Why Malinvestment, Not Chip Oversupply, Could End the AI Boom

The relentless surge in Artificial Intelligence investment has defined the current market era, but a significant warning has emerged from one of Wall Street's most watched strategists. Chris Wood, Jefferies’ Global Head of Equity Strategy, suggests that the downfall of the AI trade will not stem from a lack of chips, but from a failure to turn massive capital expenditures into meaningful profits.

The Looming Risk of Malinvestment

Unlike traditional semiconductor cycles that typically end due to inventory gluts and sudden supply increases, Chris Wood argues that the AI era faces a unique structural risk: malinvestment. He warns that the "end" of the AI trade—or at least a painful market pause—will likely be triggered when hyperscalers and leading AI labs fail to generate adequate returns on their colossal capital expenditures (capex).

Wood points to a concerning feedback loop within the ecosystem, such as Nvidia financing companies like OpenAI, which then use that capital to purchase more Nvidia chips. While this circular funding drives immediate growth, it creates a precarious foundation that could unwind rapidly once investors demand visibility on long-term earnings and capital discipline.

A Record-Breaking Capex Cycle

The scale of current AI spending is unprecedented. Wood describes the ongoing build-out as the most dramatic capex cycle he has ever witnessed. The concentration of this investment is most evident in the semiconductor industry:

  • TSMC’s Expansion: The foundry has raised its 2026 capex guidance to approximately $56 billion, up from $41 billion last year. Projections for 2027 suggest spending could reach between $65 billion and $70 billion.
  • Revenue Concentration: AI-related demand is expected to account for an estimated 31% of TSMC’s total revenue in 2026.
  • Macro Impact: This surge has fueled massive growth in Taiwan, with real GDP growth hitting 14.55% year-on-year in Q1 2026.

The Commoditisation of AI Models

Adding to the pressure on profit margins is the rapid commoditisation of Large Language Models (LLMs). Wood notes that cheaper, highly efficient models—particularly from Chinese developers—are challenging the dominance of premium Western providers.

For instance, Hong Kong-listed Z.ai’s GLM-5.2 is reportedly performing near the level of Anthropic but at just one-quarter of the cost per token. This shift is reflected in usage data: in late June, top Chinese models processed 21.37 trillion tokens on OpenRouter, a massive jump from 4.37 trillion in April, significantly outpacing the 5.76 trillion tokens processed by top US models. As token costs fall, the "moat" for premium AI providers shrinks, making it harder to recoup massive infrastructure costs.

Winners in the "Picks and Shovels" Race

Despite these long-term risks, Wood is not predicting an immediate collapse. Instead, he is repositioning portfolios toward "picks and shovels" plays—specifically memory and hardware providers—who benefit from the Jevons Paradox. This economic principle suggests that as compute becomes more efficient and cheaper, total consumption actually increases.

Memory giants like Micron, SK Hynix, and Samsung are currently in a position of strength. Micron has already secured strategic agreements covering 20% of its DRAM volume and one-third of its NAND volume, often with five-year durations, providing a buffer against the volatility expected in the software layer of the AI stack.

Key Takeaways

  • The Primary Threat: The AI trade is vulnerable to "malinvestment" risks, where massive spending by hyperscalers fails to yield sufficient returns on investment (ROI).
  • Shift in Risk Profile: Unlike previous semiconductor cycles driven by oversupply, the AI cycle's end will likely be driven by investor disillusionment with capital discipline.
  • Hardware Resilience: While software model margins face pressure from commoditisation, memory and hardware providers remain the primary beneficiaries of the ongoing capex race.