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

The artificial intelligence gold rush is currently driven by an unprecedented capital expenditure cycle, but a significant warning has emerged regarding its long-term sustainability. Chris Wood, Jefferies’ Global Head of Equity Strategy, suggests that the AI trade may not end due to a lack of demand, but rather a crisis of profitability.

The Risk of Malinvestment and Circular Funding

Unlike traditional semiconductor cycles that typically end due to sudden inventory gluts or supply shocks, Wood identifies "malinvestment" as the primary risk to the AI trade. The core concern is that hyperscalers and leading AI labs may fail to generate an adequate return on the massive capital expenditures (capex) they are undertaking.

Wood highlights a potentially precarious feedback loop within the ecosystem: circular funding arrangements where companies like Nvidia provide financing to AI labs like OpenAI, which then use that capital to purchase more Nvidia chips. While this drives short-term growth, it creates a vulnerability. If investors begin to doubt the long-term monetization capabilities of these AI investments, this feedback loop could unwind sharply, triggering a painful pause in the market.

Massive Capex and the Jevons Paradox

The scale of investment currently being witnessed is historic. TSMC has raised its 2026 capex guidance to approximately $56 billion, up from $41 billion last year, with some projections suggesting spending could reach $65–$70 billion by 2027. This surge has already significantly boosted the Taiwanese economy, with real GDP growth hitting 14.55% year-on-year in Q1 2026.

Wood views this through the lens of the "Jevons Paradox"—the idea that as the cost of a resource (in this case, computing tokens) falls due to increased efficiency, the total consumption of that resource actually rises. This paradox benefits the "picks and shovels" players, specifically memory and DRAM suppliers. Companies like Micron are already seeing this structural shift, with Micron signing 16 strategic customer agreements covering 20% of its DRAM volume and a third of its NAND volume, often with five-year tenors.

The Commoditisation of AI Models

Another mounting pressure on the economics of premium Western AI providers is the rapid commoditisation of Large Language Models (LLMs). The rise of high-quality, low-cost models—particularly from China—is challenging the dominance of US-based firms.

For instance, the launch of Z.ai’s GLM-5.2 has been described as nearly equal to Anthropic for corporate use, but at just one-quarter of the cost per token. Data from OpenRouter shows a significant shift in volume; in late June, top Chinese AI models processed 21.37 trillion tokens, significantly outperforming the 5.76 trillion tokens processed by top US models. This influx of cheap, capable alternatives puts immense pressure on the profit margins of the companies leading the AI race.

Key Takeaways

• The Primary Risk: The AI trade's end is more likely to be triggered by a realization of "malinvestment" and poor returns on capex rather than a traditional semiconductor oversupply.
• Structural Shift in Memory: DRAM and memory suppliers are currently the most insulated "picks and shovels" beneficiaries, leveraging long-term strategic agreements to maintain pricing power.
• Commoditisation Threat: The rapid emergence of low-cost, high-performance AI models (notably from China) is commoditising the LLM landscape, threatening the high-margin models of Western AI leaders.