AI venture capital logic is undergoing a transformation: from additional capital expenditure to re-evaluating returns

A recent clear signal in the market is the dramatic shift in investor attitudes toward AI venture capital. When a tech giant announces cuts in capital expenditure, it used to excite the market; now, it often leads to a sharp decline in stock prices. Behind this contrast reflects a subtle change in the underlying logic of venture capital strategies—from a narrative of optimistic capital expansion to a cautious assessment of actual investment returns. What exactly is the market worried about? We analyze three major risks in the AI sector from the perspectives of capital returns, financing pressures, and corporate interconnectedness.

Can Massive Capital Expenditures Turn Into Real Returns?

The most notable feature of this AI wave is the unprecedented scale of investment by tech companies in computing power and data centers. According to FactSet and Bloomberg, the top five cloud service providers have cumulatively spent $357.2 billion on AI-related capital expenditures over the past four quarters, with expectations to rise further to about $500 billion by 2026.

What does this mean? From a cash flow perspective, these five companies have allocated about 60% of their free cash flow to AI investments. In other words, nearly two-thirds of their generated cash is being invested in AI. Some companies are even more extreme—certain firms have capital expenditures exceeding 500% of their operating cash flow, meaning internal cash flows are insufficient to cover their investments.

The key question: can such large-scale investments truly deliver the expected returns?

Currently, AI is recognized as the most promising technological direction, but its commercialization path remains unclear, and profit models are still under exploration. This introduces two uncertainties: first, whether these investments can eventually generate substantial profits; second, as investment scales up, the marginal efficiency of AI investments is likely to decline.

This aligns with historical patterns. Economics teaches us that capital investments typically follow the law of diminishing returns—the more you invest, the slower the growth. However, the costs of AI investments have not decreased accordingly. Since 2023, prices for computers and information processing equipment have continued to rise, contrasting sharply with the declining prices of capital goods during the internet boom of the 1990s. In other words, current AI investments are still in a “scale-inefficient” stage; large investments have not resulted in proportionate cost reductions.

This phenomenon is prompting the market to reprice stock valuations. Overly optimistic expectations about capital expenditure stories will inevitably be corrected. The significant adjustment in Oracle’s stock price has already signaled this: the era of simply telling the story of capital spending is over; the market now demands real, tangible returns rather than endless capital infusions.

Such expectation adjustments are common in history. Each wave of technological revolution often involves similar price fluctuations. While AI, as a general-purpose technology, has long-term potential to boost labor productivity, technological progress tends to be phased rather than linear. In major industry cycles, multiple rounds of investment expansion and correction typically occur every 3-5 years. Stock prices, as leading indicators of investment activity, tend to be more volatile. As economist Keynes noted, stock buyers often lack full understanding of what they are purchasing; when overly optimistic fantasies break down, market prices tend to fall.

The Rising Dependence on Financing and the Hidden Credit Risks

Large-scale corporate investments often take years to complete, but the costs of labor and raw materials supporting these investments must be paid immediately. This means companies need to spend substantial funds upfront before projects generate returns. These funds come either from internal cash reserves or external financing.

The availability and cost of financing depend critically on lenders’ confidence in the company’s repayment ability and operational prospects. Once this confidence wavers, credit conditions tighten, increasing borrowing costs and threatening the continuity of investment plans.

Take Oracle as an example. Its ambitious AI-related capital expenditure plans heavily rely on external financing. According to recent financial reports, Oracle’s free cash flow has turned negative, reaching -$10 billion. From the balance sheet, the company’s earnings are $28.9 billion, but net debt stands at $97.7 billion—an enormous mismatch raising market concern.

Market re-pricing of Oracle’s credit risk has already begun. A key indicator is the credit default swap (CDS) spread—the premium lenders demand for insuring against default. Oracle’s CDS spreads have been rising steadily over recent months, surpassing 140 basis points, reaching the highest levels since the 2008 financial crisis. This indicates increased lender concern and a higher risk premium. Consequently, Oracle’s future financing costs and difficulty are expected to rise significantly.

This is not an isolated case. Other AI-related companies are also facing similar issues—revenue growth falling short of expectations, yet still requiring substantial financing. Some high-performance cloud service providers have lowered revenue forecasts due to delayed customer contract fulfillment, then issued large convertible bonds to raise funds, further heightening market worries about their financing pressures. Their stock prices have recently fallen by 37%, and their bond CDS spreads have surged from below 400 basis points to 773 basis points, indicating a clear weakening of credit conditions.

Interwoven Tech Giants: Who Bears the Associated Risks?

This wave of AI is unique in that tech giants are taking on the role traditionally played by venture investors. They are not only investing in startups but also shaping the entire industry’s development direction. On the surface, this can enhance internal industry coordination, reduce information asymmetry, and improve overall efficiency.

However, it also introduces new vulnerabilities: complex investment and financing relationships among these companies create a web of interconnected risks. The risk of one company’s failure or liquidity crisis can quickly propagate through supply chains and capital links, triggering a chain reaction.

Currently, Nvidia, OpenAI, Oracle, and others have established multi-layered collaborations—covering direct investments, cloud service procurement, chip deployment, and joint R&D. These companies are “interlinked,” forming a tightly connected network.

Specifically: Nvidia has committed to investing up to $100 billion in OpenAI, while also purchasing $6.3 billion worth of cloud services from cloud infrastructure providers, investing $5 billion in Intel, and planning joint chip R&D. OpenAI has signed a $300 billion cloud computing partnership with Oracle, and also agreed to pay up to $22.4 billion to other infrastructure providers, deploying billions of dollars’ worth of AMD chips.

This highly intertwined structure means that if any one company faces investment failure or liquidity issues, the negative impact could quickly spread to partners, potentially causing a chain reaction across the industry. After Oracle’s stock price plunged last week, related companies’ stocks also weakened. Even chip companies with strong earnings have seen significant declines. This market reaction indicates that investors are reassessing the risks embedded in the current “cooperative” AI ecosystem, and contagion risks are increasingly being recognized.

What Does AI Slowdown Mean for the US Economy?

In 2025, the US economy shows resilience, but much of this strength is driven by expansion in AI-related fixed asset investments. Estimates suggest that AI contributed about 0.7 percentage points to the YoY growth of US real GDP, accounting for roughly one-third of total growth. This implies that, excluding AI, the traditional sectors lack endogenous growth momentum, and overall economic performance may not be as robust as the headline figures suggest.

Looking ahead to 2026, if doubts about whether AI capital expenditures can deliver sufficient returns persist, and if financing conditions tighten for related firms, a reasonable expectation is that growth in AI-related fixed asset investments will slow significantly.

This risk cannot be offset solely through monetary easing, because the core constraint for AI is not financing costs but the uncertainty of realizing investment returns. Additionally, current tariffs are pushing up the prices of AI capital goods, representing supply-side constraints that monetary policy cannot address.

The wealth effect of AI is also worth noting. Data shows that nearly half of US consumer spending is contributed by the top 10% income group, which holds about 87% of US stocks. Over recent years, this high-net-worth group has benefited greatly from strong capital market returns. If a market correction diminishes this wealth effect, consumer spending could weaken.

Meanwhile, the US labor market has shown signs of ongoing softening, with uncertain employment prospects dampening consumer confidence. Historical experience indicates that during late-cycle phases, insufficient consumer demand becomes prominent. The current “K-shaped” consumption pattern—where high-income groups maintain spending while middle- and low-income groups face pressure—may be signaling such a trend, warranting continued attention.