Sep ARL - The AI Arms Race

The AI investment boom is becoming financially unsustainable, leaving investors exposed to weak long-term equity returns even though the bubble may continue for years.

• Goldman Sachs raised its estimates for hyperscaler capex in 2026 and 2027 between late May and late June; 15 months earlier, it had forecast $365 billion for 2026 across all seven Magnificent 7 companies, rather than only the five shown in its latest exhibit.
• Hyperscaler capex is accelerating relative to GDP while cash is declining relative to capex, forcing companies either to borrow more or slow investment; borrowing by megacap technology companies increased materially in 2025 after remaining limited through 2024.
• The author compares the current cash-and-capex divergence with the late 1990s, when cash ran out before technology companies cut spending and the Nasdaq 100 subsequently fell 83% from peak to trough in the dotcom bust.
• From around 2008 to the early 2020s, capex growth broadly matched cash growth, but a significant gap has opened in recent years as competition for AI leadership has overridden governance discipline.
• U.S. equity-market concentration is now comparable to levels seen before previous bubbles burst, and the comparison excludes SpaceX, which the author says has further increased concentration; high P/E multiples and concentration imply annualised 10-year returns in the low single digits at best.
• Analysts expect operating margins for the S&P 493, the S&P 500 excluding the Magnificent 7, to exceed anything previously observed, creating what the author describes as an unprecedented gap between expectations and reality.

Takeaway: Investors should not try to time the end of the AI bubble, which could arrive tomorrow or several years from now, but should expect poor 10-year equity returns and rely on patience rather than current margin and valuation optimism.