CICC: The AI wave may not be over, and the investment logic is shifting to “compete for efficiency and return”

The Zhitong Finance App learned that CICC released a research report saying that the current AI debt risk is more reflected in changes in the financing structure (from internal to external) rather than excessive debt size or deterioration in short-term solvency. The recent adjustment of AI assets is more like logical revaluation and sector rotation under the influence of multiple factors. The AI wave may not be over, but the investment logic is shifting from “competing for scale” in the first stage to “competing for efficiency and return” in the second stage: cloud vendors need to prove that capital expenditure can bring sustainable returns, chip and storage companies face a return to normalization of excess profits, while software, big model, and AI application companies compete more for cost control, product efficiency, and commercialization capabilities. In the long run, this will push the AI investment cycle into a more mature stage and help the sustainable development of AI technology.

CICC's main views are as follows:

As investment in AI computing power continues to expand, large US technology companies are gradually shifting from an “asset-light, high-cash flow” model to a “asset-heavy, high-capital expenditure” model. The scale of bond financing is growing rapidly, drawing market attention to the sustainability of AI debt. To assess risk, this paper introduces the Mins' financial instability hypothesis as an analytical framework. Minsky divides corporate financing models into three categories: hedge financing, speculative financing, and Ponzi financing to determine the stage and potential risks of the debt cycle from the perspective of debt repayment pressure.

The bank found that although the main players in AI infrastructure — the five major cloud computing companies — are issuing bonds faster, judging from their solvency, their operating cash flow can better cover the principal and interest of debt, and the pressure to repay debt is not great. Among them, the operating cash flow guarantee multiples of Microsoft (MSFT.US), Google (GOOGL.US), Meta (META.US), and Amazon (AMZN.US) are still at the leading level of the overall market, while Oracle (ORCL.US) is clearly weak. However, after deducting capital expenses, Amazon and Oracle's free cash flow turned negative, and Google's quarterly free cash flow also fell to a negative value for the first time, indicating that with the expansion of AI capital expenditure, some companies are increasing their dependence on external financing.

Looking further, the overall debt structure of the five major cloud vendors is relatively stable. On the one hand, their debt tenures are generally long. The average remaining term of bonds is about 9.3 years, which is significantly higher than the average of about 5 years for US corporate bonds; the average share of short-term debt in total debt is only 7.2%, far lower than the average of 27.4% of S&P 500 listed companies, and the pressure for short-term refinancing is relatively limited. On the other hand, in the short term, the average effective interest rate on debt of the five cloud vendors is lower than the current market interest rate, and is less sensitive to rising interest rates. Furthermore, with the exception of Oracle and Amazon, the other three companies all hold sizable cash reserves and have strong liquidity buffers.

Overall, the current AI debt risk is more reflected in changes in the financing structure (from internal to external) rather than excessive debt size or deterioration in short-term debt repayment capacity. Based on the Minsky Framework, Microsoft and Meta are still typical hedge financing; Google and Amazon are beginning to show signs of evolution from hedge financing to speculative financing, but they are still in the early stages; Oracle is relatively more vulnerable due to continued pressure on free cash flow and negative net cash reserves.

From a macro perspective, the leverage ratio and debt service ratio of the US household and corporate sector are at historically low levels. The banking system has sufficient capital, and the leverage ratio of financial institutions remains low. Furthermore, this round of AI investment relies more on bond market financing rather than bank credit expansion. The risk is mainly borne by the capital market, and the spillover impact on the banking system is relatively limited. As a result, the current AI debt risk is still within a manageable range, which is far from the true meaning of the “Minsky moment.”

The bank believes that the recent adjustment of AI assets is more like logical revaluation and sector rotation under the influence of multiple factors. The AI wave may not be over, but the investment logic is shifting from “competing for scale” in the first stage to “competing for efficiency and return” in the second stage: cloud vendors need to prove that capital expenditure can bring sustainable returns, chip and storage companies face a return to normalization of excess profits, while software, big model, and AI application companies compete more for cost control, product efficiency, and commercialization capabilities. In the long run, this will push the AI investment cycle into a more mature stage and help the sustainable development of AI technology.