Record-breaking gains in the US stock market are facing a structural threat that the market has underestimated—the sharp rise in the cost of artificial intelligence computing power is eroding the profitability of hyperscale cloud companies and raising serious questions about whether massive AI spending will be met with sufficient demand.
The growth rate of data center costs has significantly outpaced revenue growth. Microsoft has attributed $25 billion of its 2026 capital expenditure to rising component prices, and Meta CEO Mark Zuckerberg expressed similar concerns when raising the spending outlook mid-year. Of the more than $700 billion in AI infrastructure investment this year, an increasing portion is being consumed by rising input costs.
The optimistic AI narrative has been a core driver pushing the S&P 500 index to repeated all-time highs. However, as demand prospects become more uncertain and free cash flow continues to narrow, the foundation of this logic is weakening.
According to Nomura Equities data, 10 stocks contributed 69% of the nearly 20% gain in the S&P 500 index from its low for the year; Goldman Sachs estimates that AI-related investments are driving roughly 40% of the index's overall earnings per share growth projected for 2026. Should the AI momentum fade, the market would face a vacuum lacking earnings support.
Soaring Memory Costs Create a Price Bottleneck for Compute Expansion
The core pressure point for rising AI infrastructure costs lies in memory. Data from SemiAnalysis shows the proportion of hyperscale cloud company capital expenditure allocated to memory has jumped from 8% in 2023 to nearly one-third, driven by each successive generation of GPU requiring more DRAM, pushing the cost curve ever steeper.
Supply-demand imbalance has already led to a significant increase in memory chip prices. Micron has noted that demand for memory chips is so strong that some customers are only receiving about 50% of their requested supply. This situation has allowed the company to raise DRAM prices by approximately 60% and NAND prices by about 70%.
Power supply shortages are further exacerbating expansion bottlenecks. RBC research indicates that grid connection approvals currently take around 55 months, while data center construction cycles are only about two years. As Microsoft enters 2026, it has $80 billion worth of Azure orders unfulfilled, precisely because GPUs are sitting idle awaiting power connections. Power provider PJM has stated that capacity prices have surged more than tenfold over two years, influenced by growing data center load.
Dwindling Free Cash Flow Forces Hyperscalers to Borrow for Expansion
The dramatic expansion of capital expenditure is squeezing the financial flexibility of hyperscale cloud companies. According to Pimco estimates, capital expenditure is projected to absorb 94% of these companies' operating cash flow over the next two years, with free cash flow having already shrunk substantially.
This shift fundamentally alters the market risk profile of these firms. Once steady sources of stock buybacks, these hyperscalers are now increasingly reliant on credit markets to finance their spending. This not only tightens financial conditions but also implies that current stock valuation multiples do not fully reflect this change in risk.
For now, strong cloud and AI revenue growth is partially masking these pressures, but whether this cover can persist depends on whether demand can keep pace with the expansion on the supply side.
Uncertain AI Demand Outlook and a Persistent Monetization Gap
The market's assumption of "unlimited" AI demand is facing growing skepticism. Gartner predicts that by the end of 2027, over 40% of agentic AI projects will be canceled due to rising costs, unclear return on investment, and insufficient risk management. An S&P Global survey shows that in the 12 months ending last October, 42% of enterprises abandoned most of their related AI plans before projects even reached production.
The scale of the monetization gap is equally alarming. Bain estimates that the current scale of data center construction would require annual revenue of $2 trillion by 2030 to recoup costs, yet three years after ChatGPT's launch, AI-related revenue remains at only about $20 billion. Meanwhile, Forrester expects enterprises to defer a quarter of their planned AI spending to 2027, with less than a third able to link their AI investments to actual improvements in corporate profits.
Furthermore, IPO pressures facing AI platforms like OpenAI and Anthropic will likely force them to gradually abandon early strategies of acquiring users at below-cost prices and instead focus on profitability. This suggests a significant portion of the current AI compute consumption demand could subsequently disappear.
High Valuations Leave Little Room for Error
Current market valuations leave almost no room for earnings surprises. The S&P 500 index is trading at a forward price-to-earnings ratio of about 21, above its long-term average. According to Allianz research, the overall valuation of the AI sector is near 25 times EV/EBITDA, exceeding the valuation levels of the telecom sector at the peak of the 2000 dot-com bubble.
Against this valuation backdrop, any downward revision to earnings would have an amplified effect through multiple compression. For example, a 10% reduction in the earnings per share of the "Magnificent Seven," combined with a downward re-rating of their P/E multiples, would significantly drag on the overall index.
Triggering this risk does not require a systemic crisis—merely one major hyperscaler cutting its 2027 capital expenditure guidance due to weak demand could be enough to rapidly unravel the current AI trade narrative.
It is worth noting that some signs of internal market rotation in leadership have emerged, with the equal-weight S&P 500 index recently outperforming the market-cap-weighted version. However, if AI momentum fades before other sectors achieve significant earnings growth, the entire index would face the risk of lacking a solid earnings floor. This prospect strengthens the case for hedging equity returns ahead of time.
Comments