Infinite AI demand meets finite power in volatile chip markets
Tech leaders insist artificial intelligence demand is bottomless, but a shift in the industry's bottleneck from silicon to electricity is exposing the gap between genuine growth and overheated stock valuations.
Artificial intelligence executives are unified in their message: demand is effectively infinite. Lumentum, a critical supplier of optical components for data-centre connectivity, has sold out its products for the next five years. Yet the semiconductor stocks riding this infrastructure wave continue to lurch violently.
The disconnect is not a lack of buyers, but a surplus of expectations. The PHLX chip index has rallied roughly 60% this year, a level that prices in years of flawless corporate execution. When valuations run this hot, even exceptional operational results are treated as disappointments.
Recent earnings illustrate this dynamic perfectly. Samsung forecast an enormous profit rise following a 12-month rally exceeding 360%, only to see its shares fall. Cerebras doubled its revenue and watched its stock drop. Meta added to investor nerves by announcing plans to sell off excess AI computing capacity, leaving the market to debate whether this was savvy monetisation or a quiet admission of overbuilding.
The underlying business fundamentals remain robust. Unlike the dot-com era, today’s leading AI companies are extraordinarily profitable, and supplier order books are not fabricated. SoftBank’s Masayoshi Son dismissed bubble talk entirely, calling such comparisons an "insult" to a generational infrastructure project. Critics do not seriously dispute the demand. They dispute the price, noting that market concentration now exceeds year-2000 levels while the returns on hundreds of billions in capital expenditure remain unproven.
Both sides can be right at once. Demand can be genuine while stocks remain priced beyond any sensible payback timeline. For European investors watching from across the Atlantic, the decisive factor is no longer silicon, but power. Pat Gelsinger, the former Intel chief now at Playground Global, recently framed energy availability as "the only real limiter" to "almost unlimited" AI demand.
If electricity is the true bottleneck, semiconductor valuations increasingly rest on physical infrastructure that the tech sector does not control. Capital is already pivoting, with Nvidia-backed startups raising funds specifically to solve data-centre power shortages. However, upgrading electricity grids, deploying turbines and securing planning permission operate on regulatory and physical timescales that ignore quarterly earnings cycles. The tech industry has convinced itself that demand is infinite, and it may be correct. But the electricity to serve it is strictly finite, and stock markets are finally waking up to that constraint.
