AI Capex Slowdown: The S&P 500's Unspoken Leverage Event

Logic does not bleed, but capital flows leave traces. Over the past seven days, a single data point has been quietly circulating among institutional desks: AI spending is decelerating. Not collapsing, not reversing—just slowing. But in a market where the top 20 stocks now hold 50.8% of the S&P 500’s weight, a deceleration in the narrative that built that concentration is not a minor signal. It’s a structural shift. Context: The Capital Expenditure Casino The AI infrastructure buildout has been the defining macro trade of 2025. Goldman Sachs estimates annualized AI-related spending could exceed $800 billion by end of 2026. Morgan Stanley pushes it further: nearly $3 trillion by 2028, with 80% yet to be deployed. The five largest hyperscalers—Amazon, Microsoft, Google, Meta, Oracle—are set to deploy over $1 trillion in 2025–2026 alone. This is not optional investment; it’s defensive arms race. Every hyperscaler knows that halting capex means ceding narrative ground to a competitor. But the market is now asking a question that no earnings call wants to answer: what is the marginal return on that trillion-dollar bet? The Bank for International Settlements (BIS) has warned that the spending frenzy could turn into a “long-term investment crash.” The BIS doesn’t make headlines unless it sees systemic risk. Core: The Structural Deconstruction of AI’s Profit Quality Let’s break down the accounting. Mac10—a quantitative analyst whose work I’ve cross-referenced in past audits—points out that the record forward earnings growth in the S&P 500 is largely driven by AI capex flowing through income statements as a “one-time event.” In plain terms: companies are spending unprecedented cash, and that spending temporarily inflates reported earnings. It’s not sustainable operational growth. It’s a sugar high. From my own experience dissecting tokenomics in 2017, I’ve seen this pattern before. When a system’s primary value driver is capital infusion rather than revenue generation, the quality of that growth is suspect. The same logic applies here. The revenue from AI—cloud inference, API calls, Copilot subscriptions, AI agents—has not yet validated the scale of infrastructure spending. The incremental revenue-to-capex ratio remains opaque. Hyperscalers are not breaking out AI-specific revenue. They are bundling it into “cloud” and “other revenue.” Consider the storage sector. Sandisk and Western Digital are up 396% and 145% year-to-date, respectively. That’s a shadow indicator of AI-driven demand for NAND flash. But storage is historically cyclical. Any moderation in hyperscaler procurement will trigger an inventory correction that hits these stocks harder than the initial uptick. The rug is not pulled; it was never tied. Aschenbrenner’s fund collapse is the microcosm. The former OpenAI researcher’s fund grew to $45 billion, then cratered to $10 billion after AI infrastructure stocks corrected. He was inside the narrative—he bet on the AI trade with leverage. The fund’s remains were taken over by Citadel. This is not a failure of intelligence; it’s a failure of expectation management. Imagination is infinite, but liquidity is finite. Contrarian: What the Bulls Got Right Does this mean AI is a bubble ready to pop? Not necessarily. BlackRock points out that the current AI leaders—the hyperscalers—are generating real profits and have strong balance sheets. Most of the capex is funded by operating cash flow, not debt. This is a material difference from the dot-com era, where companies burned venture capital. There is also a plausible technical counter-narrative: the slowdown in AI spending might be driven by efficiency gains. If model scaling laws are indeed hitting diminishing returns, the need for brute-force compute might decline. More efficient architectures (like mixture-of-experts, or newer hardware) could reduce the per-unit cost of inference. That would be a positive signal, not a negative one. The market may be misinterpreting a natural maturation of the technology as a demand collapse. Furthermore, the hyperscalers’ overbuilding could create a second-order benefit: a glut of compute capacity would lower prices for AI startups and smaller players, accelerating application-layer innovation. The 2000 dot-com bust did exactly that for fiber optics—the excess capacity eventually enabled the modern internet explosion. The same could happen with AI. But—and this is the critical caveat—the timeline matters. The market is pricing in continuous exponential growth now. A 12-month pause in capex growth would be enough to trigger a repricing of the entire AI trade. The fund manager survey from Bank of America shows that 45% of respondents now cited AI bubble as the top tail risk, up from 28% the previous month. That’s a rapid shift in consensus. Takeaway: The Accountability Call The S&P 500 has become a proxy for the AI capex narrative. When that narrative slows, the index’s concentrated structure amplifies the shock. The question is not whether AI will transform the economy—it will. The question is whether the current pricing of that transformation is sustainable. Gas fees are the price of truth. In this market, the truth is that the return on a trillion dollars of infrastructure is still unverified. The next quarterly earnings from hyperscalers will be the first real test. If one of the top five cuts its capex guidance, the dominoes will fall faster than any analyst can model.