Equity Markets Q3 2025 Commentary

Power generation represents another issue facing AI development. After years of flat demand until around 2020, electricity consumption has been growing at a roughly 2% annualized rate. As the importance of AI inference (answering questions or performing tasks) grows relative to training, demand will increase further but it’s difficult to know how much. Utilities are overwhelmed with requests for connection to the grid, but how many of the requests are being submitted to multiple utilities? Utility companies are wary or overbuilding, being stuck with excess capacity and needing to raise customer rates. Of course, building new power infrastructure, especially grid improvements, is a time-consuming process made more difficult by the fragmented electricity market in the US. 

Finally, we arrive at most important question: Will companies investing in AI ever see a return on their investments? The question concerns not only the companies building AI systems but also the companies hoping to leverage AI to improve their operations. Indeed, if the latter cannot make it work, the former will be hung out to dry. Nvidia, the Levi Strauss of AI, has already made its return but how many AI prospectors will find gold?

A July report by the Massachusetts Institute of Technology raises concerns. The report, titled “The State of AI in Business 2025” states, “Despite $30-$40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return.” Early days, to be sure, and much of the problem appears to stem from companies not understanding how best to deploy AI as much as the shortcomings of AI itself. Regardless, learning will take time while data centers and graphics chips are rapidly depreciating assets.

As the DeepSeek saga earlier in the year demonstrated, AI investment sentiment can turn quickly. More likely than not development will follow a similar trajectory to what was experienced with railroads in the 1800s, radio and the auto industry in the 1920s, and the internet at the turn of century. Stanford computer scientist Roy Amara said in the 1960s, “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.” AI may be another example of a well-established phenomenon.

Retail sales of electricity, United States, monthly

Image

Market Outlook 

September stands as the worst performing month historically but, courtesy of 1929 and 1987, October gets all the glory. This year it will be remembered for kicking off another government shutdown. Republicans and Democrats appear far apart on key elements of funding the government, especially relating to health care. We would not be surprised were the current impasse to break the 35-day record set in 2018 over disagreement on funding a border wall. Unlike a wall, which the Republicans eventually abandoned, the GOP is hanging its hat on reducing government spending, while the Democratic Party views healthcare as a core issue and an opportunity to drive a wedge between voters and Republicans. Neither side appears open to compromise. On the other hand, the shutdown may be over by the time you read this. 

Interestingly, the stock market does not always react to disruptive events, including government shutdowns, Equity Markets Commentary in a manner one might expect. While the S&P 500 Index sunk 2.6% on the first trading day following the 2018 shutdown, it subsequently rallied over 13% by the time the government reopened. That may have had to do with the nature of the dispute, which did not include raising the debt ceiling. Investors have remained remarkably sanguine regarding the vast increase in debt over the past several years, but debt ceiling fights have the effect of bringing the issue into focus. 

Conversely, President Trump has indicated that he will approach the shutdown differently than past administrations. Rather than striving to keep as many services as possible operating, furloughing workers or asking employees to work without pay, the administration plans to engage in aggressive layoffs – a throwback to Elon Musk and the days of DOGE. Just like DOGE, it’s likely many of those employees would eventually be recalled when the necessity of their work becomes apparent. In the meantime, however, recession fears may rise at the same time tariffs are expected to exert a greater influence on inflation, exacerbating the growth versus inflation conundrum faced by the Federal Reserve.