Equity investors have spent the past 18 months watching the same AI leopard change its spots. The spots first looked like unemployment risk, then the so-called ‘SaaS Apocalypse,’ then overvaluation, with AI stocks trading on PERs in the 30–40 range. Now the latest spot is capex – whether hyperscalers can finance the infrastructure AI requires. Each bout of fear caused a 5–10% correction, and each time, AI-giddy investors have bought the dip, pushing equity indexes back to record highs – 27 for the S&P 500 alone in 2026.
Cybersecurity may be the next spot to appear. Capex remains the spot currently spreading through equity and credit pricing. On 27 August, OpenAI, Anthropic, Amazon Web Services, Microsoft, and more than 100 other companies warned in an open letter that organisations may only have months to prepare for a sharp increase in AI-enabled cyberattacks. I don’t know enough about the world’s state of cybersecurity readiness to opine on that topic. I do, however, see evidence in both equity and credit markets that the capex fear has more staying power than the previous AI scares.
During the Q2 earnings season, investors focused on hyperscaler cash-flow statements. Amazon’s trailing operating cash flow rose 33% to $161.4bn, while free cash flow swung from an $18.2bn inflow to a $7.6bn outflow as property and equipment purchases accelerated. Alphabet generated $39.1bn of operating cash in the second quarter and spent $44.9bn on capital assets, producing negative quarterly free cash flow of $5.9bn. Meta generated $31.9bn from operations and reported just $784m of free cash flow after $31.1bn of capital expenditure and finance-lease repayments. Microsoft remains the stronger counterweight, generating $182.9bn from operations against $115.9bn of annual capital expenditure.
Oracle sits on the other side of the ledger. S&P cut its rating to BBB-, citing rising AI capital requirements, weaker cash flow, leverage expected above 4x, and high customer concentration. Oracle forecasts a fiscal-2027 free-operating-cash-flow deficit near $42bn and includes $260bn of additional lease commitments in adjusted debt. Rating agencies do not need to declare the technology unpromising. They only need to suggest that leases, power agreements, guarantees and customer exposure belong on the balance sheet instead of off it. That alone would increase the financing cost of the next data centre before changing a single line of current revenue.
The credit market took notice. Spreads on high-quality hyperscaler debt over Treasuries have already widened by 20–40 basis points over the past three months, and by 100 basis points for Oracle and 40 basis points for SpaceX. At the same time, bond investors have begun to voice their discontent with the lack of fiscal responsibility across major Western economies. Long bond yields have been rising across the developed world, reaching decade highs in the process and forcing the US Treasury Secretary to come to their defence. What do wider hyperscaler credit spreads, rising long sovereign yields and resilient AI-led equity enthusiasm have in common – and should investors be concerned?
Higher sovereign demand for funding large fiscal deficits is now colliding with unprecedented hyperscaler demand for capital to build the next generation of AI infrastructure. The current AI investment thesis is simple: the nation that leads in AI will dominate the 21st century, and AI leadership runs through data centres. Funding government deficits moves the sovereign curve up. Funding data-centre capex moves the spread curve up. Everyone else pays the clearing price.
Why is the recent rise in yields and spreads likely to continue? Because governments have little choice but to seek funding at whatever yield clears the auction. Hyperscalers, meanwhile, are forecasting AI revenues and margins high enough to make them unusually tolerant of higher funding costs. That tolerance raises the reference cost of capital for the rest of the economy. Spreads will increasingly be marked against the hyperscaler curve.
The problem does not stop at the credit desk. Credit repricing reaches equities through more than the discount rate. Wider spreads reduce the present value of long-duration earnings, while higher interest, lease and depreciation charges alter the earnings path itself. Lower free cash flow also leaves less room for buybacks, dividends or a graceful retreat if returns disappoint. The transmission is not mechanically bearish. Slower capital spending could improve hyperscaler cash flow and support their bonds, while hurting semiconductor, infrastructure and power-related equities. A repricing caused by weaker counterparties or scarcer funding, however, would challenge credit and equity together, precisely when investors expect one allocation to diversify the other.
That transmission mechanism creates a diversification problem for institutional investors. Vanguard calculates that five hyperscalers supplied about 11% of US investment-grade bond issuance through July, with much of the borrowing long-dated; Guggenheim estimates technology and hyperscaler bonds represent roughly 21% of the 10-year-plus investment-grade index. Pension funds may therefore own the same AI cycle through global equities, corporate credit, private infrastructure, real estate, and insurer-backed assets, each held by a different team and labelled as diversification. The label does not change the exposure. It merely makes Total Portfolio Risk easier to underestimate.
Past infrastructure booms offer perspective, not a script. Railways, electrification, and telecommunications were all debt-financed. Each created lasting economic value. Each also distributed that value unevenly among pioneers, creditors and later entrants. AI may prove equally transformative. Transformation, however, does not waive the timing mismatch between cash spent today and revenue earned later. Pension funds can believe in the technology while declining to underwrite every structure built in its name. The leopard is still AI. The spot that matters now is financing. The AI trade does not have to fail for its financing assumptions to do so.
Olivier d’Assier is a senior director, investment decision research, at SimCorp
