The operating cash flow of leading US AI companies has approached their capital expenditure, and the market is highly focused on the sustainability of their heavy-investment model.

In our view, as an emerging industry, AI technological breakthroughs and commercialization revenue naturally carry significant uncertainty, and the sustainability of capital expenditure must be understood from multiple perspectives.

Therefore, we step outside the traditional earnings forecasting framework and, from the logic of capital allocation, break down the core funding sources behind the US AI high capital expenditure.

On the funding supply side, the long-term financialization trend in the US continues — what some scholars analogize as "cotton candy" style liquidity expansion. At the macro level, this is reflected in the US broad liquidity relative to GDP continuing to rise; at the micro level, it manifests as new financial products emerging endlessly, revitalizing existing assets and activating idle funds, providing blood transfusions to industries with potentially high returns.

On the funding demand side, the US AI industry possesses dual attributes — technology and quasi-real estate — opening up both equity and debt financing channels. Its hardcore technology attribute fits equity financing, comparable to the VC and equity financialization model during the 1990s US dot-com revolution driven by pension deregulation; its high fixed-asset, quasi-real-estate characteristic means it can conduct debt financing through asset collateralization.

Although its fixed assets cannot be fully equated with real estate, the debt financing logic of US real estate financialization after 2000 — relying on derivatives to revitalize existing assets and expand credit — has reference significance.

Currently, US AI financialization remains at an early stage. The financing cash flow of leading AI companies has just turned positive, and debt financing costs are still significantly lower than equity financing costs. New industry financing is mainly debt-based, and large-scale equity financing has not yet begun.

Reviewing history, the bursting of both the US dot-com bubble and the real estate bubble coincided with three phenomena: high leverage, deteriorating fundamentals, and systematic tightening of liquidity.

From these indicators, the US AI industry currently has no systemic risk. However, short-term liquidity risk in the US cannot be ignored.

The US economic trajectory, the game between the market and the Federal Reserve, combined with oil price uncertainty from US-Iran geopolitical conflict, means US liquidity could tighten more than expected.

If this occurs, global markets will also be affected, and market concerns about the sustainability of AI companies' capital expenditure may intensify.

Over the past two months, the US stock market has experienced significant volatility, which has also affected other regions. The volatility stems from both changes in monetary policy and market concerns about the sustainability of AI sector capital expenditure, with the former exacerbating the latter.

In short, capital expenditure at several major US AI companies has risen rapidly this year, and although growth may slow next year compared to this year, it will still be relatively fast.

On the other hand, AI companies' revenue faces significant uncertainty. Operating cash flow has become comparable to capital expenditure, and some companies have begun issuing debt to support capital expenditure.

Rising US Treasury yields have further intensified market doubts: with revenue facing uncertainty, how can capital expenditure be sustained? This is a topic of widespread market concern.

Chart 1: US AI companies' capital expenditure and operating cash flow are comparable. Note: Data in the chart is the sum of Microsoft, Amazon, Google, Meta, and Oracle. Source: Epoch AI, Bloomberg, iFinD, CICC Research Department.

AI is an emerging industry, and the timing of its technological breakthroughs and the pace of commercial revenue growth are naturally difficult to predict. We will not focus on revenue, but instead choose to explore the potential funding sources for US AI capital expenditure from the perspective of capital allocation.

In reality, investment is the source of revenue. If capital expenditure cannot be guaranteed, revenue growth will lack support. In other words, capital expenditure and revenue are not isolated variables. Therefore, discussing the financing situation of capital expenditure also helps us judge the revenue outlook.

We first analyze the advantages of the AI industry in attracting funds based on its attributes. In the second part, we analyze changes in US social funding supply since the 1990s.

Specifically, we dissect how US "financialization" has reshaped its liquidity. "Financialization" has no unified definition, and its manifestations are complex, but we believe at least two points are meaningful for analyzing AI financing.

From a macro perspective, financialization means the ratio of broad liquidity to GDP rises — in plain terms, the "money" in society gradually increases. From a micro perspective, financialization means the number and variety of new financial products increase.

Financial products (such as derivatives) make funds "active" and enhance asset liquidity. Of course, the macro and micro levels are not separate — an increase in the number of financial products itself may lead to an increase in macro-level liquidity.

Gillian Tett, Dean of King's College Cambridge and senior Financial Times journalist, in "Welcome to the Financial Cotton Candy Era," vividly explained the phenomenon of US financialization and traced the origin of the term financial "cotton candy." Therefore, we might borrow the vivid term financial "cotton candy" to approximately refer to US financialization.[1]

Although the US strengthened financial regulation after the 2008 subprime crisis, especially bank regulation, overall its financialization phenomenon has not significantly reversed.

On this basis, we review the financialization before the two bubble bursts in the US since 1990: the dot-com financialization of the 1990s and the real estate financialization before the subprime crisis. We then analyze the current stage of financialization in the US AI industry, and finally analyze the liquidity risks the US AI sector may face.

AI: Both Technology and "Quasi-Real Estate" Attributes

We believe AI has both technology attributes and "quasi-real estate" attributes (high proportion of fixed assets), meaning it can raise funds through both equity financing and debt financing (mortgage loans).

AI's technology attributes need no elaboration — its technological progress and productivity improvement seem faster than dot-com companies. One manifestation is that AI relies less on labor than dot-com companies and can even replace labor.

From mid-level data, since ChatGPT's release in 2022, non-farm employment in US telecommunications and computer industries has slightly shrunk by 6%. In contrast, from Netscape's IPO in August 1995 (the first year of the US dot-com bubble) to the bubble burst in March 2000, non-farm employment in telecommunications and computer industries rose by more than 60%.

Chart 2: Divergence in US telecommunications and computer industry employment trends between the dot-com revolution era and the AI era. Note: The AI tech revolution uses ChatGPT's release (November 2022) as T=0, and the dot-com revolution uses Netscape's IPO (August 1995) as T=0. Non-farm employment includes telecommunications, data processing, and computer system design industries; data has been standardized. Source: iFinD, CICC Research Department.

From micro data, if we set the average labor density (employees to fixed assets ratio) index of Google, Microsoft, and Meta in 2021 as 100, the index was 109 in 2019 and dropped to 37 in 2025.

Specifically, their fixed asset index rose from 100 in 2021 to 306 in 2025, while their employee count index only rose to 112.

Chart 3: Labor density of three major US tech companies. Note: Major tech companies in the chart include Google, Microsoft, and Meta. Labor density is measured by number of employees/fixed assets. Source: iFinD, CICC Research Department.

Chart 4: Fixed assets of three major US tech companies grew faster than employee count. Note: Major tech companies in the chart include Google, Microsoft, and Meta. Source: iFinD, CICC Research Department.

During the dot-com revolution, taking Microsoft as an example, its fixed asset investment and employee count increased simultaneously, with employee count growing faster, driving labor density up. Even after the dot-com bubble burst in 2000, corporate labor density continued to rise.

If we set Microsoft's labor density index in 1995 as 100, the index was 138 in 2000 and further rose to 174 in 2005. Specifically, its employee count index rose from 100 in 1995 to 220 in 2000 and 343 in 2005. During the same period, its fixed asset index only rose from 100 to 197.

Chart 5: Microsoft's labor density trend (1995-2005). Note: Labor density is measured by number of employees/fixed assets. Source: iFinD, CICC Research Department.

Chart 6: Microsoft's employee count and fixed assets trends. Source: iFinD, CICC Research Department.

From inflation-adjusted revenue per capita, current AI companies generate more than twice that of traditional internet companies, reflecting higher efficiency at AI companies.

For example, in 2026, Anthropic's revenue per capita to US GDP per capita ratio is approximately 100 times, while OpenAI and Nvidia may be 63 and 57 times respectively. In 2025, Meta, Apple, Google, and Microsoft had ratios of 28, 28, 23, and 14 times respectively. In 1999, Apple, Microsoft, and Cisco had ratios of 26, 18, and 16 times respectively.

Chart 7: Tech company revenue per capita/US GDP per capita ratio. Source: iFinD, Epoch AI, CICC Research Department.

The reason we say US AI companies also have "quasi-real estate" attributes is their high proportion of fixed assets.

For example, the average ratio of fixed assets to total assets for the five cloud providers — Microsoft, Alphabet, Amazon, Meta, and Oracle — currently exceeds 40%, while the average fixed asset ratio for four traditional companies — Walmart, ExxonMobil, Procter & Gamble, and Caterpillar — is currently 37%.

The average fixed asset ratio for dot-com era companies (Microsoft, Cisco, Amazon, WorldCom, and Lucent) in 2000 was 15%.

Among fixed assets, land and buildings account for 46% for Microsoft and Meta, 37% and 40% for Amazon and Oracle respectively, and Google has the lowest land and building ratio at 11%. Software, hardware, and equipment assets as a proportion of fixed assets are 47%, 38%, 49%, 41%, and 56% respectively for these companies.

Chart 8: Current AI companies have a relatively high proportion of fixed assets. Note: Each group calculates ratios for sample companies separately and then takes a simple average. AI companies include Microsoft, Alphabet, Amazon, Meta, and Oracle; traditional companies include Walmart, ExxonMobil, Procter & Gamble, and Caterpillar; dot-com companies include Microsoft, Cisco, Amazon, WorldCom, and Lucent. AI companies use the latest available 2026Q2 quarterly report or annual report; traditional companies use the latest annual report; dot-com companies use FY2000 annual reports. Source: iFinD, Wind, Bloomberg, CICC Research Department.

Chart 9: Fixed asset breakdown of five US AI companies. Note: Uses the latest available 2026Q2 quarterly report or annual report. Source: iFinD, Wind, Bloomberg, CICC Research Department.

Dynamically, the fixed assets to total assets ratio for these companies generally rose from 22% in 2020 to 41% in 2025, while the fixed assets to total assets ratio for the five dot-com companies remained basically at around 15% from 1995 to 2000.

Chart 10: AI companies' fixed asset ratio is rising. Note: Each group calculates ratios for sample companies separately and then takes a simple average. AI companies include Microsoft, Alphabet, Amazon, Meta, and Oracle. Source: iFinD, Wind, Bloomberg, CICC Research Department.

Chart 11: Dot-com companies' fixed asset ratio remained basically unchanged. Note: Each group calculates ratios for sample companies separately and then takes a simple average. Dot-com companies include Microsoft, Cisco, Amazon, WorldCom, and Lucent. Source: iFinD, Wind, Bloomberg, CICC Research Department.

Financialization Enhances Liquidity

Financialization has no unified definition. We might cite Epstein (2005): financial purposes, financial markets, financial entities, and financial institutions play an increasingly important role in domestic and international economic operations.[2] The implication of this definition is that finance's role in resource allocation is rising.

Different schools of thought believe the driving factors of financialization differ.[3] Neo-Marxists argue that after the 1970s, the return on physical investment in capitalism declined, and excess capital flowing into financial speculation was the main cause of capitalist financialization.

Post-Keynesians argue that corporate governance shifted to shareholder value priority, leading companies to reduce physical investment and increase stock buybacks and financial asset allocation, resulting in financial investment crowding out physical capital accumulation.

Neoclassicists argue that financial liberalization and asset securitization enhance asset liquidity and optimize resource allocation, representing progress in economic development.

Although different theoretical schools understand financialization's development differently, Luo Yu (2017) argues that the collapse of the Bretton Woods system and the global entry into a pure credit money system undoubtedly created conditions for financialization.

Specifically, he argues that in terms of payment media, the proportion of physical cash decreased significantly, transactions became account digital transfers, transaction costs decreased, and transaction scale increased. In terms of value storage, various financial assets serve a "quasi-money" value storage function, and the boundary of broad money has become broader.[4]

The main manifestations of financialization include the following aspects[5]: (1) Broad liquidity is more abundant. For example, the ratio of broad liquidity to GDP rises, the financial sector's profit share increases, and the financial sector grows relatively faster. (2) Low-liquidity assets become high-liquidity assets. Various new financial instruments emerge endlessly, securitization converts non-liquid assets into liquid assets, and financial transaction scale grows substantially. (3) Capital profit acquisition increasingly relies on financial channels rather than physical trade and commodity production. For example, corporate profits depend more on investment income. (4) The discourse power of capital market-based financial systems continues to expand. For example, stock market/GDP and debt/GDP ratios continue to rise, while bank credit/GDP ratio declines. (5) Corporate governance places more emphasis on shareholder value. For example, buybacks in the US stock market increase.

From a macro perspective, after the mid-1990s, the ratio of US broad liquidity to GDP trended upward, about 20 percentage points higher than the M2 to GDP ratio. After the financial crisis, the US strengthened financial regulation, and the US banking sector's lending business relatively slowed, but the US bank assets/GDP ratio did not shrink, and the banking sector's non-lending business continued to expand.

Chart 12: US broad liquidity to GDP ratio is rising. Note: Broad liquidity uses M2 + money market funds. Source: iFinD, Jordà-Schularick-Taylor Macrohistory Database, CICC Research Department.

Chart 13: US bank assets to GDP ratio has not shrunk. Source: iFinD, CICC Research Department.

From a micro perspective, new financial products continue to appear. From 1985 to 2014, US non-financial corporate securities issuance data shows that financial product varieties continued to expand. In 1985, about 150 types of financial products were available in the market, growing to 751 by 2014, an increase of more than 600. The proportion of non-financial corporate financing through new financial products rose, exceeding 50% during 2010-2014.

Chart 14: US non-financial corporate financing methods. Note: The left chart shows the total number of different financial products in each period from 1985-1989 to 2010-2014. The bar chart divides financial products into three categories: (1) Old products: products launched before 1985; (2) New-Entrants: products newly launched in the current period; (3) New-Young: products launched after 1985 but first issued in a previous period. The right chart shows changes in total financing amounts in each period from 1985-1989 to 2010-2014. The bar chart further decomposes financing scale into issuance financing amounts of old products, new-entrants, and new-young products. Source: Ana Babus, Matias Marzani, and Sara Moreira, "The Rise of Specialized Financial Products," NBER Working Paper 34594 (2025), CICC Research Department.

Other indicators also have reference significance. For example, the number of US mutual funds, futures, and options contracts continues to rise. In 2025, active futures and options contracts rose significantly, mainly due to the growth of event contracts, with the number of contracts on event prediction platforms such as Kalshi and ForecastEx increasing notably.

Event contracts are derivative contracts based on whether a specific event occurs or its outcome, commonly found in prediction markets, and can involve economic indicators, financial markets, weather, politics, and other events. After the 2024 US election, prediction markets gained enormous traffic, driving the growth of event contracts.

Lessons from US Dot-Com and Real Estate Financialization

The US has had two experiences of accelerated financialization since the 1980s: one is the dot-com financialization of the 1990s, and the other is the real estate financialization after 2000.

The 1990s dot-com financialization manifested as substantial growth in US VC and a hot equity financing market, increasing the ways funds could be used and enabling funds to flow. An important driving factor for financialization at this stage was the relaxation of pension investment constraints, promoting VC development, with IPOs providing exit channels, releasing VC funds, and attracting more capital entry.

The 2000s US real estate financialization both created new liquidity and increased liquidity usage methods. Its main manifestation was that through securitization, real estate lending institutions released capital, broke through balance sheet constraints, formed more loans, and created new liquidity. Securitization further enabled more financial institutions to participate in real estate financialization, increasing the ways funds could be used.

US corporate equity financing methods include private equity financing and public equity financing. In the startup phase, private equity financing is mainly used; after the growth phase, public equity financing methods are used.

After 1979, the US relaxed pension investment restrictions, promoting substantial VC growth. VC turned "internet startups without collateral" into enterprises that could more easily obtain financing. IPOs provided exit channels, releasing funds tied up in equity and forming new investments. IPO exits also helped VC attract more capital.

Specifically, before 1979, the "Prudent-Man Rule" of the Employee Retirement Income Security Act of 1974 restricted US pension participation in startup investment. In 1979, the US Department of Labor clarified the "Prudent-Man Rule," stating that pension investment judgments should consider the risk diversification effect of the entire portfolio, no longer requiring every project to be low-risk.

After the reform, pensions could increase venture capital allocation, institutional LPs entered on a large scale, profoundly changing the funding source structure of the US VC industry. US VC new fundraising scale and investment amount showed substantial growth.

From the perspective of US VC funding sources, in 1978 only about 15% came from pensions, while by 1990, about 53% came from pensions. US VC fundraising and investment amounts also began to grow rapidly in the 1990s.

Chart 15: US VC funding sources. Source: Gompers & Lerner. 1998. What Drives Venture Capital Fundraising?, Nicole R. Onorato. 1997. Trends in Venture Capital Funding in the 1990s, CICC Research Department.

Chart 16: US VC fundraising and investment amounts. Source: National Venture Capital Report 2013&2023&2025, CICC Research Department.

After funds flowed to VC, VC opened equity financing channels for internet startups lacking physical collateral. Internet startups typically have light assets, many intangible assets, negative cash flow, no stable profits, and lack bank collateral, making it difficult to finance through traditional channels — they can only use equity financing.

In the 1990s, US IPO numbers increased, the proportion of VC-backed IPOs rose, and equity financing increased substantially.

Chart 17: After the 1990s, US VC-backed IPOs rose. Source: Jay R. Ritter IPO Data, CICC Research Department.

To achieve better exits, in the late 1990s, US IPOs were heavily discounted, especially tech stocks, to attract more investors. Newly established VCs with little historical performance had strong incentives to push portfolio companies to IPO, accept higher IPO discounts, use first-day surges to create successful exit cases, demonstrate performance to LPs, and facilitate raising the next VC fund.

In the late 1990s, IPO exit returns were very substantial, further stimulating LPs to contribute to VC on a large scale. As IPO issuance increased, the average quality of listed companies declined. The average age of VC-backed listed tech companies dropped from 9 years (1991) to 4 years (1999).

After 2000, US real estate financialization both revitalized existing assets through derivatives and drove broad money expansion, closely related to debt financing. Various financial institutions, through asset securitization, converted real estate assets into tradeable, collateralizable, and leverageable financial assets, thereby connecting the real estate market with capital markets, the banking system, and the shadow banking system.

The primary role of residential mortgage MBS securitization was to revitalize existing credit assets, making existing money more active and increasing money circulation velocity; meanwhile, after banks recovered funds, they could continue to issue new loans to households, and through the commercial bank credit creation mechanism, bring about broad money expansion.

Methods for revitalizing existing assets include credit enhancement, re-securitization, and adding "insurance" to mortgage loans, turning assets that were originally difficult to "circulate" into "circulatable" assets.

Credit enhancement roughly has three methods: (1) Tranching, converting mortgage loans into securities of different risk levels to meet different investor needs; (2) Over-collateralization: the asset pool scale exceeds the total principal of all securities issued externally, with the excess portion used to absorb losses; (3) Excess spread: mortgage loan rates are higher than MBS rates, ensuring a positive spread.

CDO (Collateralized Debt Obligation) re-securitizes and tranches MBS with subprime loans as underlying assets, converting them into securities of different grades. Additionally, CDS (Credit Default Swap) adds a layer of "insurance" to mortgage loans, further enhancing their liquidity.

Data shows that after the 1980s, the US loan/GDP ratio rose substantially, with residential real estate loans rising the most. In 1995, the ratio of residential mortgage loans to GDP was less than 50%, while by 2007 it exceeded 80%.

From 2001 to 2006, the US mortgage securitization ratio rose substantially, especially for lower-quality Alt-A and Sub-prime mortgages. For example, the securitization ratio of Alt-A mortgages rose from 19% (2001) to 91% (2006), and the securitization ratio of Sub-prime mortgages rose from 46% (2001) to 75% (2006).

US mortgage holders shifted from banks to non-banks. Early US residential securitization was mainly participated in by government-sponsored agencies such as Fannie Mae, Freddie Mac, and Ginnie Mae. Later, more non-government-sponsored agencies began participating in mortgage securitization (including investment banks).

Through securitization, government-sponsored agencies (such as Freddie Mac and Fannie Mae) and ABS issuers held more than 60% by 2006.

US AI Financialization Is Still at an Early Stage

The financing cash flow of major US AI companies has just turned from negative to positive, beginning to shift toward external financing.

The investment expenditure/revenue ratio of representative US AI companies rose from about 10.3% in 2020 to about 32.8% in 2025, and further rose to about 51.7% in 2026Q2, exceeding the high point during the dot-com revolution era.

In terms of investment expenditure sources, AI companies' operating cash flow/revenue ratio remained above 40%, and the financing cash flow/revenue ratio turned from about -7% in 2025 to about 16.2% in 2026Q2.

Chart 18: AI companies' investment expenditure sources. Note: Each group calculates cash flow/revenue ratios for sample companies separately and then takes a simple average. AI companies include Microsoft, Alphabet, Amazon, Meta, and Oracle. AI companies 2020-2025 are annual reports; 2026Q1 and 2026Q2 use the trailing four-quarter aggregate as of the respective quarter to reduce seasonal effects. Source: Wind, Bloomberg, CICC Research Department.

Chart 19: Dot-com companies' investment expenditure sources. Note: Each group calculates cash flow/revenue ratios for sample companies separately and then takes a simple average. Dot-com companies include Microsoft, Cisco, Amazon, WorldCom, and Lucent. Source: Wind, Bloomberg, CICC Research Department.

Currently, the incremental external financing of AI companies is mainly debt financing, and equity financing has not yet begun on a large scale. In 2026Q2, the AI companies' net financing cash flow/revenue ratio rose to 16.2%, higher than dot-com companies in 2000 (13.2%).

Among this, the net debt financing/revenue ratio was 21.2%, and the net equity financing/revenue ratio narrowed to -3.3%, while at that time dot-com companies had net inflows from both debt and equity financing.

By comparison, the current round of AI financing expansion is more concentrated on the debt side, with equity financing still reflected as net buybacks. If future debt financing costs rise, AI companies may shift to equity financing.

Chart 20: AI companies' financing cash flow sources. Note: Each group calculates cash flow/revenue ratios for sample companies separately and then takes a simple average. Net debt financing = debt increase/issuance - debt decrease/repayment; net equity financing = equity increase/issuance - equity decrease/buybacks - dividends paid; other financing = CFF - net debt financing - net equity financing. AI companies include Microsoft, Alphabet, Amazon, Meta, and Oracle. AI companies 2020-2025 are annual reports; 2026Q1 and 2026Q2 use the trailing four-quarter aggregate as of the respective quarter to reduce seasonal effects. Source: Wind, Bloomberg, CICC Research Department.

Chart 21: Dot-com companies' financing cash flow sources. Note: Each group calculates cash flow/revenue ratios for sample companies separately and then takes a simple average. Net debt financing = debt increase/issuance - debt decrease/repayment; net equity financing = equity increase/issuance - equity decrease/buybacks - dividends paid; other financing = CFF - net debt financing - net equity financing. Dot-com companies include Microsoft, Cisco, Amazon, WorldCom, and Lucent. The dot-com companies' net equity financing/revenue 1995 average is significantly elevated mainly due to Amazon's small revenue base that year. Source: Wind, Bloomberg, CICC Research Department.

According to Dealogic data, as of June 2026, AI companies (Alphabet, Amazon, Meta, Microsoft, and Oracle) issued $159 billion in bonds globally, compared to $108 billion for all of 2025 and only $17 billion for all of 2024.

The total liabilities of these five AI companies increased substantially starting from 25Q4, with an increase of $211.5 billion in 26Q2 alone. On the margin, short-term debt grew more in 26Q2 while long-term debt growth slowed, indicating increased debt repayment pressure for these companies.

Chart 22: US AI corporate bond issuance has grown substantially. Note: Corporate bond issuance data includes Google, Amazon, Meta, Microsoft, and Oracle. Corporate bond issuance 2026 data is as of June 5, 2026. Source: WSJ, Dealogic, iFinD, Bloomberg, CICC Research Department.

Chart 23: US AI corporate debt composition. Note: AI companies use Google, Amazon, Meta, Microsoft, and Oracle. Source: iFinD, CICC Research Department.

Private credit has become an important funding source in the off-balance-sheet financing system of US AI companies. Private credit is gradually replacing traditional bank loans to support long-cycle AI infrastructure investment.

Due to the enormous investment scale and long payback period of AI projects, traditional banking systems have limited risk appetite, and large private credit institutions such as Apollo, Ares, Blackstone, and Blue Owl have become key funding providers.

According to BIS estimates, private credit loans to AI-related companies have grown to over $200 billion, accounting for nearly 8% of total private credit. Private funds involved in AI lending currently account for about 20%.

Private credit funds mainly flow in through data center SPVs, computing power operators like CoreWeave, and GPU financing platforms, with stable cash flows from long-term contracts with M7 as debt repayment support, prompting a shift in funding sources from the banking system to the non-bank financial system, effectively playing a bank-like role.

Chart 24: Private credit direct lending to US AI companies is growing rapidly. Source: Aldasoro, Iñaki, Sebastian Doerr, and Daniel Rees. Financing the AI boom: from cash flows to debt. No. 120. Bank for International Settlements, 2026., CICC Research Department.

Chart 25: The proportion of private credit involving AI loans is rising. Source: Aldasoro, Iñaki, Sebastian Doerr, and Daniel Rees. Financing the AI boom: from cash flows to debt. No. 120. Bank for International Settlements, 2026., CICC Research Department.

Currently, AI companies' debt pressure is not large, and there is still room for debt financing. The five major US cloud providers have sufficient operating cash flow, and overall debt repayment capacity is stable.

Cash flow interest coverage ratios are all above 1, with Oracle at about 7 times (the lowest), Amazon at about 48 times, and Google at about 82 times (the highest). Debt service ratios are all below 1, with Oracle at about 48%, Amazon at about 12%, and Google at only about 6% (the lowest).

Chart 26: Debt service ratio (present value of debt principal and interest/operating cash flow) of five major US cloud providers. Note: Calculated according to the BIS debt service ratio formula. Source: Bloomberg, CICC Research Department.

The effective interest rates of the five major US cloud providers are currently not high, and net cash reserve situations are divergent. In 2026Q2, the effective interest rate of the five major cloud providers was 4.9%, lower than the Baa corporate bond yield.

The financing costs of the five major cloud providers are still not high, partly due to "locking in" rates in the earlier low-interest-rate environment and issuing bonds in non-US currencies. In terms of net cash reserves, Google, Microsoft, and Meta are relatively better, while Amazon and Oracle have negative net cash, indicating greater reliance on external financing.

Chart 27: Effective interest rates of five major US cloud providers are still not high. Note: Effective debt interest rate = interest expense over the past 12 months/total debt. Source: Bloomberg, CICC Research Department.

Asset securitization may become the next step in US AI company financing. As long-term leases and capacity procurement contracts continue to accumulate, the stable cash flows of data centers and computing power assets are beginning to meet securitization conditions.

In early 2024, CyrusOne issued $1.175 billion in data center asset-backed securities (ABS)[6]; in March 2025, Sabey issued $325 million in ABS, with collateral being 7 data centers in the US[7]; in May 2025, Stack issued $345 million in ABS, with collateral being real estate, leases, and related interests of 11 US data centers[8]; in September 2025, QTS issued $600 million in ABS, with collateral being a data center in Phoenix[9]; in early 2026, Compass Datacenters issued $830 million in data center ABS, with the underlying asset being long-term contracts for leasing computing power to cloud providers[10].

AI data center loans are being packaged into ABS and distributed to institutional investors such as pension funds and insurance funds, similar to the real estate securitization process.

US AI companies' long-term equity investments have recently begun to rise, but the overall proportion is not high. AI companies' external long-term equity investments have risen since 2025, reaching about 5% of total assets by 2026Q2.

Compared to dot-com companies, AI companies' long-term equity investment proportion is not high, indicating that AI companies' equity financialization is still at an early stage. Moreover, by spinning off some heavy-asset projects through equity investment, removing them from the balance sheets of hyperscale cloud providers, this is also a way of risk isolation.

Chart 28: US AI companies' long-term equity asset proportion. Note: Each group calculates ratios for sample companies separately and then takes a simple average. AI companies include Microsoft, Alphabet, Amazon, Meta, and Oracle. Source: Wind, Bloomberg, CICC Research Department.

Chart 29: US dot-com companies' long-term equity asset proportion. Note: Each group calculates ratios for sample companies separately and then takes a simple average. Dot-com companies include Microsoft, Cisco, Amazon, WorldCom, and Lucent. Source: Wind, Bloomberg, CICC Research Department.

Short-Term Risks

The above analysis implies that under the broader context of US financialization, combined with AI companies having both stronger technology attributes than dot-com companies and "quasi-real estate" attributes, AI companies can raise funds through both equity and debt financing.

Compared with the historical dot-com revolution and real estate, US AI companies' current financialization is still at an early stage, and their funding sources need not be overly concerned about. But this is not to say we can ignore short-term liquidity fluctuation risks, especially when oil prices remain high and the US rate hike path is unclear.

Reviewing history, before the US dot-com bubble and real estate bubble burst, three phenomena appeared simultaneously: (1) High leverage. During the dot-com revolution, a large number of US Dotcom companies went public, and the Shiller PE ratio rose from 17 times (1990) to 44 times (2000). Before the subprime crisis, the US household sector macro leverage ratio rose from 70% (2000) to 98% (2007). (2) Deteriorating fundamentals. Around 2000, US internet user penetration reached an inflection point, and starting in 2006, US mortgage default rates began to rise. (3) Systematic tightening of financing conditions. The Fed began raising rates in 1998, with the federal funds rate rising from 4.75% (1998) to 6.5% (2000). Before the subprime crisis, the federal funds rate rose from 1.0% (2003) to 5.23% (2006).

Currently, for AI companies, these three phenomena are not yet significant. US corporate sector leverage is still relatively low, and stock market valuations are not particularly high; the US economy remains relatively strong, with nominal growth around 5%, and major tech companies represented by Mag 7 still have relatively high revenue and profit growth; financing conditions have not yet shown明显 sustained tightening.

However, short-term risks still need attention and may disrupt AI company financing. Currently, US financing demand is strong, and with rising inflation, funding conditions are marginally tightening.

Recently, actual financing pressure (real interest rates TIPs) has risen, and the market has concerns about the sustainability of high leverage in the AI industry chain. In September 2026, the Bank of England's Financial Policy Committee pointed out that AI debt is rising and its lack of transparency makes risk assessment difficult, and as financial conditions tighten, credit markets (including private credit) are relatively fragile[11].

In May 2026, the Federal Reserve stated in its "Financial Stability Report" that AI and private credit are risks to watch in the near term[12]. The Fed believes that as capital expenditure increasingly relies on debt financing, it increases the system's leverage effect; private credit faces increasing pressure, including investor redemptions, deteriorating sentiment, and credit quality impacts from AI.

In September 2026, S&P downgraded Oracle's rating to BBB-, just above junk level[13]. After the Fed raised rates in September 2026, by the end of September, yields on some CoreWeave high-yield bonds rose by more than 0.5 percentage points[14].

In July 2026, Meta issued $12 billion in new bonds to finance data centers, with a bond yield of 7.5%, 0.4 percentage points higher than when it issued $27 billion in bonds in October 2025[15].

Chart 30: Bubble bursts generally occur after systematic tightening of financing conditions. Source: iFinD, CICC Research Department.

Specifically, two types of risks need attention in the short term. First, tightening financing conditions weakens private investment momentum. If the Fed raises rates, or banks tighten lending standards, corporate financing costs and difficulty in obtaining funds may rise, dragging down private investment expansion.

Second, short-term financing pressure intensifies, amplifying liquidity risk. Repo and asset-backed commercial paper market spreads remain above normal levels, indicating that short-term financing pressure for some market participants has not yet subsided. If monetary policy tightens further at this point, it may increase debt rollover difficulty and amplify short-term funding market volatility.