they believe valuations have overshot fundamentals. While individual sales can reflect diversification needs, coordinated patterns warrant attention.
Portfolio Protection Strategies
Investors concerned about AI bubble risks have several tactical options:
Stress Testing AI Sector Exposure: Calculate the percentage of your portfolio directly or indirectly exposed to AI through tech stocks, corporate bonds, and private credit funds. Model a 30-40% drawdown scenario to assess total portfolio impact. If losses would exceed your risk tolerance, rebalance positions.
Rebalancing Triggers: Establish clear rules for reducing AI exposure. For example: “If credit spreads exceed 100 basis points, reduce tech bond allocation by 25%” or “If aggregate capex-to-revenue reaches 30%, trim AI equity positions by 20%”. Rules-based approaches remove emotion from decisions during volatile periods.
Defensive Positioning Options: Rotate toward quality companies with fortress balance sheets, minimal debt, strong cash flows, diversified revenue streams. Nvidia exemplifies this profile with less than $10 billion debt and massive free cash flow. Defensive sectors like utilities, consumer staples, and healthcare typically outperform during tech corrections.
Quality Factor Emphasis: Academic research shows quality factors (high return on equity, stable earnings, low leverage) outperform during risk-off periods. Tilt portfolios toward companies with superior financial metrics rather than pure growth stories dependent on AI hype.
Bank of America’s Michael Hartnett suggested sophisticated investors consider short positions on hyperscaler bonds. While shorting requires expertise and carries unlimited loss potential, put options on tech ETFs or bond funds offer defined-risk hedging alternatives. Consult a qualified financial advisor before implementing complex strategies.
Frequently Asked Questions
What is an AI debt bubble and why are Goldman Sachs and Bank of America warning about it?
An AI debt bubble involves unsustainable debt accumulation by technology companies financing artificial intelligence infrastructure. Goldman Sachs warns the AI trade resembles 1997, years before the dot-com crash, with five warning signals reappearing: peak investment spending, declining profitability, rapid debt growth, Fed rate cuts, and widening credit spreads. Tech companies issued $75 billion in AI debt during September-October 2025 alone, more than double typical annual averages.
How much debt are tech companies raising for AI investments?
Tech giants are raising unprecedented AI debt levels. Bank of America documented $75 billion in investment-grade debt from AI companies in September-October 2025 alone. Meta’s $30 billion bond offering—the year’s largest corporate debt sale drew $125 billion in orders, while Oracle’s total debt reached $96 billion after securing $18 billion in bonds plus a $38 billion loan. JPMorgan projects AI hyperscalers need $1.5 trillion in bonds over five years, yet a $1.4 trillion funding gap remains. The Big Five tech companies will spend $349 billion on capital expenditures in 2025, with industry-wide AI capex reaching $571 billion by 2026 according to UBS.
Is the AI boom really comparable to the dot-com bubble?
Goldman Sachs compares the AI boom to 1997 tech dynamics before the 2000 crash—noting similarities including rapid infrastructure debt, widening credit spreads, and competitive pressure driving spending. CEO David Solomon warns “there will be a lot of capital that’s deployed that will turn out to not deliver returns.” However, critical differences exist: today’s AI leaders like Meta, Microsoft, and Alphabet generate massive real revenues unlike dot-com startups with negligible cash flow. Current debt-to-profit ratios remain “significantly lower” than the bubble peak, and firms finance substantial capex with free cash flow. Goldman suggests we’re in a “1997 moment,” not the peak “1999 moment.
What are the five warning signals Goldman Sachs identified?
Goldman Sachs strategists Dominic Wilson and Vickie Chang identified five warning signals from the late 1990s bubble that investors should monitor:
- Peak investment spending: Technology capital expenditure surging to unsustainable levels, with AI capex reaching $349 billion in 2025 and projected $571 billion by 2026.
- Declining corporate profitability: Rising investment combined with falling profitability pushing corporate financial balance into deficit, forcing companies to fund gaps through debt.
- Rapid corporate debt growth: Accelerating debt accumulation, particularly when debt grows faster than profits, with $75 billion issued in just two months in 2025.
- Federal Reserve rate cuts: The Fed cut rates by 25 basis points in October 2025 with another cut expected in December, providing monetary fuel that can inflate asset bubbles.
- Widening credit spreads: Credit spreads on hyperscaler bonds widening from 50 to 80 basis points over Treasuries, signaling deteriorating investor confidence in debt serviceability.
Goldman emphasized that these warning signs appeared “at least two years before the dot-com bubble burst,” suggesting investors have time to adjust positions if patterns repeat.
Which tech companies are borrowing the most for AI?
Several major technology companies are leading the AI debt surge:
Meta: Completed a $30 billion bond offering in November 2025 ($27 billion debt, $3 billion equity) for Louisiana data center construction, drawing $125 billion in orders. The company reportedly explored an additional $29 billion private credit deal with up to $26 billion in debt. Meta’s total debt stands at approximately $37 billion against $60 billion in cash reserves.
Oracle: Secured $18 billion in bond issuance plus a $38 billion loan in late 2025, bringing total debt to approximately $96 billion. The company is building data centers for the Stargate AI initiative in New Mexico.
Amazon, Microsoft, Alphabet, and Apple: These companies collectively are projected to spend $349 billion on capital expenditures in 2025, much of it AI-related. Microsoft announce
