One of the world’s most popular economists on why AI is ‘undoubtedly going to crash’: It’s built off ‘digital lettuce’—and the U.S. will be just fine anyway

Economist David McWilliams cautioned that the U.S. AI-led “boominess” could crash because much investment is concentrated in rapidly obsolescing GPUs—what he calls “digital lettuce”—arguing AI doesn’t create jobs and firms may be overpaying for perishable capacity. His critique echoes Michael Burry’s concerns about lengthening depreciation schedules, although market commentators such as Ed Yardeni and Nvidia’s recent earnings have pushed back on an imminent collapse. For investors the takeaway is heightened risk of revaluation in capex‑intensive AI and data‑center names, focus on useful‑life and impairment assumptions in financial statements, and a political/regulatory tail risk if chips are recast as strategic assets—risks McWilliams believes are mitigated by America’s deep innovation and risk‑taking culture.

Analysis

David McWilliams, a widely followed Irish economist, argued that the current U.S. AI-led "boominess" is vulnerable to a crash because much investment is concentrated in GPUs that he calls "digital lettuce," claiming chips become outdated quickly and that AI does not create jobs. His interview occurred before Nvidia's most recent quarterly results, which the article notes helped assuage some market fears about an imminent bubble.

McWilliams’s critique echoes Michael Burry’s “useful life” concerns over lengthening depreciation schedules and potential under-recognition of impairments, while Ed Yardeni and recent earnings commentary counter that many data centers predate the AI surge and continue operating with original chips. The broader tone from the article and signal data is mildly negative (sentiment_score -0.3) with limited immediate market disruption indicated (market_impact_score 0.3); NVDA’s per-ticker sentiment is slightly negative (-0.1).

Practical implications are heightened revaluation risk for capex-intensive AI and data-center names, a need to scrutinize useful-life and impairment assumptions in financial statements, and a conditional political/regulatory tail risk if chips are recast as strategic assets—an outcome McWilliams judges unlikely given current political incentives. His wider point is that U.S. innovation and a risk-embracing culture reduce the chance of a systemic collapse even if specific assets are overinvested in and shorten product lifecycles.

Key Decisions for Investors

• Reassess exposure to GPU-heavy, capex-intensive names (including NVDA) and stress-test valuation models under scenarios with shorter useful lives and higher impairment rates
• Prioritize companies with diversified revenue streams or long-term capacity contracts rather than pure-play GPU capacity providers, and consider trimming positions where earnings depend heavily on ongoing GPU re-spending
• Monitor upcoming earnings and capex guidance for signs of slowing GPU demand and explicitly check management disclosures on depreciation schedules, useful-life assumptions, and impairment risk
• Track political and regulatory developments that could reclassify chips as strategic assets and consider policy hedges (options or reduced notional exposure) ahead of potential election-driven shifts, even though the article judges that outcome unlikely