The Thesis
AI risk is no longer contained within Big Tech. The buildout is becoming more interconnected, concentrated and dependent on external finance. As capital moves through an increasingly connected web of companies and creditors, a shock in one part of the AI ecosystem can travel much further than before.
Four mechanisms are widening the transmission channel.
Concentration. AI exposure is clustering in the companies that dominate global markets, making any repricing harder to isolate.
Circularity. Capital, infrastructure spending and customer demand increasingly reinforce one another. Any weakness in one part of the system can feed back into the rest.
Credit expansion. More of the buildout is moving from corporate cash flows into bond markets, private credit and institutional portfolios.
Opacity. The growing use of structured and off-balance-sheet financing obscures where exposures ultimately sit and how losses could propagate through the financial system.
The Signal
Three developments worth watching this week.
What happened. The latest hyperscaler results showed AI infrastructure spending beginning to press directly against internal cash generation. Five US companies sit at the centre of the buildout: Amazon, Alphabet, Meta, Microsoft and Oracle. Alphabet drew particular attention in Q2, when it spent $44.9 billion on capital expenditure against $39.1 billion of operating cash flow. This pushed free cash flow negative for the first time since its 2004 listing. It also raised its full-year capex guidance to $195 billion to $205 billion. Meta showed a similar squeeze, with quarterly free cash flow falling 91% year on year to just $784 million. Following the results, Goldman Sachs raised its estimate for global AI investment in 2026 to more than $1 trillion, underscoring how quickly the capex race has expanded beyond even the largest corporate balance sheets.
Why it matters. The shift is in scale, not solvency. Alphabet and Meta still hold substantial cash reserves, but AI infrastructure requirements are growing quickly enough that even the world’s most cash-generative technology companies are increasingly combining operating cash flow with debt, equity and leases. That financing is now large enough to alter the structure of credit markets themselves. Roughly two-thirds of the value of $10 billion-plus US investment-grade bond deals across 2025 and 2026 has come from Big Tech. The financing question is moving from whether hyperscalers can afford AI to how much of the investment they choose to place outside their own cash flows.
Second-order effect. The constraint can move from technology to capital. As the buildout becomes more difficult and expensive to finance, marginal data-centre and compute projects become harder to justify. This could slow the infrastructure expansion on which future AI adoption and productivity gains depend. Governments pursuing domestic AI capacity may then face pressure to support investment that markets are becoming less willing to fund.
What happened. The surge in hyperscaler borrowing is beginning to test investor appetite. Apollo finds that the cover ratio on new hyperscaler bonds, the value of investor orders relative to the amount issued, fell from nearly 5x in February to below 2x by July 2026. The same pressure is visible in secondary markets: a Reuters analysis of LSEG data found that 78 of 91 hyperscaler bonds issued in 2026 were trading at higher yields by 28 July than at issuance, with a median increase of about 22 basis points. That does not mean demand has disappeared, but it suggests increasingly large deals may require wider spreads to clear. The shift is already visible in individual projects: Meta’s latest $12 billion data-centre financing came at a higher borrowing cost than a comparable deal completed less than a year earlier.
Why it matters. Hyperscalers have so far benefited from exceptional credit quality and deep investor demand. Falling cover ratios test how far that advantage can stretch as issuance accelerates. In Bank of America’s August fund manager survey, 38% of respondents identified hyperscaler capex as the most likely source of a systemic credit event, ranking it first for a second consecutive month. The concern is not that the buildout is about to stop, but that capital commitments could continue rising even as the market becomes less willing to finance them on the same terms.
Second-order effect. As AI exposure migrates into bonds, private credit, bank balance sheets and institutional portfolios, the risk no longer sits where the investment decision was made. A correction could reveal that institutions holding different instruments are ultimately exposed to the same underlying AI bet.
What happened. The Bank of England’s July Financial Stability Report brought the AI investment cycle explicitly into its assessment of financial stability. The Bank warned that risks from the infrastructure investment channel are “growing rapidly” as AI companies turn to external finance, particularly debt. It also modelled how a reassessment of AI productivity and profitability could travel beyond technology equities through credit markets and into the wider economy. Similar concerns had already emerged in the US. Months earlier, senators urged the Financial Stability Oversight Council to investigate AI-related debt, citing estimates of a $1.6 trillion financing gap for data-centre investment by 2028. They noted that $800 billion of this was projected to be filled by private credit. The intervention shows that concern over the financial risks of the buildout is increasingly international.
Why it matters. The Bank of England’s concern is not that AI debt is already large enough to threaten financial stability. Outstanding exposures remain relatively modest and much of the new issuance comes from highly rated borrowers. The concern is the direction of travel: investment forecasts keep rising while financing is spreading through an increasingly interconnected credit ecosystem. The Financial Policy Committee also noted that AI-related equities had continued to rise on strong earnings, but that “on some metrics, valuations have also become more stretched”. That leaves the financial system more sensitive to any reassessment of expected AI productivity or profitability.
Second-order effect. AI risk may increasingly need to feature in bank stress tests, government fiscal planning and corporate risk models. A major correction could hit credit conditions and tax revenues at the same time. That would make AI assumptions relevant far beyond the technology sector.
The Metric
What it measures. The modelled shortfall in UK output at the trough of the Bank of England’s AI correction scenario, and the split of that shortfall by the channel that carries it into the real economy.
Why it matters now. Credit spreads carry more of the damage than equity markets do. The exposure that matters most is therefore the one running through debt markets rather than the one visible in technology share prices, which is the same shift the first two signals describe. Despite rising debt-to-GDP ratios, the Bank’s scenario finds that US Treasury and UK gilt markets continue to function normally. The larger financial-stability risk emerges only if the shock begins to impair the sovereign markets that underpin global liquidity and collateral.
Source. Bank of England, Financial Stability Report, July 2026.
The Playbook
Four moves institutions can take as AI financing expands.
Do not stop at direct holdings in hyperscaler debt or equity. Trace exposure through private credit, data-centre financing, structured vehicles, suppliers and funds whose returns depend on continued AI investment. The important question is not which instruments an institution owns, but how many ultimately depend on the same AI growth assumptions.
Model a scenario in which AI revenues disappoint, valuations fall and credit spreads widen at the same time. Banks, asset managers and large corporates should test how that shock moves through their financing and portfolios. The aim is to capture how different exposures can reinforce one another rather than assessing each in isolation.
Defaults will come late. Track bond cover ratios, spreads, private-credit terms and the share of capex funded externally. Those indicators show whether the market is becoming less willing to finance the buildout before credit losses appear.
Central banks should incorporate AI-related shocks into stress testing. Governments should test the fiscal impact of weaker equity markets, lower tax receipts and pressure on sovereign borrowing. Corporate boards should treat assumptions about AI adoption and financing costs as part of their wider risk planning, not just their technology strategy.
The Verification Test
“The AI buildout is becoming a financial-system exposure rather than a self-funded Big Tech investment cycle.”
Test. Track the next two quarters of hyperscaler capex, free cash flow and external financing. Do this alongside their share of investment-grade bond issuance and new-deal cover ratios. The key question is whether rising infrastructure spending continues to push financing outside company cash flows and deeper into credit markets.
Pass criteria. Capex continues to grow faster than internal cash generation, external borrowing remains elevated and AI-related debt becomes a larger presence across public and private credit markets. Investor demand also weakens or requires wider spreads as issuance grows.
Fail smell. AI revenues and cash generation catch up with spending, hyperscalers reduce their reliance on external finance and bond-market demand remains strong despite continued issuance. That would suggest the current financing surge is temporary rather than a structural shift in where AI risk sits.
The Lens — Horizon Search Institute
AI is turning expectations of future compute demand into long-lived infrastructure today. As the buildout becomes more dependent on external finance, more capital is being committed to data centres and chips before their future economic life is known. The risk is not only overinvestment, but infrastructure built against assumptions that may change faster than the assets themselves.
The financing footprint is expanding faster than the institutions responsible for overseeing it. Bonds and off-balance-sheet structures spread AI exposure across banks, investors and governments that may not think of themselves as technology investors. As that exposure grows, AI assumptions increasingly become part of financial-stability policy, and stress testing and public-finance planning become critical.
The productivity case for AI is becoming part of the credit case. Investment made today ultimately depends on adoption translating into higher revenues and output tomorrow. If those gains arrive more slowly than expected, weaker organisational returns could feed back into financing conditions, which could in turn slow the deployment on which those same productivity gains depend.
Links Worth Your Time
-
The AI debt reshaping the US bond market
The Financial Times’ running account of how hyperscaler borrowing is reshaping the investment-grade bond market, including the concentration data behind this issue’s Big Tech bond-share figure.
-
The AI investment race
Phurichai Rungcharoenkitkul models the AI buildout as a winner-take-most contest among hyperscalers, and finds that debt and circular equity ties are what turn ordinary over-investment into the kind of financial fragility that ends booms badly.
-
AI revenues are growing fast, but not fast enough
The Economist lays out how fast the spending curve is climbing industry-wide, then weighs that pace against how quickly AI revenue is actually showing up.
- Alphabet Inc. Alphabet Q2 2026 earnings release. July 22, 2026.
- Bloomberg. Big Tech earnings slam into a market in revolt over AI spending. July 26, 2026.
- Bloomberg. Alphabet posts cloud sales beat, raises capex guidance. July 22, 2026.
- Meta Platforms, Inc. Meta reports second quarter 2026 results. July 29, 2026.
- Goldman Sachs. Global AI investment is forecast to exceed $1 trillion in 2026. August 7, 2026.
- Financial Times. The AI debt reshaping the US bond market. 2026.
- Apollo. Cover ratios for hyperscaler bonds declining. July 15, 2026.
- Reuters. Hyperscaler debt binge pushes yields up as investor demand cools. July 29, 2026.
- Financial Times. Meta faces higher borrowing costs in latest $12bn data-centre financing. July 24, 2026.
- Investing.com. Investors are getting extremely bullish, BofA’s August survey shows. August 18, 2026.
- Bank of England. Financial Stability Report, July 2026. July 7, 2026.
- US Senate Committee on Banking, Housing, and Urban Affairs. Letter to Secretary Bessent on AI debt and financial stability risks. January 22, 2026.
- Bank of England. Financial Policy Committee record, July 2026. July 7, 2026.
- The Economist. AI revenues are growing fast, but not fast enough. July 28, 2026.
- Rungcharoenkitkul, P. The AI investment race. BIS Working Paper No. 1367, 2026.
- Vanguard. The AI buildout comes to the bond market. August 19, 2026.