Zhongjin: AI capital expenditure spurs $3.5 trillion financing demand, debt repayment capacity becomes a key test.

CICC's report states that AI capital expenditure is rapidly expanding, with the capital expenditure of large cloud vendors as a percentage of revenue expected to increase from 12% in 2023 to over 40% by 2027. In the next five years, this growth is projected to generate approximately $3.5 trillion in external financing needs, primarily through investment-grade bonds and private equity. The core challenge lies in the fact that AI applications need to generate around $1 trillion in revenue each year to cover debt costs, with a required profit margin of about 50% and a depreciation period of approximately 5 years. The bank's calculations show that the peak period for large cloud vendor bond maturities will be between 2027 and 2032, with an annual maturity size of around $28 billion, representing a 60% increase compared to 2024-2026, adding to refinancing pressure. Key points from CICC include: AI infrastructure capital spending: from cash flow to debt The proportion of capital expenditure of large cloud vendors as a percentage of revenue is expected to increase from about 12% in 2023 to around 23% in 2025. It is expected that by 2027, this proportion may further increase to over 40%, reaching or even surpassing historical capital expenditure peaks in the internet and energy industries. Under the pressure of AI capital expenditure, cloud vendors' AI investments may shift from operational cash flow to greater reliance on external financing. Based on market expectations for the operating cash flow and capital expenditure of large cloud vendors, AI capital expenditure is expected to generate approximately $3.5 trillion in external financing needs over the next five years. Where will the $3.5 trillion come from? In the base case scenario, we expect the $3.5 trillion in external financing for AI capital expenditure to be covered by public equity markets ($0.4 trillion), investment-grade bonds ($1.5 trillion), leveraged financing ($0.3 trillion), asset securitization products ($0.3 trillion), and private equity ($1.1 trillion). Investment-grade bonds and private equity are the cornerstones of financing, with the former relying on the balance sheet expansion and cash flow debt repayment ability of cloud vendors, while the latter can meet the financing needs of high-risk and large-scale projects and can also supplement the financing gap more flexibly through the structuring of off-balance sheet financing. A trillion-dollar question: How will AI debt be repaid? To meet debt repayment requirements and shareholder returns, assuming a return on invested capital of 10%, we estimate that AI applications ultimately need to generate sustainable revenue of around $1 trillion per year. Assuming this revenue scale is reached by 2030, it means that AI application revenue needs to nearly double annually over the next five years. More important than revenue scale are profitability and asset life: with an EBITDA profit margin of about 50% in the mature stage, infrastructure depreciation needs to be around 5 years for investments to generate positive returns; even with a longer asset life, it will be difficult to cover capital costs if the profit margin is below 20%. Can refinancing replace cash flow? Large cloud vendors typically issue long-term bonds with maturities of 10-30 years, and infrastructure fund investments usually have maturities of over 10 years. After data centers are put into operation, project bonds and securitization can be used to replace construction phase financing. However, refinancing can only buy time and cannot replace cash flow: if project utilization, profit margins, and asset life are lower than expected in the long term, continued rolling financing will increase leverage and financing costs. The bank's calculations show that the peak period for large cloud vendor bond maturities will be between 2027 and 2032, with an annual maturity size of around $28 billion, representing a 60% increase compared to 2024-2026, adding to refinancing pressure. Opportunities and risks for financial institutions Banks can earn income from equity and debt underwriting, trading, M&A advisory, and project financing, while private equity and insurance institutions can acquire new long-term assets. In the second quarter of 2026, non-interest income for the six largest banks in the United States grew by 32% year-on-year; as of June 2026, bank loans to non-bank financial institutions grew by about 25% year-on-year, with AI financing being a significant increment. Bank income is usually recognized in the financing and construction stages, while credit risk will only manifest in the project operation and refinancing stages, showing a "income first, risk later" characteristic. Is AI a financial "bubble"? Currently, AI investments are still dominated by large cloud vendors with strong cash flow, and equity and subordinate capital can absorb losses before bank senior loans. Therefore, even if some projects do not meet expected returns, risks are more likely to be manifested as adjustments in enterprise valuations, slowing of capital expenditure, and partial credit losses, rather than immediately leading to a financial system crisis. However, on the other hand, the rapid growth of external financing, particularly private equity and leverage financing, has also created financial spillover effects; if commercial growth continues to lag behind capital expenditure, risks will shift from valuation corrections gradually to deteriorating credit quality, and will transmit through channels such as rolling financing, private credit, and securitization in the financial system. The scale of financing itself is not the issue; the real test is whether a cash flow cycle can be formed before financing costs rise and assets depreciate. Chart 1: Where does the $3.5 trillion in AI financing come from? Note: AI capital expenditure and OCF support portions are based on market consensus expectations; assume public equity market share of 10%; investment-grade bond issuance size is calculated based on constraints such as cloud vendor debt ratings, leverage ratios, and bond concentration; leveraged financing includes high-yield bonds and leveraged loans, securitized assets include ABS and CMBS, estimated based on market capacity and acceptability; the gap not covered by the above financing forms is assumed to be covered by private equity capital, including infrastructure funds, private equity funds, private credit, real estate funds, etc. Illustrative calculations, not representative of actual forecasts Sources: Financial statements of listed companies, Bloomberg, CICC Research Department Risks AI commercialization progresses slower than expected, AI capital expenditure and financing needs are lower than expected, and capital market financing conditions tighten.