All scenarios

AI Bubble Risk: Strong Demand, Fragile Financing

Strengthening

The evidence points to elevated bubble risk rather than proof of a completed bubble. AI demand and revenue growth are exceptionally strong, but investment is increasingly dependent on debt, private credit, large future commitments, and potentially circular financing. A slowdown in earnings, funding, or infrastructure demand could therefore trigger a sharp repricing.

October 25, 2026 — AI fundamentals remain powerful, but financial conditions surrounding the build-out are becoming more fragile. Valuations are described as stretched, investment may exceed socially efficient levels, and capital spending is outpacing cash flows. The central risk is not that AI has no value, but that expected returns and deployment timelines fail to justify the scale and financing structure of current commitments.

Geography: United States, Taiwan

Portfolio implications

Your exposure

How this scenario reads against the holdings in your portfolio: which positions carry more exposure, and which look relatively resilient.

Log in to see how this scenario affects your portfolio

Developments

  1. Broad market correction reflects growing skepticism about AI valuations

    Strengthens the thesis

    The S&P 500 fell 10% from its October high as investors questioned whether AI-company valuations had risen too far and whether demand for processors, memory, and data-center equipment could weaken if AI profitability disappointed.

    Why it matters: The correction indicates that concerns about AI valuation and infrastructure-demand assumptions are affecting broader market pricing, increasing the risk of further repricing if earnings or demand expectations deteriorate.

    Potential beneficiaries

    • Investors positioned defensively
    • Potential buyers of AI-related assets after a repricing

    Potential losers

    • AI-company shareholders
    • Suppliers dependent on sustained processor, memory, and data-center demand

    Market evidence of heightened sensitivity to AI profitability and demand expectations; it signals concern but is not by itself proof of a completed bubble.

  2. Record-scale GPU debt financing raises leverage concerns

    Strengthens the thesis

    Apollo and Goldman Sachs arranged approximately $44 billion in debt for SpaceX to purchase Nvidia GPUs and related equipment, including senior and subordinated notes with senior secured yields of 7.66% and 8.259%.

    Why it matters: This shows that AI infrastructure expansion is being financed at unusually large scale through debt, increasing the financial exposure attached to continued demand for AI computing capacity.

    Potential beneficiaries

    • Apollo
    • Goldman Sachs
    • Nvidia

    Potential losers

    • SpaceX if AI infrastructure demand or returns weaken
    • Debt investors exposed to the financing

    A significant new indicator of leverage in AI infrastructure financing, though the evidence does not establish that the financing is distressed or unsustainable.

  3. Data-center credit conditions are tightening

    Strengthens the thesis

    Global lenders are reducing or reassessing data-center loan exposure, shifting risk to insurers, and demanding higher funding costs from data-center asset-backed securities.

    Why it matters: This is fresh evidence that financing fragility is becoming an active constraint, potentially raising borrowing costs and limiting capital available for continued infrastructure expansion.

    Potential beneficiaries

    • Insurers assuming transferred credit exposure
    • Lenders able to reprice or selectively retain stronger borrowers

    Potential losers

    • Highly leveraged data-center developers
    • Investors in data-center asset-backed securities
    • Projects dependent on cheap refinancing

    Meaningfully strengthens the financing-risk component of the scenario, though it does not establish a broad collapse in credit availability.

  4. Policy is accelerating infrastructure buildout while shifting energy costs to data centers

    Weakens the thesis

    The Senate approved legislation to speed permitting for power and transmission infrastructure while requiring data centers to pay for their energy consumption and associated transmission upgrades.

    Why it matters: Faster approvals could ease a key physical bottleneck and support continued capacity expansion, but direct cost responsibility may reduce project economics or expose weaker developments to higher operating and connection costs.

    Potential beneficiaries

    • Data centers that can absorb or pass through energy and transmission costs
    • Power and transmission infrastructure developers

    Potential losers

    • Marginal data-center projects with high energy requirements
    • Customers or investors if the added costs are passed through

    Partly offsets infrastructure-supply constraints but has mixed implications for bubble risk: it supports real demand expansion while increasing the costs borne by data-center operators.

  5. AI infrastructure investment is becoming more dependent on external financing

    Strengthens the thesis

    A Federal Reserve analysis finds that data-center builders are relying more on debt while reducing shareholder payouts and cash accumulation, and warns that sustaining investment will increasingly require profitability, retained earnings, and continued access to external funding.

    Why it matters: The funding mix leaves infrastructure spending more exposed to weaker earnings, reduced cash generation, or a loss of investor and lender appetite.

    Potential beneficiaries

    • Semiconductor producers funding investment largely from operating cash flow
    • Credit providers if risk-adjusted returns improve

    Potential losers

    • Debt-dependent data-center builders
    • Shareholders facing lower payouts and greater financial risk

    Strengthens the case that the investment boom has a more fragile financing base than the semiconductor segment, while stopping short of showing that profitability or external funding has already failed.

  6. AI demand remains strong, but expansion is increasingly exposed to capacity and policy constraints

    Weakens the thesis

    Alibaba’s AI cloud and compute revenue rose 53% year over year, while capital expenditures increased 60% as it expanded AI infrastructure. The company nevertheless warned that U.S. restrictions on remote access to advanced chips could affect global data-center plans.

    Why it matters: The evidence confirms substantial operating demand but also shows that monetization requires heavy upfront spending and can be disrupted by technology-access restrictions.

    Potential beneficiaries

    • AI cloud and compute providers with compliant chip access
    • Suppliers serving domestic Chinese AI infrastructure

    Potential losers

    • Companies dependent on restricted advanced-chip access
    • Projects whose economics rely on unrestricted global deployment

    Demand remains a support, but policy and infrastructure constraints make the growth path less predictable.

  7. Tighter monetary conditions could raise the cost of sustaining AI infrastructure spending

    Strengthens the thesis

    Federal Reserve officials broadly anticipated a December rate increase and expressed concern that inflation could persist even if the AI-infrastructure buildout weakened.

    Why it matters: Higher-for-longer rates would make debt-funded expansion more expensive and could reduce the valuation support available to companies whose investment plans depend on continued access to capital.

    Potential beneficiaries

    • Cash-rich companies
    • Lenders able to command higher rates

    Potential losers

    • Debt-dependent AI infrastructure investors
    • High-growth companies valued on distant cash flows

    A developing macroeconomic headwind that increases the sensitivity of AI financing and valuations to interest rates.

  8. Hyperscaler cash-flow strain and cross-market concentration are now explicit stability concerns

    Strengthens the thesis

    The ECB reported that most U.S. hyperscalers’ free cash flow turned negative in 2026 as AI infrastructure spending outpaced operating cash flow, increasing reliance on external equity and debt. It also identified growing concentration of AI exposure across equity, bond, private-credit, and private-equity markets.

    Why it matters: This indicates that strong infrastructure spending is already creating financing needs and that a downturn could transmit losses across multiple funding channels rather than remaining confined to individual technology companies.

    Potential beneficiaries

    • Providers of external equity and debt financing
    • Investors positioned defensively against concentrated AI exposure

    Potential losers

    • Highly leveraged hyperscalers
    • AI-focused equity, bond, private-credit, and private-equity investors if sentiment reverses

    A material escalation in evidence that AI investment is generating systemic financing vulnerability.

  9. Anthropic’s IPO exposes financing fragility and attracts official scrutiny

    Weakens the thesis

    Anthropic is advancing toward an IPO while disclosing $6.2 billion in debt. State treasurers and comptrollers have separately warned that the offering could amplify exposure to circular financing in the AI sector.

    Why it matters: The combination of substantial issuer debt, a major public offering, and concerns from public officials suggests that market access and investor confidence are becoming important transmission points for AI-related financial risk.

    Potential beneficiaries

    • Investors seeking greater disclosure of AI-sector leverage
    • Officials and regulators focused on financial-system exposure

    Potential losers

    • Highly leveraged AI companies
    • Investors exposed to a repricing of AI IPOs and related financing

    This strengthens the case for elevated financing risk, but it does not establish that circular financing is occurring or that Anthropic’s IPO will fail.

  10. Local opposition is beginning to constrain AI infrastructure deployment

    Weakens the thesis

    Salt Lake City imposed a 180-day moratorium on new data centers, creating a temporary barrier to expansion of the physical infrastructure supporting AI workloads.

    Why it matters: Even with strong demand and available capital, local permitting and political constraints can delay capacity additions, increase development costs, or disrupt the pace assumed in infrastructure investment plans.

    Potential beneficiaries

    • Residents and local governments seeking time to assess data-center impacts
    • Existing data centers facing less immediate local competition

    Potential losers

    • Data-center developers and hyperscalers planning near-term expansion
    • Financiers relying on rapid infrastructure deployment

    The moratorium is a localized and temporary constraint, so it provides evidence of implementation risk rather than a broad reversal in AI infrastructure demand.

  11. AI infrastructure expansion is becoming increasingly dependent on credit markets

    Weakens the thesis

    Morgan Stanley estimates that credit markets could fund more than half of approximately $3.8 trillion in hyperscaler spending from 2026 through 2028. The New York Fed president also compared the sector’s increasingly complex financing arrangements with those preceding the 2008 financial crisis.

    Why it matters: A large debt-funded share of planned infrastructure spending increases sensitivity to interest rates, credit availability, and lender risk appetite. Further rate increases could raise the cost of sustaining the buildout.

    Potential beneficiaries

    • Banks and private lenders arranging AI infrastructure financing
    • Credit investors able to price and manage the associated risk

    Potential losers

    • Hyperscalers and infrastructure developers dependent on continued borrowing
    • AI projects whose economics rely on low financing costs

    This materially strengthens the financing-fragility concern, although the evidence describes projected spending and risk comparisons rather than realized losses or defaults.

  12. Nvidia’s financing role increases ecosystem concentration risk during a leadership transition

    Strengthens the thesis

    Nvidia is reported to provide equity investments, co-investments, and debt backstops to companies including CoreWeave and Lumentum while also undergoing a leadership transition.

    Why it matters: The combination of financial support from a central supplier and executive transition could make ecosystem health more dependent on Nvidia’s continued capacity and willingness to support counterparties.

    Potential beneficiaries

    • Companies receiving Nvidia-backed financing
    • Nvidia ecosystem participants

    Potential losers

    • Counterparties if Nvidia reduces financial support
    • Investors exposed to concentrated AI financing relationships

    The evidence identifies potential financing interconnectedness, but does not establish that these arrangements are loss-making or circular.

  13. AI infrastructure commitments are scaling toward a $2 trillion 2027 market

    Strengthens the thesis

    Omdia projects AI infrastructure spending will exceed $2 trillion in 2027, with AI-optimized-server spending rising sharply from $527.6 billion in 2026 to $957.2 billion. The forecast comes alongside continued chip and memory constraints, indicating that the expansion depends on sustained demand and available supply.

    Why it matters: The projected spending surge raises the scale of capital exposed to any disappointment in AI demand, deployment economics, or supply availability.

    Potential beneficiaries

    • AI infrastructure suppliers
    • Chip and memory manufacturers

    Potential losers

    • Investors exposed to AI infrastructure if projected demand falls short
    • AI developers facing constrained chip and memory availability

    This is a forward-looking industry projection rather than evidence that the spending will occur, but it materially increases the potential magnitude of both upside and downside.

  14. Long-term capacity contracts are being paired with project-level financing

    Strengthens the thesis

    AIB Data Centers secured a 12-year, 50-megawatt agreement with Nebius and simultaneously engaged advisers on project financing involving project-level debt and preferred equity for its CLT1 facility.

    Why it matters: Long-duration customer commitments can support infrastructure investment, but the use of debt and preferred equity embeds financing obligations that may become vulnerable if project execution or customer demand changes.

    Potential beneficiaries

    • AIB Data Centers
    • Project lenders and preferred-equity investors
    • Nebius through secured capacity

    Potential losers

    • AIB Data Centers if financing or construction underperforms
    • Debt and preferred-equity providers facing project-specific risk

    The customer agreement provides a source of revenue visibility, while the financing structure shows that AI capacity expansion remains dependent on external capital.

  15. Regulators are identifying hidden leverage and asset-value risks across the AI financing chain

    Weakens the thesis

    The OCC warned that specialized chips and data centers may be difficult to repurpose or may depreciate rapidly, while delays, power constraints, opposition, obsolescence, and changing risk premiums could leave loans mispriced before repayment or refinancing. Separately, the Federal Reserve highlighted more than $3 trillion in potential off-balance-sheet obligations for the seven largest AI firms and over $100 billion in cumulative equity investments by hyperscalers and GPU producers in AI labs and cloud startups.

    Why it matters: The combination of hard-to-redeploy collateral, large contingent commitments, and supplier-to-customer investment links could transmit losses beyond reported balance-sheet debt if growth or customer revenue disappoints.

    Potential beneficiaries

    • Creditors with transparent exposure and robust collateral protections
    • Investors positioned to benefit from clearer disclosure of contingent obligations

    Potential losers

    • Lenders exposed to obsolete or difficult-to-repurpose AI assets
    • AI startups dependent on supplier financing and future customer revenue

    This is a significant escalation in the potential transmission mechanism, although the cited figures include maximum nominal obligations and do not represent expected losses.

  16. AI infrastructure borrowing has become a major share of corporate debt issuance

    Weakens the thesis

    Advanced-computing-related debt issuance exceeded $680 billion by October 2026, about five times the total issued in 2023. It represented roughly 31% of nonfinancial investment-grade corporate debt issuance in 2026, up from less than 6% in 2023.

    Why it matters: The rapid concentration of investment-grade borrowing in AI-related activity raises the financial consequences of any deterioration in infrastructure demand, project economics, or refinancing conditions.

    Potential beneficiaries

    • Credit investors during continued AI expansion
    • Large issuers able to access investment-grade markets

    Potential losers

    • AI borrowers facing a synchronized repricing
    • Corporate bond markets exposed to concentrated AI infrastructure risk

    This materially strengthens the financing-risk case by showing scale and concentration, but issuance growth alone does not establish that the debt is mispriced or unserviceable.

  17. AI-linked debt is beginning to reprice ahead of a potential policy slowdown

    Weakens the thesis

    Bond investors have become more cautious about AI-related debt: Galaxy Helios Data Center bonds fell to 89.25 cents on the dollar, while yields rose to 13% from a 9.875% offering yield. The shift was linked to expectations that a Democrat-controlled Congress could slow data-center expansion.

    Why it matters: This provides a concrete market signal that financing costs and political expectations are already affecting AI infrastructure credit, increasing pressure on highly leveraged projects.

    Potential beneficiaries

    • Bond investors demanding higher risk premiums
    • Data-center lenders with strong covenants or senior claims

    Potential losers

    • Highly leveraged data-center developers
    • AI infrastructure projects dependent on cheap refinancing

    A meaningful early warning of credit-market stress, though it is based on a single reported bond example and political expectations rather than broad evidence of systemic distress.

  18. Secondary-market activity signals reduced lender confidence in data-center debt

    Strengthens the thesis

    A roughly $500 million portfolio of data-center loans, including exposure tied to the Stargate project, was offered in a rare auction, while banks and credit investors became more reluctant to add AI-debt exposure after a market selloff.

    Why it matters: Difficulty distributing or retaining infrastructure loans can expose valuation gaps, weaken refinancing prospects, and amplify repricing if lenders demand higher risk premiums or seek to exit.

    Potential beneficiaries

    • Potential buyers able to acquire loans at discounted prices
    • Lenders reducing or limiting exposure before further deterioration

    Potential losers

    • Existing holders of AI-related loans
    • Projects reliant on smooth loan syndication and refinancing
    • Borrowers facing higher financing costs

    Strong evidence of emerging credit-market caution, though the auction alone does not establish widespread default or a completed bubble.

  19. AI infrastructure credit is entering a formal risk-review phase

    Strengthens the thesis

    Banks and regulators are moving beyond routine lending scrutiny: the ECB is considering a review of data-center exposures, while lenders are stress-testing project cash flows and customer commitments amid greater caution toward AI-related debt.

    Why it matters: This indicates that financing fragility is becoming an institutional risk-management concern, potentially tightening credit and raising the cost of capital for AI infrastructure.

    Potential beneficiaries

    • Banks with limited data-center exposure
    • Credit investors able to negotiate stronger protections
    • Regulators monitoring financial-system risks

    Potential losers

    • Highly leveraged data-center projects
    • AI infrastructure borrowers dependent on continued credit expansion
    • Investors exposed to deteriorating project cash flows

    Moderate-to-strong evidence that financing risk is broadening from individual transactions into bank and regulatory oversight.

  20. AI selloff becomes a global repricing event

    Strengthens the thesis

    AI-linked equities experienced a sharp synchronized decline, with major U.S. technology and semiconductor indexes falling heavily and losses spreading to Japan and South Korea. The scale and international reach of the move show that investor reassessment is no longer confined to a narrow group of companies.

    Why it matters: A broad repricing raises the probability that elevated AI valuations and spending assumptions are vulnerable to changing expectations, particularly given the sector’s high borrowing and investment levels.

    Potential beneficiaries

    • Investors holding cash or defensive assets
    • Potential buyers of AI-related assets at lower valuations

    Potential losers

    • AI-linked equities
    • Semiconductor companies
    • Investors exposed to leveraged AI infrastructure spending

    This is direct evidence of heightened market sensitivity and repricing risk, but not by itself proof that the underlying AI investment cycle has ended.

  21. Demand and financing assumptions face simultaneous pressure

    Strengthens the thesis

    Disappointing Intel results challenged the assumption that semiconductor demand from AI is insulated from normal industry cycles. At the same time, Firmus abandoned a planned public listing and turned to private funding, indicating that at least one AI infrastructure developer could not proceed with its intended public-market financing.

    Why it matters: Cyclical demand concerns emerging alongside a retreat from planned equity-market financing increase the risk that projects dependent on continued growth and accessible capital will be reassessed or delayed.

    Potential beneficiaries

    • Private financiers able to negotiate new funding terms
    • Companies with strong balance sheets and less dependence on external capital

    Potential losers

    • AI infrastructure developers reliant on public listings
    • Semiconductor suppliers facing weaker-than-expected demand
    • Investors relying on uninterrupted AI demand growth

    The combination strengthens the fragile-financing scenario, although the evidence covers individual developments rather than demonstrating a sector-wide funding freeze or demand collapse.

  22. Policy uncertainty is raising doubts about future data-center returns

    Strengthens the thesis

    AI and data-center stocks declined after a Democratic midterm sweep increased concerns about tougher scrutiny and slower expansion. UBS estimated that about 30% of expected U.S. data-center capacity additions in 2028 could be exposed to potential construction pauses.

    Why it matters: Potential delays or cancellations could leave developers with committed capital and reduced utilization prospects, increasing the risk that planned investment fails to generate expected returns.

    Potential beneficiaries

    • Regulators
    • Communities seeking tighter oversight of data-center construction
    • Existing data centers facing less new competition

    Potential losers

    • AI and data-center developers
    • Data-center equipment suppliers
    • Investors in expansion-dependent companies

    Strengthens the downside-risk case by adding policy and execution risk to already large future capacity plans.

  23. AI borrowing has not yet displaced Treasury demand

    Weakens the thesis

    The Treasury Borrowing Advisory Committee said AI-related corporate borrowing had not crowded out demand for long-term Treasuries. It did note that technology investment may be influencing interest rates through stronger growth and could affect Treasury issuance conditions and hedging flows at the margin.

    Why it matters: The finding reduces evidence of an immediate, broad fixed-income funding displacement caused by AI debt issuance, although AI investment is still large enough to have some macro-financial effects.

    Potential beneficiaries

    • U.S. Treasury market
    • Investors in long-term Treasuries
    • AI companies benefiting from continued credit availability

    Potential losers

    • Investors expecting an immediate AI-driven Treasury-market disruption

    Moderately weakens the near-term systemic-financing concern, but leaves open the possibility that sustained AI investment could affect rates and market conditions.

  24. Big-tech guarantees are becoming part of AI infrastructure financing

    Strengthens the thesis

    Google provided a credit backstop for TeraWulf’s data-center project with Anthropic, alongside an expanded construction agreement with Fluor. The arrangement adds to evidence that smaller AI infrastructure developers may rely on financial support from major technology companies to secure funding.

    Why it matters: Credit guarantees can facilitate continued buildout, but they also indicate that project financing may depend on the balance sheets and strategic commitments of large technology firms rather than standalone project economics.

    Potential beneficiaries

    • Google
    • Anthropic
    • TeraWulf
    • AI data-center developers

    Potential losers

    • Smaller developers without major technology backers
    • Creditors exposed to project-finance concentration

    Strengthens the financing-fragility component of the scenario, while not by itself demonstrating that demand or projects are uneconomic.

  25. Unrecognized AI obligations broaden the financing overhang

    Strengthens the thesis

    Goldman Sachs estimates that five major technology companies have accumulated approximately $3.2 trillion in off-balance-sheet AI-related commitments, including $1.7 trillion in leases. More than 75% of those leases have not yet commenced, indicating that a substantial portion of the spending burden remains ahead.

    Why it matters: The commitments suggest that reported balance sheets may understate future AI-related obligations and that investment exposure could rise materially as leases begin and commitments are executed.

    Potential beneficiaries

    • AI infrastructure providers
    • Technology companies with strong access to capital

    Potential losers

    • Investors underestimating future technology-sector obligations
    • Technology companies facing weaker financing conditions

    Strongens the case for elevated financing and execution risk, though the commitments alone do not establish that spending will fail to generate adequate returns.

  26. AI expansion is increasingly tied to continued capital raising

    Strengthens the thesis

    Financial leaders expect technology companies to raise more capital to meet AI-industry demand, even as combined AI-related spending is projected to exceed $730 billion in 2026. They also warned that the sector remains exposed to a possible correction.

    Why it matters: The combination of very large planned spending, additional fundraising needs, and an explicit correction warning increases sensitivity to tighter capital markets or a reassessment of AI growth expectations.

    Potential beneficiaries

    • Capital providers
    • Financial institutions arranging technology-sector funding

    Potential losers

    • Highly leveraged or capital-dependent technology companies
    • Investors exposed to an AI-sector correction

    Provides direct market-level confirmation that financing needs and correction risk are becoming central concerns around the AI investment cycle.

  27. AI-related borrowing is beginning to create broader credit-market pressure

    Strengthens the thesis

    Rising borrowing by U.S. technology companies is increasing corporate-debt supply in European bond markets, prompting investors to demand pricing concessions that could raise borrowing costs for other companies and governments.

    Why it matters: This indicates that AI financing needs may be affecting funding conditions beyond the technology sector, increasing the potential for tighter credit and higher financing costs if borrowing accelerates.

    Potential beneficiaries

    • Investors able to demand higher yields
    • Credit analysts monitoring corporate leverage

    Potential losers

    • Other corporate borrowers
    • Governments refinancing in European bond markets
    • Highly leveraged AI-related companies

    A concrete sign of financing spillover, though the evidence does not establish a broad credit-market disruption.

  28. Demand remains strong but is accompanied by capacity strain and a less predictable spending path

    Weakens the thesis

    Compute demand continues to exceed supply, and strong AI-equipment demand is supporting U.S. factory growth. At the same time, supply-chain strain and higher prices are emerging, while Goldman Sachs’s CEO warned that infrastructure spending may proceed through stops, starts, and recalibrations rather than a straight line.

    Why it matters: The evidence supports continued near-term demand while also identifying operational bottlenecks and spending volatility that could make future growth less smooth and increase the risk of abrupt adjustments.

    Potential beneficiaries

    • AI-equipment manufacturers
    • Suppliers serving data-center and compute demand

    Potential losers

    • AI infrastructure developers facing higher input costs
    • Customers affected by constrained compute supply
    • Investors assuming uninterrupted infrastructure growth

    Demand is currently a stabilizing factor, but capacity constraints and the prospect of uneven spending make the expansion more vulnerable to pauses or repricing.

  29. Funding capacity is falling short of the capital implied by data-center investment plans

    Strengthens the thesis

    OpenAI reduced the target for its latest funding round from $30 billion to $20 billion, while Gartner estimated that data-center owners would need $8.5 trillion in cumulative revenue by the end of the decade to justify their investments, compared with projected AI revenue of $3.6 trillion.

    Why it matters: The combination of a smaller major funding target and a large gap between required and projected revenue highlights the risk that infrastructure investment assumptions may depend on capital and monetization that are not yet secured.

    Potential beneficiaries

    • Investors demanding stronger investment discipline
    • Established firms with lower infrastructure funding needs

    Potential losers

    • Data-center owners reliant on sustained capital inflows
    • Companies seeking very large AI financing rounds
    • Creditors exposed to infrastructure investment assumptions

    A strong warning about funding and revenue sufficiency, but the estimates are projections rather than proof that investments will fail.

  30. Alphabet expands bond financing for AI spending

    Strengthens the thesis

    Alphabet launched a seven-part euro bond sale with maturities extending to 19 years, explicitly linking the proceeds to artificial-intelligence spending. This adds evidence that major platform companies are using longer-term debt markets to fund AI investment.

    Why it matters: Debt-funded expansion increases the importance of financing conditions and future cash flows to the sustainability of AI capital spending.

    Potential beneficiaries

    • Alphabet
    • Euro bond investors seeking relatively high-quality corporate debt

    Potential losers

    • Alphabet if AI returns or financing conditions deteriorate
    • Investors exposed to long-duration AI infrastructure spending

    A concrete new example of hyperscaler AI investment being supported by bond issuance; moderately strengthens the financing-fragility concern.

  31. Hyperscaler AI capital spending approaches economy-wide significance

    Strengthens the thesis

    The five largest AI-data-center companies are projected to spend about $900 billion on AI in 2026, while Meta, Google, and Amazon are already absorbing substantial investment costs in their earnings.

    Why it matters: The scale of planned spending raises the potential macroeconomic impact of any slowdown in AI demand, monetization, or infrastructure utilization.

    Potential beneficiaries

    • AI chip and data-center suppliers
    • Construction, power, and infrastructure providers

    Potential losers

    • Hyperscalers if spending fails to generate proportional revenue
    • Economies and suppliers exposed to a pullback in AI capital expenditure

    The size and earnings burden of the investment cycle increase the consequences of a disappointing AI spending payoff, strengthening the downside-risk case.

  32. AI infrastructure financing is increasingly opaque and concentrated

    Strengthens the thesis

    BIS-related analysis indicates that AI data centers are increasingly financed through long-term debt and off-balance-sheet vehicles, with exposures concentrated among a small group of large companies. The five largest U.S. hyperscalers were expected to exceed $800 billion in 2026 capital spending.

    Why it matters: Hidden or difficult-to-map obligations can obscure leverage and make stress transmission harder for investors to assess, while concentration increases vulnerability to problems at a few major borrowers or operators.

    Potential beneficiaries

    • Large hyperscalers able to access structured financing
    • Lenders and arrangers earning fees from data-center financing

    Potential losers

    • Investors with incomplete visibility into AI-related liabilities
    • Lenders and counterparties if data-center cash flows weaken

    This provides direct evidence of opacity, leverage, and concentration in AI infrastructure finance, materially strengthening the risk assessment.

  33. AI infrastructure spending is intensifying alongside severe server-capacity pressure

    Strengthens the thesis

    Intel is reallocating production from consumer products to server CPUs in response to what it describes as unprecedented AI demand, while planning about $20 billion in capital expenditure in 2026 and substantially more in 2027.

    Why it matters: The scale-up confirms powerful near-term demand, but it also increases reliance on sustained AI infrastructure spending and raises the consequences if demand or funding conditions later deteriorate.

    Potential beneficiaries

    • Server CPU suppliers
    • Data-center operators
    • Companies providing AI infrastructure

    Potential losers

    • Consumer-product segments competing for production capacity
    • Investors exposed to infrastructure spending that depends on continued demand growth

    The evidence strengthens the demand side of the scenario while also showing growing capital intensity and concentration of capacity around AI-related spending.

  34. Political risk is becoming a material constraint on planned AI infrastructure

    Weakens the thesis

    State-level political and regulatory decisions could affect rules governing 67.4 gigawatts of planned data centers, representing approximately $2.5 trillion in proposed AI infrastructure investment.

    Why it matters: A large portion of anticipated AI infrastructure deployment is exposed to election outcomes and regulatory changes, increasing uncertainty around project timing, approvals, and investment realization.

    Potential beneficiaries

    • State governments with leverage over data-center regulation
    • Existing data centers and operators with approved projects

    Potential losers

    • Developers of unapproved or politically contentious data centers
    • Suppliers dependent on the full buildout of planned capacity

    This adds a significant policy and execution risk to the infrastructure investment pipeline, but the evidence does not show that projects have been canceled or demand has weakened.

  35. AI usage gains are showing signs of weaker monetization

    Strengthens the thesis

    Google’s AI Overviews increased search usage but generated lower margins and cannibalized some core advertising activity. Google’s market value fell by more than $500 billion after its earnings report, indicating that strong AI engagement did not prevent a severe investor repricing when profitability concerns emerged.

    Why it matters: This provides a concrete example of AI demand translating imperfectly into earnings and exposing valuation sensitivity to margin pressure.

    Potential beneficiaries

    • Investors seeking evidence of earnings discipline
    • Competing search and advertising platforms if monetization proves stronger

    Potential losers

    • Google shareholders
    • AI businesses whose valuations depend on rapid high-margin monetization

    A meaningful warning sign for the scenario: demand can remain strong while profitability and market confidence deteriorate.

  36. Anthropic’s growth is paired with exceptionally large contractual and financing commitments

    Strengthens the thesis

    Anthropic reported $11.6 billion in second-quarter revenue, alongside $430.6 billion of minimum payments under four take-or-pay computing contracts and $110.8 billion of net cash from financing activities in the first half of 2026.

    Why it matters: The combination highlights how substantial future infrastructure obligations and financing flows can sit alongside rapid revenue growth, increasing exposure to funding conditions and sustained demand for computing capacity.

    Potential beneficiaries

    • Cloud and computing providers with long-term contracted demand
    • Financiers participating in AI infrastructure funding

    Potential losers

    • AI companies exposed to large fixed or minimum-payment commitments
    • Investors if revenue growth fails to support contracted obligations

    The figures materially reinforce concern about fragile financing and the scale of obligations embedded in AI expansion, without by themselves proving that the commitments cannot be met.

  37. A major AI-spending repricing signal has emerged in public markets

    Strengthens the thesis

    Alphabet's loss of more than $500 billion in market value during a sharp one-day decline renewed investor doubts about whether the spending scale of major technology companies can be justified.

    Why it matters: A sharp valuation reaction at a major AI investor shows that confidence in the economic payoff from AI spending can shift rapidly, raising the risk of broader repricing if earnings or infrastructure demand disappoint.

    Potential beneficiaries

    • Investors reducing AI-related exposure before further declines
    • Defensive or non-AI sectors during a rotation

    Potential losers

    • Alphabet shareholders
    • AI-related equities and companies dependent on sustained technology-sector valuations

    Provides market evidence of heightened sensitivity to AI-spending concerns, but does not by itself establish that a broad bubble has burst.

  38. AI infrastructure financing is becoming more credit-dependent and risk-transfer oriented

    Strengthens the thesis

    Reported plans for Nvidia to support debt financing for an Anthropic-linked data center, alongside Barclays exploring insurance protection for more than $10 billion of AI-infrastructure debt, indicate that major projects increasingly rely on credit enhancement and risk redistribution. Both arrangements remain preliminary, but they suggest financing structures are becoming more complex and dependent on third-party balance sheets.

    Why it matters: This increases the potential for financing stress to propagate across chipmakers, lenders, insurers, and AI infrastructure projects if expected demand or cash flows weaken.

    Potential beneficiaries

    • Insurers offering infrastructure-debt protection
    • Lenders able to transfer or hedge AI-infrastructure credit exposure
    • AI infrastructure developers receiving credit support

    Potential losers

    • Unhedged lenders
    • Investors exposed to AI infrastructure without transparent credit-risk allocation
    • Projects unable to secure credit enhancement

    Strengthens the scenario's concern about fragile financing, though the evidence describes exploratory or unsigned arrangements rather than completed transactions.

  39. Hyperscalers are relying on unusually large external funding volumes

    Strengthens the thesis

    Alphabet, Amazon, Meta, Microsoft, and Oracle collectively raised $326 billion through stock and bond issuance during 2026 to finance AI expansion.

    Why it matters: The scale of issuance shows that even the largest platforms are supplementing internal resources with capital-market funding, increasing their exposure to market access and financing costs.

    Potential beneficiaries

    • Banks and investors underwriting or purchasing the new securities
    • Companies providing capital to AI expansion

    Potential losers

    • Issuers if financing costs rise or markets become less receptive
    • Equity and debt investors exposed to increased issuance

    The amount is substantial and reinforces the financing dimension of the scenario, but the evidence does not show that the capital cannot be serviced.

  40. Credit markets are repricing AI infrastructure risk

    Strengthens the thesis

    Investors are demanding sharply higher yields from AI-linked borrowers and data-center financings, with CoreWeave’s unsecured bonds reaching about 16% and related infrastructure bonds also weakening.

    Why it matters: This indicates that financing conditions are becoming differentiated rather than uniformly supportive, raising the cost of capital for companies most dependent on AI infrastructure funding.

    Potential beneficiaries

    • Higher-quality borrowers with strong cash generation
    • Investors able to demand higher credit spreads

    Potential losers

    • CoreWeave
    • Highly leveraged data-center operators
    • Holders of AI-linked corporate debt

    The move is a concrete sign of growing credit-market selectivity, though it does not by itself establish broad-based financial stress.

  41. AI spending signals are diverging across major platforms

    Strengthens the thesis

    Microsoft reported strong cloud growth, demand exceeding supply, and a large commercial backlog, while Alphabet missed revenue and profit expectations, posted negative free cash flow of $9.07 billion, and warned of higher capital spending and near-term margin pressure.

    Why it matters: The divergence suggests that strong sector demand is not translating evenly into cash flow or profitability, increasing the importance of company-specific execution and financing capacity.

    Potential beneficiaries

    • Cloud providers with established commercial demand and stronger cash generation
    • Investors favoring profitable AI exposure

    Potential losers

    • Companies facing rising AI costs without equivalent revenue conversion
    • Investors exposed to weaker cash-flow outcomes

    The evidence is mixed: Microsoft supports the durability of demand, while Alphabet highlights the cash-flow and margin risks of meeting that demand.

  42. AI infrastructure demand reaches a three-decade high-growth pace

    Weakens the thesis

    Demand for AI infrastructure is reportedly expanding at its fastest pace in 30 years, providing unusually strong near-term support for infrastructure-related revenue and investment expectations.

    Why it matters: The evidence reinforces the fundamental-demand side of the scenario and reduces the case that current AI investment is driven solely by speculative enthusiasm. It does not, however, resolve the scenario’s concerns about leverage, private credit, or future funding commitments.

    Potential beneficiaries

    • AI infrastructure manufacturers
    • Cloud and data-center operators
    • AI hardware suppliers

    Potential losers

    • Investors positioned for an imminent collapse in AI infrastructure demand

    This is a meaningful counterweight to bubble concerns, but a single demand indicator cannot establish that financing structures are sustainable or that valuations are justified.

  43. Anthropic’s IPO filing highlights severe profitability and accounting risk

    Weakens the thesis

    Anthropic reported more than $16 billion in first-half 2026 revenue but no profit, alongside an $80.7 billion net loss that included a $72.4 billion one-time charge tied to financing agreements. The filing could make the company a major public-market test of whether exceptional AI revenue growth can support credible earnings and valuation.

    Why it matters: A high-profile IPO with large reported losses and complex financing adjustments could expose investors to substantial uncertainty about underlying economics and provide a sharp repricing catalyst if public-market expectations diverge from sustainable profitability.

    Potential beneficiaries

    • Investors seeking greater disclosure through public listings
    • Competing AI companies with stronger underlying margins

    Potential losers

    • Anthropic shareholders if the IPO is discounted
    • Investors relying on revenue growth as a proxy for profitability

    Strengthens the case that AI valuation risk is tied not only to demand, but also to earnings quality and financing-related accounting complexity.

  44. AI infrastructure financing is becoming more layered and collateral-dependent

    Weakens the thesis

    A consortium of major banks and Blackstone announced a $63 billion special-purpose vehicle to finance Broadcom accelerators and networking equipment leased to Anthropic. The structure combines first- and second-lien financing with Broadcom residual-value support, placing greater emphasis on equipment collateral and future lease performance.

    Why it matters: Layered financing can expand infrastructure deployment without equivalent upfront cash, but it also introduces leverage, collateral-value, and counterparty risks. A slowdown in Anthropic’s demand or equipment utilization could transmit stress across lenders, lessors, and suppliers.

    Potential beneficiaries

    • Banks and private-capital providers earning financing fees and interest
    • Equipment suppliers able to accelerate sales through leasing structures

    Potential losers

    • Lenders and investors exposed to lease repayment or residual-value risk
    • AI infrastructure owners if utilization or equipment values decline

    Shows that AI capacity growth is increasingly supported by structured financing arrangements whose stability depends on sustained demand and asset values.

    Sources

  45. Meta’s AI expansion is moving toward equity-funded, potentially circular capital formation

    Weakens the thesis

    Meta outlined a plan to raise $35 billion through a stock sale for AI investment, with the possibility of an additional $60 billion through a new share class. Reports that Nvidia, AMD, and Broadcom might purchase $5 billion of the offering raised concerns that AI suppliers could help finance the customer demand supporting their own valuations.

    Why it matters: The structure could blur the distinction between independent end-user demand and supplier-supported financing, while the scale of the proposed issuance creates dilution and increases the market’s sensitivity to future AI spending returns.

    Potential beneficiaries

    • AI hardware suppliers if the funding sustains orders
    • Meta if the capital supports productive AI investments

    Potential losers

    • Existing Meta shareholders facing potential dilution
    • Investors if supplier participation is interpreted as circular financing

    Adds a concrete example of the circular-financing concern and raises the stakes around whether AI investment is being funded by durable cash generation or by ecosystem-wide capital recycling.

  46. Investor scrutiny is shifting toward near-term earnings proof as AI borrowing needs rise

    Weakens the thesis

    Investors are moving from AI-related equities into government bonds, demanding higher compensation from neocloud borrowers and focusing on whether major hyperscalers can justify unprecedented spending. Hyperscaler debt issuance is projected at $420 billion in 2027, while major technology-company capital expenditures are expected to reach $1 trillion and many leading spenders may remain in negative free cash flow through 2027.

    Why it matters: The combination of higher funding needs, expensive credit and delayed cash-flow recovery makes upcoming earnings and financing conditions more immediate repricing catalysts for the AI buildout.

    Potential beneficiaries

    • Government bonds
    • Investors favoring cash-generative technology companies
    • Credit investors able to demand higher risk premia

    Potential losers

    • Highly leveraged AI and neocloud companies
    • Hyperscalers reliant on debt-funded capital expenditure
    • AI-related equities vulnerable to disappointing earnings or guidance

    The evidence indicates a broadening market test of AI economics, though it does not establish that investment returns will ultimately fail.

  47. Utilities are filtering out a large share of planned AI data-center demand

    Weakens the thesis

    Utilities and grid operators are requiring collateral, longer contracts and higher fees before granting power connections. About half of planned U.S. data-center buildouts are reportedly delayed, stalled or canceled, while Wood Mackenzie estimates utilities may commit to only about 28% of requested connections.

    Why it matters: The data suggests that announced AI infrastructure demand is materially less firm than headline construction plans imply, increasing the risk of overbuilding, delayed revenue realization and stranded financing commitments.

    Potential beneficiaries

    • Utilities imposing stronger credit protections
    • Existing data centers with secured power access
    • Developers with funded and contractually committed projects

    Potential losers

    • Speculative data-center developers
    • Equipment and infrastructure suppliers dependent on canceled projects
    • Lenders financing projects without firm power commitments

    A significant weakening signal because power access and project credibility are emerging as binding constraints rather than merely longer-term infrastructure issues.

  48. AI infrastructure financing is becoming a live private-credit stress test

    Weakens the thesis

    CoreWeave’s bonds have fallen to distressed levels while its roughly $50 billion debt load places private-credit exposure to AI infrastructure under direct scrutiny. Private credit could ultimately be asked to fund about $1 trillion of AI-compute construction, creating concentration risk if highly leveraged operators struggle to refinance or generate expected returns.

    Why it matters: This converts financing risk from a forward-looking concern into an observable credit event with potential spillovers across private-credit portfolios and AI infrastructure lenders.

    Potential beneficiaries

    • Government bonds
    • Credit insurers
    • Lenders with conservative exposure to AI infrastructure

    Potential losers

    • CoreWeave and other highly leveraged neoclouds
    • Private-credit funds concentrated in AI infrastructure
    • AI infrastructure borrowers facing higher spreads

    A concrete weakening development, supported by distressed bond pricing and estimates of substantial future private-credit exposure.

  49. AI growth is being supported by increasingly fragile financing

    Strengthens the thesis

    The largest firms’ capital expenditure is outpacing cash flows, increasing reliance on debt and private credit. Across 14 publicly disclosed AI transactions since 2023, the IMF identified $300 billion in equity investments, $650 billion in debt, and $1.4 trillion in future revenue commitments.

    Why it matters: Debt and contractual commitments can magnify losses if AI revenue or profitability disappoints, turning an equity-market correction into a broader financing shock.

    Potential beneficiaries

    • Creditors with strong underwriting and collateral protections
    • Companies with substantial internal cash generation

    Potential losers

    • Debt-dependent AI projects
    • Private-credit lenders exposed to weaker AI cash flows

    The scale and structure of financing materially increase the consequences of an AI investment slowdown.

  50. AI investment and valuations show clear signs of overheating risk

    Strengthens the thesis

    The BIS described AI-related equity valuations as stretched and warned that slower investment or earnings could cause sharp repricing. It also estimated that baseline investment could exceed the socially efficient level by about 50%.

    Why it matters: These assessments indicate that prices and capital deployment may be assuming unusually strong future growth, increasing downside if expectations are revised.

    Potential beneficiaries

    • Investors maintaining liquidity
    • Businesses with disciplined AI capital budgets

    Potential losers

    • Highly valued AI-related equities
    • Highly leveraged AI infrastructure projects

    This is the strongest direct evidence supporting an AI-bubble concern, although it does not establish that prices have already entered a full bubble.

  51. Exceptional AI demand provides fundamental support, but commitments may outrun realizable demand

    Weakens the thesis

    Quarterly revenue rose 106% year over year to $96.2 billion, while data-center revenue rose 117% to $89.0 billion. However, supply and capacity commitments had risen to $279 billion, and customers were warned that infrastructure or funding constraints could delay purchases.

    Why it matters: Very strong realized growth argues against viewing the entire AI boom as purely speculative, but large forward commitments and possible customer delays show that expectations remain vulnerable to financing and execution constraints.

    Potential beneficiaries

    • AI infrastructure providers with real current revenue growth
    • Suppliers positioned to meet sustained demand

    Potential losers

    • Projects dependent on uninterrupted customer financing
    • Companies carrying excess capacity if orders are delayed

    This is important counterevidence to a simple bubble thesis, but it does not eliminate the risk that future capacity and valuations exceed sustainable demand.

Thesis assessment

ComponentDirectionNote
Valuation riskelevatedAI-related equity valuations are described as stretched, with sharp repricing possible if investment or earnings slow.
Fundamental demandstrengtheningReported revenue and data-center growth remain exceptionally strong.
Financing fragilityelevatedCapital expenditure is outpacing cash flow, while debt, private credit, and large future commitments are expanding.
Contagion riskelevatedCircular financing and interconnected relationships could spread disappointing returns across the AI ecosystem.
Bubble diagnosisno decisive new evidenceThe evidence supports elevated risk of an AI bubble and correction, but strong current revenue growth means it does not prove that a full bubble has already formed.

Bottom line

There is credible evidence of an AI investment and valuation bubble risk, especially in the financing structure and scale of future commitments. However, the evidence does not support concluding that AI demand is fictitious or that a collapse is inevitable. The most defensible assessment is an economically valuable boom with increasingly bubble-like financial characteristics.

Favor diversification, balance-sheet strength, and exposure that is less dependent on uninterrupted AI capital spending. Treat highly valued, debt-dependent, and interconnected AI infrastructure exposures as vulnerable to a sharp repricing if earnings, funding, or customer purchases slow.