The demand for AI is 'boundless'. So why are stock prices for chip companies declining?

      Executives at TL;DRAI assert that demand is “almost unlimited,” with Pat Gelsinger pointing out that energy is the primary constraint and Lumentum indicating that its products are sold out for the next five years. However, chip and data center stocks continue to fluctuate, as the ~60% year-to-date increase in the PHLX chip index reflects the expectation of perfect execution: Samsung anticipated a significant profit increase yet its shares declined, and Meta’s intention to sell its surplus computing capacity had mixed implications. The issue lies in expectations rather than in demand.

      Leaders driving the AI surge are steadfast in their beliefs. They claim demand is virtually endless, even as the stocks tied to it experience volatility, according to CNBC.

      Pat Gelsinger, the former Intel CEO now at Playground Global, states very clearly that he views AI demand as nearly boundless, with energy availability being “the only genuine limiter.”

      Evidence from order books supports his claim. Lumentum, which provides optical components for data center connectivity, has reported that its products are sold out for the next five years.

      So what explains the stock market's instability?

      The answer is that the current prices already factor in high expectations. The PHLX chip index has seen a roughly 60% increase this year, anticipating years of impeccable execution.

      At such levels, positive news may not suffice. Samsung projected a significant profit increase, yet its stock still dropped following a 12-month rally exceeding 360%.

      A similar situation occurred with Cerebras, which doubled its revenue only to observe its stock decline. When expectations are this elevated, even a good outcome can be interpreted as a disappointment.

      Meta contributed to the unease by announcing plans to offload its surplus AI computing capacity. Investors could interpret this either as a savvy monetization move or as an indication that the company over-invested in computing resources.

      The bullish and bearish perspectives

      The bulls back their claims with concrete figures. Unlike the dot-com era, the firms fueling this surge are highly profitable, and the demand indicators from suppliers are authentic. SoftBank’s Masayoshi Son has taken it a step further, asserting that labeling AI as a bubble is an affront. He views this expansion as a once-in-a-generation infrastructure initiative, rather than a fleeting craze.

      The bears, while not disputing the demand itself, contest the valuations, pointing out that market concentration now surpasses levels seen in 2000, and the returns on significant capital expenditures remain untested.

      Both perspectives can coexist. Demand can be real, while stocks may still be priced beyond what this demand will deliver on a realistic timeline.

      The constraint that cannot be circumvented

      Gelsinger's caution is critical to consider. If energy is the limiting factor, then chips are no longer the chokepoint, and the industry's valuations hinge on infrastructure that is beyond its control.

      Capital is flowing toward addressing this gap, with Nvidia-backed startups securing funding to tackle data center energy issues. However, the development of grids, turbines, and securing planning permissions occur on timelines that disregard quarterly profits.

      This is the uncomfortable reality behind the market’s volatility. The industry may believe that demand is endless, and it could be correct, but electricity is a finite resource, and share prices are not.