Financing the AI boom: from cash flows to debt [pdf]

BIS Bulletin No 120 Iñaki Aldasoro, Sebastian Doerr and Daniel Rees 7 January 2026 Financing the AI boom: from cash flows to debt BIS Bulletins are written by staff members of the Bank for International Settlements, and from time to time by other economists, and are published by the Bank. The papers are on subjects of topical interest and are technical in character. The views expressed in this publication are those of the authors and do not necessarily reflect the views of the BIS or its member central banks. The authors thank Jon Frost, Pablo Hernández de Cos, Phil Wooldridge and Liang Yu for helpful comments, Ilaria Mattei and Jose María Vidal Pastor for statistical support, and Nicola Faessler and Maja Viscek for administrative support. The editor of the BIS Bulletin series is Hyun Song Shin. This publication is available on the BIS website (www.bis.org). © Bank for International Settlements 2026. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated. ISSN: 2708-0420 (online) ISBN: 978-92-9259-919-5 (online) BIS Bulletin 1 Iñaki Aldasoro Inaki.Aldasoro@bis.org Sebastian Doerr Sebastian.Doerr@bis.org Daniel Rees Daniel.Rees@bis.org Financing the AI boom: from cash flows to debt Key takeaways • Investment related to artificial intelligence (AI) is surging – both in nominal amounts and as a share of GDP – and currently accounts for a substantial share of economic growth. • The size of anticipated investment needs will require firms to shift the source of financing from operating cash flows to debt, with private credit playing a rapidly increasing role. • While macroeconomic and financial stability risks from the AI boom appear moderate, the boom’s sustainability hinges on AI firms meeting high earnings expectations. The fact that equity prices have run far ahead of debt market pricing underscores this tension. Rapid advances in artificial intelligence (AI) appear set to reshape economies, industries and financial markets and AI firms have been a major driver of equity market developments over the past year (BIS (2025)). Yet AI-related innovations demand not only groundbreaking research but also substantial investment in infrastructure.

At the heart of this transformation lies a surge in capital expenditures to build the physical infrastructure for AI, eg data centres, and related technological infrastructure such as computer servers, networking hardware, cooling systems, grid connections and power stations. These investments are key in supporting the enormous demand for computational resources and data storage facilities to train and operate AI models. The need to finance these investments is causing a shift in sourcing the financing from cash flows to debt. Leading firms in the information technology (IT) sector have historically financed much of their investments internally out of operating cash flows. However, the scale of current and anticipated AI-related investment needs is now so vast that firms are seeking external sources of funding. They are therefore increasingly financing AI investment via debt, a shift that is not only reshaping corporate balance sheets but also raises important questions about credit standards and financial stability. This Bulletin explores the AI investment boom. It first focuses on the boom’s macroeconomic dimensions. It then examines the evolving financing landscape and highlights the interplay between the surging demand for AI infrastructure and the financial mechanisms that support it, in particular private credit. The Bulletin concludes with a discussion of the possible consequences if the current optimism regarding the future returns on AI-related capital expenditures turns out to be unfounded. The macroeconomic dimension of AI investment The rise of AI has triggered a wave of investment in the digital infrastructure needed to support its development and deployment. Much of this investment has occurred in the United States (Haag (2025)). Conveniently, this is one of the few jurisdictions with sufficiently detailed data to pinpoint AI-components 2 BIS Bulletin of investment. Accordingly, in this Bulletin we focus on AI-related investment in the United States. The trends we document are likely to exist in other economies too, albeit to a lesser extent.1 AI-related investment takes a number of forms. The most direct is expenditures on data centres, which house the specific IT infrastructure needed to train, deploy and deliver AI applications and services. Such expenditures include the construction costs of building the physical facilities, as well as spending on IT and other electrical equipment needed for their operation, including servers and networking equipment. Beyond data centres, AI-related investment can also encompass IT manufacturing facilities, which produce the specialised chips and hardware that power these systems. Finally, advances in AI may also spur broader investment in IT products, for example if AI prompts businesses to upgrade their computer hardware or purchase new software.

The macroeconomic impact of the AI investment boom has become increasingly relevant. Since 2022, AI-related investment has accounted for a rising share of US GDP (Graph 1.A). Initially, this investment reflected spending on domestic semiconductor manufacturing, spurred in part by the passage of the US Creating Helpful Incentives to Produce Semiconductors (CHIPS) and Science Act as well as by a desire to strengthen supply chain resilience and reduce reliance on foreign suppliers (blue area). More recently, the surge in demand for AI-powered applications has led to a sharp increase in data centre construction (red 1 Beyond investments in data centres outside the United States (see eg the recent report by the Central Bank of Malaysia), the AI investment boom is also supporting exports and growth, especially in Asia, given the importance of the semiconductor industry (eg semiconductor demand played a prominent role in recent growth revisions by the Bank of Korea). Demand for raw materials required to support AI infrastructures, such as copper, is also likely to be a boon for producing countries. AI-related investment is growing and contributing materially to GDP growth Graph 1In per cent A. AI-related investment accounts for a rising share of US GDP, …1 B. …has contributed materially to recent GDP growth…2 C. …and is projected to rise further in the coming years3 1 Data centre construction investment sourced from the US Census Bureau Construction Spending release. Data centre equipment investment is estimated to equal three times data centre construction investment, based on Noffsinger et al (2025). Other IT equipment investment is estimated as IT equipment investment in the US national accounts minus our estimate of data centre equipment investment. IT manufacturing facilities are sourced from the Computer/electronic/electrical manufacturing facility component of the US Census Bureau Construction Spending release. 2 Other IT expenditures include business investment in both hardware and software. 3 High demand scenario is based on the incremental growth in AI capacity in the “continued momentum” scenario in Noffsinger et al (2025). Medium demand scenario is based on the “Base Case” projections in IEA (2025). Sources: Bureau of Economic Analysis; US Census Bureau; International Energy Agency; authors’ calculations.

95 00 05 10 15 20 25 0 1 2 3 4 Data centres (incl equipment) IT manufacturing facilities Other IT equipment investment Software 16 17 18 19 20 21 22 23 24 25 −2 0 2 4 Data centres IT manufacturing facilities Other IT expenditures Other 20 25 30 0.0 0.2 0.4 0.6 0.8 1.0 1.2 Data centre investment Five-year projections: Medium demand scenario High demand scenario BIS Bulletin 3 area). 2 By mid-2025, expenditures on IT manufacturing facilities and data centres (including both equipment and construction costs) were equivalent to 1% of GDP. This, in turn, has seen total IT-related investment, including investment in other IT equipment and software, rise to 5% of GDP, exceeding its previous peak at the height of the dot-com boom in 2000. Unlike that earlier episode, which was driven almost entirely by spending by firms using IT products, the current boom is driven by IT-producing firms. AI-related investment has emerged as an important driver of GDP growth in the United States. From a negligible contribution before 2022, expenditures on semiconductor manufacturing facilities and data centres have contributed on average 0.4 percentage points (pp) to GDP growth over the subsequent three years (Graph 1.B).3 Total IT investment, which also includes spending by businesses on equipment and software to facilitate AI use, has accounted for almost half of GDP growth in recent quarters, helping to limit the negative adverse effects of trade tariffs on growth. These contributions could remain sizeable in the coming years. Analyst forecasts indicate that annual spending on data centres alone could increase by between $100 billion and $225 billion in the next five years. This would see data centre spending rise to between 0.8 and 1.3% of GDP, up from 0.5% today (Graph 1.C). The financing dimension of the AI boom The increasing importance of AI is already evident in the financial strategies of major IT firms. Firms that are currently driving the AI investment boom have historically operated with substantially less debt than other firms (Graph 2.A).

Instead, they have relied on their highly profitable operations to generate the cash flows needed to fund investments. However, these companies have ramped up their capital expenditures significantly, with investments growing both in absolute terms and as a share of revenues (Graph 2.B). The sheer size of these actual and anticipated investments, combined with dwindling free cash flows in some cases, are testing the limits of expansion based on cash flows. Indeed, free cash flows have recently lagged capital expenditures in absolute amounts. Equity financing may in turn be neither timely nor cost-effective at the current juncture: AI valuations are volatile and concentrated, issuance windows are narrow, and new stock sales can be costly and dilutive for long-dated, asset-heavy projects. As the need for AI-related investment grows, firms are increasingly turning to external sources of financing. Debt financing, through corporate bonds, leasing arrangements or loans, allows investors to spread costs over time and align financing maturities with the long economic life of data centre assets. Specific risks concerning construction, power availability and tenant concentration, however, mean that financing may fall outside the scope of traditional bank and bond financing (notwithstanding reportedly record bond issuance). A particularly fast-growing source of external financing is private credit. Private credit typically refers to non-bank credit extended by specialised investment vehicles (funds) mostly to small or medium-sized non-financial firms. Deals are directly negotiated between lenders and borrowers, and lenders hold the loan on their balance sheets until maturity. Private credit is characterised by bespoke covenant structures, greater certainty, faster execution and more flexible renegotiation. 4 The ability to provide bespoke financing arrangements that accommodate construction and operational risks arguably make private credit well placed to fund large, asset-heavy AI projects. 2 Investment in data centres includes spending on the building that houses the centre as well as the IT equipment inside it. A standard rule of thumb is that the physical structure accounts for one quarter of the spending on a new data centre and IT equipment accounts for the remaining three quarters (eg Noffsinger et al (2025)). 3 These figures may overstate the net contribution of this investment to GDP growth, as some data centre equipment is imported.