Editor’s Note: This article draws on a detailed analytical bulletin issued by the Bank for International Settlements and authored by Iñaki Aldasoro, Sebastian Doerr and Daniel Rees, which examines how the global artificial intelligence investment boom is reshaping macroeconomic dynamics, corporate financing structures and credit markets. The bulletin explores the accelerating shift from internally generated cash flows toward debt financing, the rapid rise of private credit as a funding channel for AI infrastructure, and the growing tension between elevated equity valuations and more cautious debt market pricing. By translating this research into a business and financial perspective, the following column highlights the opportunities, risks and structural implications embedded in the financialisation of the AI expansion.
Artificial intelligence is no longer a speculative sideshow in financial markets; it has become a capital-hungry industrial project that is reshaping national investment patterns and corporate balance sheets. The BIS Bulletin makes this explicit when it states that investment related to artificial intelligence is surging, both in nominal terms and as a share of GDP, and now accounts for a substantial share of economic growth. What is unfolding is not merely a software revolution but a heavy-asset expansion driven by data centres, high-density servers, cooling systems, grid connections and dedicated power capacity. At the heart of this transformation lies a surge in capital expenditures to build the physical infrastructure required to train and operate increasingly large AI models, a reminder that digital ambition rests on very material foundations.
The United States has emerged as the central theatre of this build-out because it offers sufficiently granular data to isolate AI-related investment. By mid-2025, spending on IT manufacturing facilities and data centres, including both equipment and construction, had already reached the equivalent of about 1% of GDP. When broader IT investment such as software and other equipment is included, the total rises to roughly 5% of GDP, exceeding the previous peak recorded during the dot-com boom of 2000. The structural difference is significant. The late-1990s boom was driven largely by firms consuming technology; the current surge is driven by firms producing it, concentrating capital intensity, operational risk and financing pressure within a relatively small cluster of mega-corporations and their supply chains.
AI investment has also become a tangible driver of economic growth. Before 2022 its contribution was negligible, but since then spending on semiconductor manufacturing facilities and data centres has contributed on average around 0.4 percentage points to annual GDP growth. In recent quarters, total IT investment has accounted for almost half of GDP growth, offsetting headwinds from trade tariffs and global uncertainty. Looking ahead, the trajectory appears even steeper. Analysts forecast that annual spending on data centres alone could increase by between $100 billion and $225 billion over the next five years. If realised, this would lift data-centre investment from around 0.5% of GDP today to between 0.8% and 1.3%. The AI boom, therefore, is evolving into a macroeconomic force rather than a narrow technology cycle.
From Cash Flows to Borrowed Money
The more fragile dimension of this expansion becomes evident when corporate financing strategies are examined. The technology firms leading the AI investment wave have historically operated with comparatively low leverage, relying on strong operating cash flows to fund growth. That model is now under strain. Capital expenditures have accelerated sharply, rising both in absolute terms and as a share of revenues, while free cash flows have begun to lag behind investment needs. In several cases, cash generated from operations no longer covers the scale of infrastructure build-out required to sustain competitive advantage in AI.
Equity markets offer limited relief. Valuations in the AI sector are volatile and concentrated, issuance windows are narrow, and fresh equity issuance can be costly and dilutive for long-dated, asset-heavy projects. As a result, firms are increasingly turning to debt financing through bonds, leasing arrangements and loans. Borrowing allows companies to spread costs over time and match financing maturities with the long economic life of data-centre assets. However, AI infrastructure carries risks that often sit outside traditional bank and bond comfort zones, including construction delays, power availability constraints and tenant concentration risk. These frictions have opened the door for alternative sources of capital.
Private credit has emerged as the fastest-growing channel of external financing. Typically extended by specialised non-bank funds, private credit is characterised by bespoke covenants, faster execution and greater flexibility in renegotiation. These features make it well suited to funding large, asset-heavy AI projects with complex risk profiles. Outstanding private credit loans to AI-related sectors have surged from near zero a decade ago to more than $200 billion today. The share of AI loans in total private credit volumes has climbed from less than 1% to nearly 8%. Based on projected AI investment growth of 50% to 300%, outstanding private credit exposure to AI firms could reach between $300 billion and $600 billion by 2030.
Notably, loan characteristics in AI are not radically different from those in other sectors. Loans to AI firms are typically larger, averaging about $169 million compared with roughly $90 million elsewhere, but maturities remain similar at around five years. Rate spreads are also comparable, hovering near six percentage points, and the proportion of secured lending is broadly aligned with non-AI sectors. Exposure remains modest at the fund level, with only about one-fifth of private credit funds currently investing in AI-related borrowers and the average fund allocating roughly 5% of its portfolio to these loans. For now, systemic concentration appears limited.
Debt, Valuations and the Risk of Self-Deception
The deeper tension lies in the disconnect between debt pricing and equity optimism. The Bulletin stresses that the sustainability of the boom depends on AI firms meeting exceptionally high earnings expectations, while equity prices have surged far ahead of what debt markets appear to be pricing in. If loan spreads accurately reflect underlying risk, lenders are effectively treating AI projects as no riskier than an average private credit borrower. Equity markets, by contrast, are embedding assumptions of extraordinary future profitability. This divergence implies either that lenders are underestimating risk just as their exposures are rising, or that equity investors are overestimating the cash flows AI will ultimately generate. Neither interpretation inspires confidence.
There are additional structural vulnerabilities. Some financing arrangements may shift leverage off corporate balance sheets, creating the appearance of reduced risk while leaving economic exposure intact. Leverage does not vanish simply because it is hidden. Investor commentary already reflects concerns about the long-term collateral value of data centres underpinning large deals. In a sector characterised by rapid technological obsolescence and escalating performance requirements, today’s premium infrastructure can quickly lose strategic relevance and resale value.
From a historical perspective, the AI investment boom is not unprecedented in scale. At roughly 1% of GDP, it mirrors the size of the US shale expansion in the mid-2010s and remains smaller than the dot-com surge of the 1990s. Other economies have absorbed far larger investment waves, such as Japan’s commercial property boom in the 1980s or Australia’s mining expansion in the 2000s. Yet history also suggests that the end of investment booms often coincides with GDP slowdowns of more than one percentage point on average, and that such booms rarely translate into sustained medium-term growth acceleration. The dot-com episode, despite its modest size relative to GDP, still produced the sharpest post-boom contraction.
If a slowdown in AI investment were accompanied by a sharp equity correction, spillovers could exceed historical benchmarks. Investors have disproportionately used US equities to gain AI exposure, while opaque leverage within private credit structures could transmit stress into less transparent segments of the financial system. The macroeconomic risks may appear moderate in aggregate, but the financial system is increasingly intertwined with a narrow technological narrative that assumes uninterrupted growth, rising productivity and scalable profitability.
The AI boom, therefore, is no longer merely a story of innovation and productivity promise. It is an experiment in debt-funded industrial expansion, one that carries familiar hazards of mispriced risk, asset inflation and delayed adjustment. The physical infrastructure is being built at breathtaking speed, the financing model is tilting decisively toward leverage, and expectations embedded in equity markets remain exceptionally high. Whether this transformation becomes a durable engine of growth or a costly lesson in technological exuberance will depend not on rhetoric or valuation multiples, but on whether the promised earnings materialise fast enough to service the expanding mountain of borrowed capital.
