Shenwan Hongyuan (SWHY) has released a research report indicating that as major cloud providers' free cash flow turns negative, market concerns over the sustainability of AI capital expenditure are growing. The market's pricing logic for AI is shifting from "capex expansion" to "financing and cash flow constraints." Looking ahead, the key to sustaining AI's "cash-burning model" lies in whether profit expectations can be met at the micro level and whether the Federal Reserve will raise interest rates at the macro level. The firm believes that, at the micro level, compared to the dot-com bubble era, current U.S. tech giants have solid financial metrics, with profit growth far outpacing market value gains. At the macro level, given weak employment, cooling inflation, and slowing economic momentum, the fundamentals do not yet support an immediate Fed rate hike.
Since late 2025, as leading cloud providers' free cash flow has turned negative, market concerns about the sustainability of AI capital expenditure have intensified. As corporate leverage rises and liquidity tightens, where is the boundary for AI's "cash-burning" model of capex?
Key Thought: The Boundary of the U.S. AI "Cash-Burning" Model?
AI Financing: Shifting from Internal Cash Flow to External Funding, with Corporate Bonds and Private Credit as Key Channels
The market's pricing logic for AI is transitioning from "capex expansion" to "financing and cash flow constraints." Taking the Q2 2026 earnings report as an example: when tech giants raised their capex guidance but experienced a significant compression in free cash flow, and the market questioned the monetization of cloud and computing assets, it punished "capex" instead, leading to a stock price decline (e.g., Alphabet Inc. (GOOGL.US) and Meta Platforms Inc. (META.US)). As AI capex scales up, internal cash flow becomes insufficient, increasing reliance on external financing. As of August 11, the market expects five major cloud providers to spend $769.2 billion on capex in 2026, a $40.9 billion upward revision from three months ago. Their capex may exceed operating cash flow, and AI investment is expected to remain elevated for the next five years (per BIS forecasts). AI's importance across various financing channels is growing, and in the future, corporate bonds and private credit may become key tools for AI infrastructure funding. In the first half, AI accounted for 34% of investment-grade corporate bonds, 40% of high-yield bonds, 85% of venture capital, and 58% of IPOs and follow-on offerings. According to FSB projections, AI infrastructure spending could reach $2.9 trillion from 2025 to 2028, mostly funded by external sources.
AI Financing Pressure: Risk Indicators Like Credit Spreads and Private Credit Default Rates Have Issued "Warning" Signals
Corporate bonds are the primary financing tool for major cloud providers, and increased issuance has widened spreads, with stock prices showing an inverse relationship with credit default swaps (CDS). Year-to-date, the five major cloud providers have issued over $200 billion in investment-grade corporate bonds, with AI bonds making up 40% of 10-year+ issuance. The widening of investment-grade tech credit spreads primarily reflects increased supply and uncertainty about returns. The private credit market is expanding and has become a key off-balance-sheet financing channel for major cloud providers. As of Q3 2025, global private credit reached over $2 trillion (72% in North America), potentially growing to $4.5 trillion by 2030. As of Q2 2026, the five major cloud providers' off-balance-sheet liabilities stood at $2.6 trillion. A rise in private credit default rates and an increase in Payment-in-Kind (PIK) notes from renegotiated deals in Business Development Companies (BDCs), along with the prior decline in software service stocks, have sent "warning" signals. The firm has constructed a comprehensive, high-frequency AI "financing risk" monitoring system. First, "existing leverage": although the expected debt-to-equity ratio of top cloud providers has been revised upward, the expected net debt-to-EBITDA ratio has been revised downward, indicating strong profitability. Second, "external financing": this includes high-frequency indicators such as credit spreads, equity financing volume, BDC stock prices, and data center ABS issuance volume.
The Tipping Point of the AI "Cash-Burning" Model: Fulfillment of Profit Expectations is Key, and a Fed Rate Hike Could Be a Major Catalyst
What rate of return is necessary for AI capex to be "worth it"? The firm modeled capex against annualized revenue from large models, estimating that the large model and cloud computing segments would each retain a portion of the operating profit from the revenue chain. The firm projects that by 2027, if an additional $350 billion in Annual Recurring Revenue (ARR) is achieved, bringing ARR to $500-600 billion, it could support over $1.4 trillion in capex. The firm has established a high-frequency monitoring system for AI "profitability outlook." The tipping point of the U.S. AI "cash-burning model" likely depends on the realization of profit expectations. Therefore, the firm built an AI "profit expectation" monitoring system, incorporating ARR and the profit expectations of leading cloud providers and the tech sector. Over the past year, these indicators have been trending upward or being revised upward. Looking ahead, the key to sustaining the AI "cash-burning model" is whether profit expectations can be realized at the micro level and whether the Fed will raise rates at the macro level. At the micro level, compared to the dot-com bubble era, current U.S. tech giants have solid financial metrics, with profit growth far outpacing market value gains. At the macro level, given weak employment, cooling inflation, and slowing economic momentum, the fundamentals do not yet support an immediate Fed rate hike.
Risk Warning
The risk of the oil price center shifting upward more than expected; the "hawkish" policy stance of Kevin Warsh; and the risk of a sharper-than-expected slowdown in the U.S. economy.
Comments