Executive Summary
AI is powering markets, not the economy. Valuations have soared as AI giants dominate the S&P 500, but productivity gains and profits lag behind the hype. Rising energy costs and shrinking free cash flows signal risks ahead. The long-term upside is immense, but near-term earnings disappointment could hit those stocks currently benefitting from the AI thematic hard. This is one of several concerns we currently have regarding US stock market valuations and why we are still recommending an underweight position to US stocks.
The AI Theme: Stocks vs the Real World
The artificial intelligence (AI) boom has been one of the most significant drivers of equity markets over the past few years. The combined market capitalization of leading AI-related companies including Nvidia, Microsoft, Google, Amazon, and others – has surged to $18 trillion, now representing one-third of the S&P; 500.
However, the economic impact of AI has been less visible so far. Capital expenditure (capex) on AI infrastructure has risen sharply, but much of this investment is in imported equipment, muting the net benefit to domestic GDP. In short, AI’s influence has been more of a financial market theme than an economic growth engine.
Data centers exemplify this tension. Their rapid expansion has led to soaring energy consumption, with U.S. data centers using 4.4% of national electricity in 2023 – a share projected to potentially triple this decade. This surge is straining power grids, raising wholesale electricity prices, and indirectly pressuring households and businesses with higher costs. The U.S. Department of Energy expects that without additional renewable capacity, energy bottlenecks could hamper both AI expansion and other industries reliant on affordable power.
Another concern is whether AI can deliver sustained profitability. Unlike software firms that benefit from near-zero marginal costs, scaling AI requires expensive computing power. This raises doubts about whether AI firms can maintain dominant market positions or if the industry could resemble capital-intensive sectors such as airlines or shale oil – essential but persistently low-margin. Currently, hyperscalers enjoy strong profits due to demand exceeding supply, but free cash flow is trending lower, a potential early signal of overinvestment risks. Productivity gains remain elusive. Despite hype, economy-wide data shows little evidence of AI-driven efficiency improvements. A recent MIT study found that 95% of organizations reported no measurable return on generative AI investments. This echoes the “Solow paradox” of the 1980s, when computers were ubiquitous but productivity statistics lagged. If AI follows a similar timeline, it could be another decade before its benefits materialize in a measurable way.
For investors, this means stock valuations may be vulnerable to disappointment if expectations for near-term profit boosts prove overstated. Longer-term potential remains immense, particularly if today’s large language models (LLMs) serve as stepping stones to Artificial Superintelligence (ASI). In that case, the economic pie could expand dramatically, rendering current concerns about profit distribution less relevant. Yet there are signs of plateauing. Apple’s June research paper argued that even the latest models, including GPT-5, struggle with tasks beyond their training scope, suggesting limits to current architectures. In summary, AI has reshaped financial markets but has not yet transformed the broader economy. Investors should closely monitor hyperscaler cash flows, energy infrastructure constraints, and signs of productivity growth to judge whether AI’s promise is sustainable. The technology’s eventual impact could be transformative, but in the short run, its risks may rival its opportunities.