Seeing Ghosts

AI remains a demanding but fundamentally supported investment build-out rather than a market bubble, because skepticism is high while revenues, demand and cash generation continue to expand.

• The S&P 500 forward P/E fell from 23.8x last October to 18.3x on November 5, 2026, while Nvidia traded at 14.3x and Micron at 4.4x; consensus still expects 18% earnings growth next year after 40% growth this year.
• AI market concentration is high, yet the leading companies combine infrastructure, distribution, data and recurring revenues, with more than 850 acquisitions between them, making them closer to seven sectors than seven individual stocks.
• Demand remains ahead of supply: Elon Musk reports persistent shortages of compute, while Amazon raised its capital expenditure guidance and still left demand unserved; this contrasts with the unused fiber capacity that followed the dotcom boom.
• Hyperscalers carry about 0.3x net debt to EBITDA versus 1.5x to 2.0x for the S&P 500 excluding financials, while rental prices for older GPUs are rising and Oracle says current re-rental prices could be about 20% above its original contract price.
• Hyperscaler operating cash flow is estimated to grow 30% this year and 32% next year, with Oracle expecting more than $30 billion of free cash flow in fiscal 2029; AWS, already above $180 billion in revenue, is guided to 40% growth in both 2027 and 2028.
• The main risks are a renewed rise in bond yields, tougher data-center regulation, falling GPU rental prices, slowing operating cash flow alongside rising capital expenditure, wider neo-cloud credit spreads or cloud revenue below guidance.

Takeaway: Investors should treat concentration and financing risks seriously while the evidence for a bubble remains limited, and monitor compute pricing, cash flow, credit spreads, cloud revenue and Washington policy for signs that the investment case is weakening.