Besides the bubble, a big AI risk could be growing in the shadows

Synopsis

Investors are worried about potential financial risks linked to Big Tech's investments in AI infrastructure. The technology sector recently experienced a significant sell-off, indicating growing nervousness among shareholders. S&P Global predicts considerable expenses on data centres for major companies like Amazon and Microsoft through 2030.

Friday's sharp sell-off in US technology stocks is another indication of how nervous investors are becoming about Big Tech's enormous AI spending spree. The tech-heavy Nasdaq fell 2.86%, its biggest one-day decline since June 5, after disappointing results from Intel and Amazon intensified concerns about how quickly big investments can generate returns.

Though the biggest concern surrounding the AI boom remains whether the billions being poured into data centres, chips and computing infrastructure will generate returns quickly enough, the profitability question may not be the only financial risk building around the AI boom.

As the infrastructure buildout expands, an increasingly complex web of financing arrangements is connecting hyperscalers, chipmakers, data-centre operators and AI startups, potentially creating financial exposures that are not immediately visible to investors. These arrangements are not an immediate threat, but could become a more serious vulnerability if the AI investment boom falters.

Besides the conventional AI bubble

The recent sell-off in technology stocks has brought renewed scrutiny of Big Tech's AI spending. Investors are increasingly questioning whether the expected returns can justify the extraordinary capital expenditure.

Live Events

That remains the most immediate concern for credit-rating agencies too. In an October 19 interview with Reuters, Mark Habib, S&P Global's head of digital infrastructure, said the biggest risk to hyperscalers' credit ratings remained whether expected profits from AI investments fail to materialise. S&P estimates that six major US hyperscalers, including Amazon, Microsoft, Alphabet, Meta and Oracle, alongside SpaceX, will spend more than $7 trillion on data centres and other AI-related infrastructure between 2025 and 2030. The agency also expects these companies to experience negative free cash flow until at least next year.

But Habib flagged another concern -- the possibility that the technology giants have more AI-linked debt and financial liabilities than currently visible.

"It is safe to assume that there is more (debt) out there that we are not seeing," he told Reuters, pointing to interconnected financing structures through which hyperscalers effectively support large data-centre projects or smaller AI companies.

Habib acknowledged that S&P cannot quantify the additional exposure because it does not necessarily have visibility into transactions that are not reported or that fall outside its rating coverage. Importantly, S&P does not currently see a disconnect between market perceptions and its credit ratings that would warrant immediate rating changes.

That makes the issue less an established credit crisis than a potential vulnerability building alongside the AI investment boom.

The growing web of AI financing

The financing arrangements behind AI infrastructure are becoming more complicated than conventional corporate borrowing.

S&P Global Ratings' report from August 27 projected combined capital expenditure by the six hyperscalers to exceed $1.3 trillion by 2027. Importantly, it also highlighted the growing use of debt, leases, joint ventures, special-purpose vehicles and residual-value guarantees to finance infrastructure expansion.

The agency expects all six companies to generate negative free operating cash flow in 2026 and 2027, with recovery not projected until 2029. This does not mean they are running out of cash or cannot service their obligations. It does, however, make the financing of their investment plans increasingly important to their future credit profiles.

One example of these increasingly interconnected financing arrangements involves Google and Anthropic. Google is both an investor in the AI company and a major provider of its cloud-computing capacity. This creates a relationship in which a technology giant is not only financing an AI developer but also supplying the computing infrastructure it needs to operate.

A more complex financing structure was outlined by LSEG in an October 21 analysis. A special-purpose vehicle called Compute SPV was set up to purchase Google's custom AI chips, supplied with Broadcom, and lease them to Anthropic. The roughly $35 billion financing involved third-party debt led by Apollo and Blackstone, with Broadcom providing credit enhancement. Separately, Google has backed lease obligations for data-centre projects involving infrastructure provider Fluidstack. These are related but distinct financing arrangements, illustrating how several companies can become financially connected through the same AI infrastructure buildout.

Such structures are not inherently problematic. They allow companies to finance expensive infrastructure and spread the associated costs and risks among investors, lenders and technology providers. But they also make it harder to establish where the ultimate financial exposure lies, particularly when companies are simultaneously investors, suppliers, customers and financial backers within the same ecosystem.

Why circular financing deserves attention

The issue becomes more complicated when investments, commercial agreements and financing arrangements reinforce one another. A chipmaker may invest in an AI company that subsequently uses the money, directly or indirectly, to purchase its chips. A cloud provider may invest in a model developer that becomes a major customer for its computing services. A data-centre operator may borrow against long-term commitments from a large technology company.

These arrangements can support genuine business growth. But they also create questions about how much reported demand is ultimately backed by independent customers rather than financing relationships within the same ecosystem.

A recent October 21 analysis by LSEG's investment research team argues that AI financing is better understood as a web of relationships than a simple circular loop. It identifies self-reinforcing valuations, concentration risk and limited transparency as potential sources of instability.

The analysis also cautions against drawing direct comparisons with the 2008 financial crisis. There is not yet evidence of the same kind of synthetic replication and complex derivatives that amplified risks during that episode. A more relevant historical comparison may be the telecom infrastructure boom of the late 1990s, when heavy investment eventually contributed to overcapacity as demand failed to meet expectations.

The risks may emerge elsewhere in the financial system

The exposure is not limited to technology companies or their shareholders. Private-credit funds, infrastructure investors and banks are increasingly financing the physical infrastructure required for AI.

A Reuters analysis published on October 29 highlighted how much of this lending takes place through private-credit funds and infrastructure vehicles. Unlike publicly traded bonds, these loans are generally valued periodically using models rather than continuously through market prices. This can make emerging losses less visible during periods of market stress.

The article cited OECD estimates that AI-related transactions accounted for 34% of global private-credit deals in 2025. The Bank of England has estimated that around $700 billion of data-centre spending between 2026 and 2028 could rely on private credit.

There is also a potential mismatch between the life of the financing and the assets being financed. Data-centre buildings may be designed to last decades, but the chips and servers inside them can become technologically outdated much sooner. If the equipment loses value faster than expected, lenders could find that their collateral is worth less than anticipated.

The Bank of England flagged this issue in its July 2026 Financial Stability Report, noting that shorter asset lifecycles could complicate the servicing of longer-term debt.

A vulnerability, not yet a crisis

None of this suggests that major technology companies are facing an imminent debt problem. Their substantial cash-generating businesses and access to capital markets give them a financial cushion that smaller AI infrastructure companies do not enjoy.

Nor does the existence of interconnected financing automatically mean that reported AI demand is artificial. Much of the infrastructure being financed is supporting real and growing computing requirements.

The issue is what happens if the economics change. If AI revenues disappoint, infrastructure utilisation falls or the value of computing assets declines, financial commitments spread across the ecosystem could become more difficult to sustain. A problem initially concentrated among smaller infrastructure providers could then affect their lenders, investors and major commercial partners.

For now, S&P's central concern remains whether AI investments will generate sufficient profits. But the growing complexity of the financing behind those investments raises another question -- if the AI boom falters, will investors have a clear picture of where all the financial risks actually sit?

Can new data protection law guard against SBI-like data breaches?

KYB and KYC: Quick commerce, the Damani premium and what DMart must deliver – 3rd and concluding part

The tough business behind the walls

KYC and KYB: Inside Damani’s DMart, the retail chain for the middle class. And stock for the long term? – Part 2 of 3

Will it be a regulatory nightmare if an exchange lists on its own platform?

KYC and KYB: Inside Damani’s Avenue Supermarts that refuses to charge more – Part 1 of 3