"Similar to the Dot-Com Bubble, but Different": Three Risks in AI Financing

Based on Actual Demand and Long-Term Financing... Unlike the Dot-Com Bubble and Financial Crisis

Rising Reliance on External Funding, More Off-Balance-Sheet Financing, and Risks from Circular Deals in the Supply Chain

Watch for Delayed Monet

An analysis has found that risks stemming from the artificial intelligence (AI) industry could be triggered not only by weakening profitability expectations but also by credit risks in vulnerable sectors materializing. Disruptions in meeting equipment leasing or purchase commitments by neoclouds (cloud providers specializing in AI), as well as defaults, could cause losses for counterparty companies and financial institutions, with shocks spreading through interconnected financial channels. South Korea, one of the biggest beneficiaries of the AI boom, is closely tied to global supply chains, raising the possibility that real-economy and financial shocks originating in AI could be transmitted with greater force than in the past. The analysis says potential vulnerabilities and the possibility of risks emerging therefore warrant closer scrutiny.

An engineer works at an Amazon Web Services (AWS) AI data center. Photo by AP-Yonhap News

According to a report titled "Key Features of AI Industry Financing and Risk Assessment" by Kim Yoon-kyung and Wang Jae-gon, posted on the Bank of Korea's website on Oct. 28, financing in the AI industry has recently become increasingly reliant on external funding amid intensifying competition. Other notable features include a rise in hidden debt not reflected on balance sheets (off-balance-sheet financing) and deepening financial interconnectedness through investment and financial support within an AI supply chain whose participants are mutually dependent.

Markets expect the ratio of operating profit to capital expenditure by five major hyperscalers—companies that operate large-scale data centers—to fall from 1.6 in 2025 to 1.1 in 2026 and 0.9 in 2027, meaning capital expenditure is expected to exceed operating profit next year. Of the US$2.9 trillion hyperscalers are forecast to invest in data centers from 2025 to 2028, US$1.5 trillion—nearly half—is expected to come from external funding.

So far this year, hyperscalers have sharply increased external financing through long-term corporate bonds, while neoclouds have done so mainly through loans secured by graphics processing units (GPUs) or convertible bonds. Kim Yoon-kyung, a director in the Bank of Korea's International Affairs Team in the Research Department, said, "Hyperscalers have a sound capacity to absorb losses, but neoclouds face greater risks because their high debt ratios and revenue concentrated among specific customers make them vulnerable to changes in external conditions, such as rising interest rates and slowing demand." Hyperscalers, too, have become more reliant on long-term debt in the AI industry. If revenue generation is delayed or technological advances quickly render existing equipment and facilities obsolete, cash flows and asset values could decline before debts mature, significantly weakening AI companies' financial soundness.

The increase in off-balance-sheet financing is also a risk factor. To ease financing constraints and spread risk, AI companies are actively using off-balance-sheet borrowing, including setting up special-purpose vehicles (SPVs) and joint ventures (JVs) to raise funds for data center construction. Goldman Sachs forecasts that the outstanding balance of project finance (PF) bonds issued by hyperscalers through joint ventures will grow from more than US$108 billion this year to around US$300 billion next year. The Bank for International Settlements (BIS) says these transactions have the characteristics of "shadow borrowing," creating obligations that are effectively similar to debt outside the balance sheet. Complex loss structures may also make it difficult to determine who ultimately bears the risk, raising the possibility that risks will be underestimated.

Kim said, "If profitability deteriorates, the value of hyperscalers' stakes in SPVs or JVs could fall, and they could face obligations to repay their debt or honor long-term contracts instead. But it is difficult to identify these risks in a timely manner from financial statements alone." He added, "If an SPV or JV cannot repay its debt or the value of its collateral falls, losses could spread to the financial institutions and private equity funds that provided loans or investments."

Deepening financial interconnectedness within the AI supply chain is another concern. Hyperscalers and AI chipmakers are investing in or providing funding to companies in their supply chains, enabling those companies in turn to purchase their cloud services or AI chips and thereby securing future revenue. The scale of these circular transactions has grown substantially. The BIS estimates that circular financing among AI labs, hyperscalers and AI chipmakers accounts for more than 50%. Kim said, "Given that demand for computing continues to grow steadily, the risks are lower than during the dot-com bubble. However, this could lead to overestimation of expected returns in the AI industry and increase the risk of shocks spreading to the financial system and the real economy."

Given the sound financial positions of large leading companies, the likelihood of an AI-triggered financial risk emerging immediately was assessed as low. However, vulnerabilities could accumulate in an environment of rising long-term interest rates. The main potential triggers identified were ▲delays in monetizing AI ▲distress among weaker links such as neoclouds ▲the risk of shocks from circular financing spreading to the financial and real economies. Kim said, "The expected channels through which risks could spread are considered to share some features with the dot-com bubble and the global financial crisis. However, current AI investment is based on actual demand and relies mainly on long-term financing, distinguishing it from the excessive investment of the dot-com era and the short-term funding market freeze during the global financial crisis."

This content was produced with the assistance of AI translation services.