As the AI narrative begins to waver, long-overlooked risks in the market are quietly emerging

The sharp sell-off in the stock market reveals that investors' patience for AI investment returns is wearing thin, and multiple risks previously masked by strong growth expectations are now emerging simultaneously.

Citadel Securities warned in its latest report that as AI trading logic is shaken, the market's margin for error has become extremely limited. Higher interest rates have made the return on AI infrastructure investments a more urgent issue, while the continued expansion in investment scale has made finding answers to these questions ever more critical. The report points out that this scrutiny runs through the entire industry chain—from client productivity gains, to the pricing power of model developers, and whether infrastructure investments can sufficiently cover financing costs and depreciation.

Meanwhile, fiscal policy, tariffs, rising energy prices, and geopolitical tensions are building on each other, putting the Federal Reserve in a dilemma. Citadel Securities believes these risks have long been obscured by the strong narrative of AI and economic growth, but as market confidence in this story becomes wobbly, the mounting pressures will be increasingly hard to digest.

The massive spending on AI infrastructure has evolved from a pure growth signal into a double-edged sword.

Take Meta as an example. Its third-quarter revenue grew 26% year-on-year, but with capital expenditures (including principal payments on finance leases) reaching $34.9 billion—exceeding the $32.8 billion operational cash flow for the same period—quarterly free cash flow turned negative by $2.1 billion. The annual guidance implies an additional $49–59 billion in investment for the fourth quarter.

Citadel Securities points out that while the eventual returns may justify this spending, investors need more confidence that the relevant capacity can be financed, built, and commercialized on schedule. This constitutes the so-called "bidirectional risk of CapEx": cutting spending raises doubts about the sustainability of the investment cycle; continuing to ramp up brings questions about financing capability, execution, and ultimate returns.

On the demand side, expenditure data sends more complex signals. Ramp’s figures show that although token usage continues to grow, related spending was nearly flat from August to September. This divergence is mainly explained by price declines and efficiency gains. Spending hit a new high in early October, but the growth rate has visibly slowed.

The release of DeepInfer by DeepSeek further demonstrates that software-level improvements can boost the efficiency of existing hardware, adding uncertainty to the incremental capacity scale required for future AI demand—even as lower inference costs encourage wider adoption.

Citadel Securities believes that cheaper intelligent services improve client-side economics and broaden the scope of viable applications; but for suppliers, profitability depends on whether costs decrease faster than prices, and the dynamic balance between incremental usage and the required supporting investments. A successful technology can create enormous economic value while delivering disappointing returns to some producers. This logic continues to favor diversified platforms with mature cash flows and multiple monetization paths—even though shareholders of such platforms also need to distinguish between "funding the next phase of growth" and "deriving attractive returns" as fundamentally different goals.

The AI industry is facing these economic challenges with limited political goodwill.

According to a Reuters/Ipsos poll this month, 69% of registered Democratic voters support curbing data center construction, while among Republican voters, this proportion also reaches 43%, showing that opposition crosses party lines. Citadel Securities notes that the November midterm elections may lead to stricter congressional oversight, while state and local elections will directly affect project permits and development conditions. Even in the absence of a sweeping federal restriction, additional compliance requirements, local building limitations, and lengthy approval processes are enough to make AI deployment slower and more expensive. For projects funded with leverage, delays mean additional interest costs before revenue starts to flow in, and possibly facing a dramatically altered competitive landscape when capacity comes online.

This week's cybersecurity news has further intensified political risks. AI-assisted researchers discovered a critical vulnerability, over a decade old, in widely-used PHP software. Meanwhile, reports indicate Amazon has detected attempts to exploit the flaw and warns that AI is shortening the time from vulnerability information disclosure to the creation of working attack tools.

Citadel Securities expressed clear concerns: Legitimate security worries could become the pretext for broad restrictive measures, with the economic costs of such policy responses potentially far exceeding the risks of the vulnerabilities themselves—at a time when the speed of AI adoption is crucial for growth prospects, this poses additional risks to return timelines.

The risks above do not exist in isolation but are occurring amid a policy environment already pushing up the costs of economic financing and operations significantly.

Citadel Securities notes that pro-cyclical fiscal easing has stoked demand and pushed up financing needs, while tariffs have increased costs and uncertainty, collectively making it harder to bring inflation steadily back to target. Meanwhile, higher real interest rates do partly reflect stronger growth and productive investment, but the cumulative burden cannot be ignored. As debts are gradually refinanced, higher rates will increasingly affect a broader array of borrowers; higher input costs will squeeze profit margins or be passed on to consumers; and persistent inflation makes monetary easing harder to implement.

Energy prices are another major variable. Brent crude climbed above $117 a barrel this week, with Saudi Arabia hit by attacks and disruptions to Gulf shipping. Despite continued strong regional oil exports, fears of supply interruptions have been rekindled in the market. Persistently high energy costs will erode purchasing power and push up operating expenses.

Citadel Securities believes that if an AI-led investment slowdown coincides with an energy shock, the Federal Reserve will face the dilemma of deteriorating growth prospects alongside persistently high inflation. In addition, reports suggest that military action against Iran may occur post-election, while a divided Congress will limit Washington's ability to manage the fiscal path.

The report concludes: A temporary stock market correction may be manageable, but if investment and financing activity see sustained contraction, it would threaten the growth momentum, corporate profitability, and household wealth base that have so far sustained the U.S. economy in challenging times. Even if Treasury yields decline, it may not offer substantive relief to borrowers whose credit spreads blow out sufficiently—risk-free yields can fall, while the cost of risk financing could still rise. These risks have been accumulating for some time, merely hidden by robust growth and the AI narrative; as confidence in this narrative starts to falter, and with both the economy and the stock market so heavily staked on AI trading, the margin for error is now vanishingly small.